201033602 六、發明說明： 【發明所屬之技術領域】 本發明係關於檢查偏光膜或相位差膜等光學膜（特別 是捲成捲筒狀被保管•輸送的長尺寸光學膜）等成形薄片 的缺陷之缺陷檢查裝置。 【先前技術】 φ 從前的成形薄片之缺陷檢查裝置，使用稱爲線狀感測 器（line sensor )之1次元攝影機，以螢光燈管等線狀光 源照明成形薄片，沿著成形薄片之長邊方向由長邊方向之 一端至另一端以1次元攝影機掃描成形薄片表面取得1枚 靜止影像資料，根據此1枚靜止影像資料來檢査成形薄片 之缺陷。於此靜止影像資料，通常包含線狀光源像。線狀 光源像，係在線狀光源及攝影機與反射面之間被配置成形 薄片的場合，由線狀光源射出而藉由成形薄片正反射而到 〇 達攝影機的光的影像，在線狀光源與攝影機之間被配置成 形薄片的場合，則是由線狀光源射出而透過成形薄片到達 攝影機的光的影像。在此缺陷檢查裝置，成形薄片的寬幅 很寬的場合，以可檢査成形薄片的寬幅方向全區域的方式 ，把複數台線狀感測器在寬度方向上並列使用。 然而，在此從前的缺陷檢查裝置，根據針對成形薄片 全區域之1枚靜止影像資料（以下，簡稱「影像資料J ) 檢査成形薄片的缺陷，所以影像資料之檢査對象畫素與線 狀光源像之位置關係，成爲一種已決定的位置關係。缺陷 -5- 201033602 ，只會在檢查對象畫素（注目畫素）與線狀光源像的位置 關係在特定的位置關係時會出現於影像資料上。例如，缺 陷之一種之氣泡，常常只在線狀光源像的周緣或附近的場 合才會出現於影像資料上。因此，缺陷常因其位置而未被 檢測出。亦即，前述從前的缺陷檢查裝置，只具備有限的 缺陷檢測能力。 因此，本案申請人申請了可對前述從前之缺陷檢查裝 置提高缺陷識別能力之成形薄片之缺陷檢查裝置（參照專 利文獻1 )。在此缺陷檢査裝置，以螢光燈管等線狀光源 照明成形薄片，把成形薄片於特定方向連續搬送，同時使 用稱爲區域感測器（area sensor )的2次元攝影機取得動 畫資料（在成形薄片上之攝影位置不同的複數枚影像資料 )，而根據此動畫資料檢查成形薄片之缺陷者。在此缺陷 檢查裝置，可以根據檢查對象畫素與線狀光源像之位置關 係不同的複數枚影像資料而判定是否有缺陷，所以比起從 前的缺陷檢查裝置可更確實地檢測出缺陷。亦即，此缺陷 檢查裝置比起從前的缺陷檢查裝置更爲提高缺陷檢測能力 。又，利用此動畫資料的話，也可以看到缺陷對照明像移 動之模樣。 〔先前技術文獻〕 〔專利文獻〕 〔專利文獻1〕日本國公開專利公報「特開2007-218 62 9號公報（2007年8月30日公開）」 .201033602 【發明內容】 〔發明所欲解決之課題〕 然而’根據本案發明人的檢討，可知記載於專利文獻 1之缺陷檢查裝置，也還有改善缺陷檢測能力的餘地。 亦即，在記載於專利文獻1的缺陷檢查裝置，從藉由 區域感測器攝影的（多値之）複數枚影像資料之各個，藉 由以下之影像處理而檢測出缺陷（參照專利文獻1之〔 φ 0032〕〜〔0035〕）。 首先，將多値之影像資料2値化，標記白區域及黑區 域作爲檢測對象。接著，由檢測對象之白區域，把具有超 過特定値（配合於線狀光源像的面積之比較大的値；例如 2 5 00畫素）的面積（畫素數）之白區域，視爲線狀光源像 而予以排除。同樣地，由檢測對象之黑區域，把具有超過 特定値（配合於背景區域的面積之比較大的値）的面積之 黑區域，視爲背景區域（成形薄片之沒有缺陷的區域之影 φ 像）而予以排除。進而，由檢測對象之白區域及黒區域， 把具有不滿特定値（接近1個畫素之比較小的値；例如9 畫素）的面積之白區域及黑區域，視爲雜訊而予以排除。 接著，把未從檢測對象之白區域及黑區域排除掉之剩下來 的區域作爲缺陷進行檢測。 然而，記載於專利文獻1之缺陷檢査裝置，係根據明 暗之反轉而檢測出缺陷者，所以影像資料上之對比低的缺 陷沒有辦法檢測出。 此處，說明於本發明作爲檢測對象之種種缺陷。本發 201033602 明，主要是檢測伴隨著成形薄片表面的微小凹凸（特別是 數从m程度高的凹凸）之缺陷（外觀缺陷）。作爲伴隨著 微小凹凸的缺陷，例如可以舉出起因於氣泡或異物而於成 形薄片表面產生的微小凹凸；打痕（點狀按壓產生的壓痕 ):被折彎之痕（被稱爲「折曲」（音譯：kunick ))； 於製造成形薄片時，以搬送卷搬送時產生的搬送輥導致之 壓痕（被稱爲「折線」（音譯：suji ))等。這些伴隨著 微小凹凸的缺陷，對使用從前的線狀感測器之缺陷檢査裝 @ 置來說是非常難檢測出的。本發明主要目的爲檢測出這些 種類之缺陷。 在本案說明書，爲了方便，把微小凹凸局部集中的（ 凸部直徑在lmm程度以下（攝影裝置的分解能在200 /畫素的場合爲數個畫素程度以下））缺陷，例如，氣泡 、異物、打痕等稱爲點缺陷，微小凹凸連接成線狀超過 1mm大者稱爲線缺陷。折線等那樣典型的線缺陷，具有超 過10mm程度的寬幅（攝影裝置的分解能在200 // m/畫 © 素的場合爲超過數十個畫素程度的寬幅），典型者爲數十 cm程度，有時還具有超過數十cm之寬幅。折曲，具有 10mm程度以下（攝影裝置的分解能在200 # m/畫素的場 合爲數十個畫素程度以下）之寬幅，典型者具數mm程度 之寬幅，具有點缺陷與典型的線缺陷之間的中間性質。 於專利文獻1記載之缺陷檢查裝置之攝影部被攝影的 缺陷影像（動畫）之例，如專利文獻1之圖14〜圖1 5所 示。在專利文獻1之圖14〜圖15，構成動畫的連續5枚 -8- 201033602 圖框依照時間順序爲（a)〜（e)。此動畫若爲通常的電 視用動畫的話，圖框間的時間間隔（圖框速率）爲1/30 秒。圖框間之時間間隔依存於攝影部的特性。在專利文獻 1之缺陷檢查裝置，攝影部，係以由攝影部朝向成形薄片 的攝影區域（專利文獻1的圖1之成形薄片表面以虛線表 示之矩形）的中心之方向與成形薄片的搬送方向夾銳角的 方式，且於攝影區域之一部分包含照明像（線狀光源之反 φ 射像），以在攝影區域之照明像的兩側存在無照明像的區 域（背景區域）的方式被配置。亦即，專利文獻1之圖14 〜圖15所示之動畫之上方向，與成形薄片之搬送方向一 致。成形薄片之攝影區域之中，缺陷會移動於成形薄片的 搬送方向，所以於專利文獻1之圖14〜圖15所示之動畫 ，缺陷係由下往上移動（照明像，係照成白色帶狀之區域 )° 專利文獻1之圖14，係包含藉由攝影部攝影的氣泡（ ❹ 點缺陷）之攝影區域（與搬送方向直交的方向之尺寸： 5mm )之動畫例，氣泡係可見到明暗反轉的部分。在此動 畫’在第1個圖框（a)無法見到氣泡，點缺陷在接近照 明像之第2個圖框（b )開始看到氣泡，在氣泡位於照明 像的邊緣之第3個圖框（c )氣泡可以較清楚地看到，氣 泡進入照明像中之第4以及第5圖框（d )( e )因氣泡被 埋入照明光所以看不見。 專利文獻1之圖15，係包含藉由攝影部攝影的折曲之 攝影區域（與搬送方向直交的方向之尺寸：5 mm )之動畫 -9" 201033602 例，並沒有藉由攝影部攝影的折曲導致的明暗反轉’隨著 折曲通過，本來應該拍出矩形白區域之照明像隨時間而扭 曲。 專利文獻1之圖1 5，係包含藉由攝影部攝影的折線（ 線缺陷）之攝影區域（與搬送方向直交的方向之尺寸： 200mm)之動畫例。在第1〜第3個圖框（a)〜（c)， 照明像多少扭曲成弓狀，但此程度之弓狀的扭曲並不是起 因於缺陷，而是拉扯成形薄片所產生的。另一方面，在第 @ 4個圖框照明像呈現S字形，扭曲地比較大。照明像扭曲 超過此程度，是因爲在扭曲的位置有折線的緣故。亦即， 可望把照明像扭曲超過此程度之扭曲部分檢測出作爲缺陷 〇 然而，專利文獻1之缺陷檢查裝置，可知於2値化處 理隨著區分亮區域（線狀光源像以及線狀光源像內部之缺 陷區域）與暗區域（背景區域以及背景區域內部之缺陷區 域）之閾値，而有影像資料上之對比很低而無法檢測出缺 @ 陷的情形。 亦即，於專利文獻1之缺陷檢查裝置，缺陷係影像上 之對比（缺陷導致之亮度變化）比較高者的場合，如圖3 之垂直方向（成形薄片搬送方向）亮度輪廓圖所示，缺陷 導致之亮度變化以跨過2値化處理的閾値的方式被觀測到 。更詳細說明之，在線狀光源像中作爲暗區域被觀測到的 缺陷所對應之極小點（圖3之谷部分）之亮度値變得更大 ，且其極小點兩側之極大點的亮度値變得更小。此外，2 -10- 201033602 値化處理之閾値，在線狀光源像以外作爲亮區域被觀測到 的缺陷所對應之極大點（圖3之山部分）之亮度値變得更 大，且其極大點兩側之極小點的亮度値變得更大。而且， 在線狀光源像中作爲暗區域觀測到的缺陷，或是在線狀光 源像之外作爲亮區域觀測到的缺陷也被檢測出。亦即，專 利文獻1之缺陷檢査裝置，可以確實檢測出對比比較高的 缺陷。 φ 另一方面’於專利文獻1之缺陷檢查裝置，缺陷在影 像資料上之對比很低的場合，缺陷部分之垂直方向亮度輪 廓圖與2値化處理之閾値之關係變成圖4之垂直方向亮度 輪廓圖所示的關係，亦即，缺陷導致之亮度變化變成不跨 過2値化處理的閾値的關係。作爲大體的推測，在滿足次 式的場合，（缺陷導致之亮度變化量）<{(線狀光源像 區域之亮度位準）-(背景區域之亮度位準）} / 2…（1 )成爲圖4所示的關係。成爲圖4所示的關係時，會看漏 _ 缺陷。 但是，影像資料上之缺陷的對比很低，即使滿足式（ 1)的場合，圖5的垂直方向亮度輪廓圖所示之缺陷之亮 度變化若是跨過2値化處理的閾値而被觀測到的場合，可 以檢測出缺陷。亦即，專利文獻1之缺陷檢查裝置，即使 在滿足式（1)的場合，亦可藉由缺陷所致之亮度變化與2 値化處理的閾値之關係而可以檢測出缺陷。 進而，專利文獻1之缺陷檢查裝置，使用動畫像，而 在動畫像，缺陷像於1個個畫格內被觀測到接近線狀光源 -11 - 201033602 像、通過線狀光源像內，由線狀光源像遠離之動向。亦g卩 ，即使是影像資料上之缺陷的對比很低，滿足式（1 )的 場合，即使於構成動畫像的複數影像之中有1個，是缺1¾ 之亮度變化與2値化處理的閾値之關係成爲圖5所示的_ 係之影像的話，就可以檢測出缺陷。 如以上所述，在專利文獻1之缺陷檢查裝置，以式（ 1)爲界，隨著缺陷導致之亮度變化量變小，看漏缺陷的 可能性變高。亦即，專利文獻1之缺陷檢查裝置，針對對 _ 比低的缺陷之檢測的確實性，還有改善的餘地。 本發明係有鑑於前述問題點而爲之發明，目的在於提 供可以更確實檢測出種種缺陷，可以提供成形薄片之缺陷 檢査裝置。 〔供解決課題之手段〕 相關於本發明之缺陷檢查裝置，爲了解決前述課題， 係檢測出成形薄片之缺陷的缺陷檢查裝置，其特徵爲具備 @ :複數次攝影前述成形薄片之2次元影像產生複數2次元 影像資料之攝影手段，以對前述成形薄片之被攝影的區域 的一部份投影線狀光源之像的方式，供照明前述成形薄片 之用的線狀光源，以改變前述成形薄片之前述線狀光源之 像被投影的位置的方式，使前述成形薄片與前述線狀光源 之至少一方，移動在與前述線狀光源之長邊方向交叉，且 直交於前述成形薄片的厚度方向之方向上的移動手段，及 從藉由前述攝影手段產生的複數2次元影像資料檢測出線 -12- 201033602 缺陷之線缺陷檢測手段；前述線缺陷檢測手段，係藉由以 函數曲線套合（fitting)前述2次元影像資料之線狀光源 之像的邊緣，把線狀光源之像的邊緣與函數曲線之距離在 第1閾値以上之處所檢測出作爲線缺陷之線缺陷檢測演算 法，或者是針對前述2次元影像資料之線狀光源之像的邊 緣，求出各畫素之附近區域之曲率，把曲率在第2閾値以 上之處所檢測出作爲線缺陷之線缺陷檢測演算法來檢測出 φ 線缺陷。 根據前述構成，可以更確實檢測出線缺陷。 前述缺陷檢查裝置，最好是進而具有從藉由前述攝影 手段產生的複數之2次元影像資料來檢測出點缺陷之點缺 陷檢測手段。藉此，不僅線缺陷也可以檢測出點缺陷。 於前述缺陷檢查裝置，最好是前述點缺陷檢測手段， 係把前述2次元影像資料之依存於沿著一直線上的位置之 亮度變化表示爲亮度輪廓圖，把亮度輪廓圖之數據點群假 φ 設爲以使數據點間之移動時間成爲一定的方式移動之質點 ，而由注目數據點之前2個數據點間之前述質點的速度向 量與前述注目數據點之前3個之數據點間之前述質點的加 速度向量來預測前述注目數據點的亮度値，藉由把預測的 亮度値與實際的亮度値之差在第3閾値以上之處所檢測出 作爲點缺陷之點缺陷檢測演算法，或者使前述2次元影像 資料平滑化，把被平滑化的2次元影像資料與原來的2次 元影像之料之差分求出作爲差分影像資料，把差分影像資 料之亮度値在第4閾値以上之處所及亮度値在第5閾値（ -13- 201033602 第5閾値比第4閾値還要小）以下之處所檢測出作爲點缺 陷之點缺陷檢測演算法來檢測出點缺陷。藉此，與專利文 獻1之技術相較，可以更確實地檢測出對比低的點缺陷（ 例如圖4那樣顯示小的亮度變化之缺陷）。亦即，可以確 實檢測出點缺陷及線缺陷雙方。 相關於本發明之缺陷檢查裝置，爲了解決前述課題， 係檢測出成形薄片之缺陷的缺陷檢查裝置，其特徵爲具備 :複數次攝影前述成形薄片之2次元影像產生複數2次元 @ 影像資料之攝影手段，以對前述成形薄片之被攝影的區域 的一部份投影線狀光源之像的方式，供照明前述成形薄片 之用的線狀光源，以改變前述成形薄片之前述線狀光源之 像被投影的位置的方式，使前述成形薄片與前述線狀光源 之至少一方，移動在與前述線狀光源之長邊方向交叉，且 直交於前述成形薄片的厚度方向之方向上的移動手段，及 從藉由前述攝影手段產生的複數2次元影像資料檢測出點 缺陷之點缺陷檢測手段；其中前述點缺陷檢測手段，係把 〇 前述2次元影像資料之依存於沿著一直線上的位置之亮度 變化表示爲亮度輪廓圖，把亮度輪廓圖之數據點群假設爲 以使數據點間之移動時間成爲一定的方式移動之質點，而 由注目數據點之前2個數據點間之前述質點的速度向量與 前述注目數據點之前3個之數據點間之前述質點的加速度 向量來預測前述注目數據點的亮度値，藉由把預測的亮度 値與實際的亮度値之差在第3閾値以上之處所檢測出作爲 點缺陷之點缺陷檢測演算法，或者使前述2次元影像資料 -14 - 201033602 平滑化，把被平滑化的2次元影像資料與原來的2次元影 像之料之差分求出作爲差分影像資料，把差分影像資料之 亮度値在第4閾値以上之處所及亮度値在第5閾値（第5 閾値比第4閾値還要小）以下之處所檢測出作爲點缺陷之 點缺陷檢測演算法來檢測出點缺陷。 藉此，與專利文獻1之技術相較，可以更確實地檢測 出對比低的點缺陷（例如圖4那樣顯示小的亮度變化之缺 ⑩ 陷）。 前述缺陷檢查裝置，最好是進而具有從藉由前述攝影 手段產生的複數之2次元影像資料來檢測出線缺陷之線缺 陷檢測手段。藉此，不僅點缺陷也可以檢測出線缺陷。 〔發明之效果〕 如以上所述，本發明，發揮可以提供可更確實檢測出 種種缺陷，可以提供成形薄片之缺陷檢査裝置之效果。 Φ 【實施方式】 〔實施型態1〕 參照圖面於以下說明本發明之一實施型態。 相關於本實施型態之缺陷檢查裝置，係檢測成形薄片 的缺陷者。相關於本實施型態之缺陷檢査裝置，適用於透 光性之成形薄片，特別是熱塑性樹脂等樹脂所構成之成形 薄片的檢查。由樹脂所構成之成形薄片，例如可以舉出由 壓出機壓出之熱塑性樹脂通過輥之間隙而於表面賦予平滑 -15- 201033602 或光澤之處理，冷卻於搬送輥上同時藉由拉取輥拉取而成 形者。適用於本實施型態之熱塑性樹脂，例如有甲基丙烯 酸樹脂、甲基丙烯酸甲酯-苯乙烯共聚合物、聚乙烯或聚 丙烯等聚烯烴，聚碳酸酯'聚氯乙烯、聚苯乙烯、聚乙烯 醇、三醋酸纖維素樹脂等。成形薄片，亦可由這些熱塑性 樹脂中僅1種來構成，亦可以是這些熱塑性樹脂之層積複 數種類者（層積薄片）。此外，相關於本實施型態之缺陷 檢查裝置，適於檢查偏光膜或相位差膜等光學膜，特別是 捲成捲筒狀被保管·輸送的長尺寸光學膜。此外，成形薄 片可以是具有任何厚度者，一般而言稱爲薄膜之比較薄的 厚度者，或是一般稱爲板之比較厚者均可。 作爲成形薄片之缺陷之例，可舉出氣泡（成形時產生 者等）、魚眼（fish eye)、異物、輪跡痕、打痕、傷痕 等點缺陷；折曲、折線（由於厚度之差異而產生者等）等 〇 以下根據圖1及圖2說明相關於本實施型態之缺陷檢 查裝置1的構成。圖1係顯示缺陷檢查裝置1之主要部位 的功能方塊圖。圖2係顯示缺陷檢查裝置1的槪觀之模式 圖。又，於圖2，以容易識別重疊於成形薄片的構件的方 式反轉成形薄片表面之明暗而顯示。亦即，圖2之成形薄 片表面的黑色區域，實際上爲明（亮）區域，圖2之成形 薄片表面之白色區域，實際上爲暗區域。 缺陷檢查裝置1，係藉由搬送裝置（移動手段）3把 矩形之成形薄片2搬送於一定方向，同時藉由線狀光源4 -16- 201033602 照明之成形薄片2藉由η個（η爲2以上之整數）之攝影 部（攝影手段）51〜5„攝影複數次而以攝影部51〜5η之各 個產生複數之2次元影像資料，根據產生的2次元影像資 料使解析裝置6檢測出成形薄片2的缺陷者。 缺陷檢測裝置1，具備：搬送成形薄片2的搬送裝置 (移動手段）3，對成形薄片2之攝影區域（藉由攝影部 51〜5η攝影的區域；圖2之成形薄片2表面以虛線表示之 φ 矩形）之一部分被投影線狀光源4之像的方式照明成形薄 片2之用的線狀光源4，複數次攝影線狀光源4的反射像 (作爲來自線狀光源4的直接光藉由成形薄片2反射而到 達攝影部5!〜5η的結果而被形成的線狀光源4之像）及成 形薄片2的反射像（作爲來自線狀光源4的散射.光藉由成 形薄片2反射而到達攝影部51〜511的結果而被形成的成形 薄片2之像）之攝影部5l〜5n，及根據複數2次元影像資 料，藉由影像處理演算法（缺陷檢測演算法）檢測出成形 〇 薄片2的缺陷之解析裝置6。 搬送裝置3，以改變成形薄片2之線狀光源4之像被 投影的位置的方式使成形薄片2搬送於直交於其厚度方向 的方向，特別是於其長邊方向。搬送裝置3，例如具備使 成形薄片2搬送於一定方向之送出輥與承接輥，藉由旋轉 編碼器（rotary encoder )等計測搬送速度。搬送速度例如 被設定爲2m〜12m/分程度。搬送裝置3之搬送速度，藉 由未圖示之資訊處理裝置等來設定及控制。 線狀光源4,係以其長邊方向成爲與成形薄片2的搬 -17- 201033602 送方向交叉的方向（例如與成形薄片2的搬送方向直交的 方向）的方式，且線狀光源4的反射像橫切成形薄片2之 攝影區域，於攝影區域之線狀光源4的反射像之兩側存在 沒有線狀光源像的區域（背景區域）的方式被配置的。線 狀光源4，只要是發出對成形薄片2的組成及性質不造成 影響的光之光源即可，沒有特別限定，例如有螢光燈（特 別是高頻螢光燈）、金屬鹵素燈、鹵素傳送燈等。又，將 線狀光源4，夾著成形薄片2配置在對向於攝影部5!〜5η φ 的位置，藉由攝影部攝影包含線狀光源4的透過像（來自 線狀光源4的直接光透過成形薄片2到達攝影部51〜511而 形成的線狀光源4之像）以及成形薄片2的透過像（來自 線狀光源4的散射光透過成形薄片2到達攝影部51〜511而 形成的成形薄片2之像）之2次元影像亦可。 攝影部之各個，攝影複數次包含線狀光源4的 反射像及成形薄片2之反射像的2次元影像，產生而輸出 複數之2次元影像資料》攝影部5^5««，係以攝影2次元 @ 影像之 CCD ( Charge Coupled Device )或是 CMOS ( Complementary Metal-Oxide Semiconductor)等攝影元件 構成之區域感測器所構成。藉由缺陷檢查裝置1檢測出的 缺陷的尺寸，依存於攝影部51〜5„的分解能，所以配合欲 檢測出的缺陷的尺寸而選定攝影部51〜5„的分解能即可。 又，藉由缺陷檢查裝置1檢測出的缺陷的立體形狀（寬幅 對高度之比），基本上不依存於攝影部Si-Sn的分解能， 所以沒有隨著欲檢測出的缺陷的種類不同而選擇攝影機分 -18- 201033602 解能的必要。 攝影部51〜5n，係以從攝影部SixSn朝向成形薄片2 的攝影區域的中心的方向與成形薄片2的搬送方向夾銳角 的方式被配置的。攝影部5 i〜5n，係以成形薄片2的寬幅 方向（直交於成形薄片2的搬送方向，且直交於成形薄片 2的厚度方向之方向）之全區域被攝影部Si-Sn之至少1 個拍攝到的方式’沿著成形薄片2的寬幅方向並列配置的 φ 。藉由攝影部5!〜5n攝影成形薄片2的寬幅方向之全區域 ，可以檢查成形薄片2之全區域的缺陷。 攝影部51〜5n之攝影間隔（圖框速率），可以是固定 的，也可以隨著使用者操作攝影部自身而改變，或 藉由使用者進行操作而可變更被連接於攝影部5i〜5n的資 訊處理裝置（未圖示；可省略）亦可。此外，攝影部5i〜 5n之攝影間隔，亦可爲數位相機之連續攝影的時間間隔之 數分之一秒等，爲了提高檢查的效率，最好是以短的時間 ❿ 間隔，例如一般的動畫資料之圖框速率之1/3 0秒等。 此處，各攝影部攝影1枚2次元影像之後至攝影次一 2次元影像爲止的期間內成形薄片2被搬送的距離（搬送 距離），係被設定爲沿著成形薄片2的搬送方向之攝影區 域的長度之至少l/m(m爲2以上）。藉此’包含成形薄 片2之同一處所的2次元影像被攝影m次。m最好是比2 更充分地大。藉由增加成形薄片2的同一處所之攝影次數 ，可以高精度地檢査缺陷。 解析裝置6，如圖1所示，接收由各攝影部5!〜5„輸 -19- 201033602 出的複數2次元影像資料’輸出根據複數之2次元影像資 料檢測缺陷的檢測結果（檢査結果）之線缺陷用影像解析 部（線缺陷檢測手段）6 1 1〜ό 1 n以及點缺陷用影像解析部 (點缺陷檢測手段）02,〜62n ’及顯示檢測結果（檢查結 果）之顯示部64，與統括控制這些各部之控制CPU63。 在各攝影部SixSn產生的複數之2次元影像資料，分 別被輸入至線缺陷用影像解析部61!〜61n及點缺陷用影像 解析部62,〜62n。 線缺陷用影像解析部點缺陷用影像解析部 62 !〜62n之各個’藉由線缺陷檢測演算法，由成形薄片2 上之線狀光源像位置不同的複數之（複數圖框之）2次元 影像資料檢測出缺陷’將其結果作爲檢査結果而輸出。點 缺陷用影像解析部62 62n之各個，藉由點缺陷檢測演算 法’由成形薄片2上之線狀光源像位置不同的複數之（複 數圖框之）2次元影像資料檢測出缺陷，將其結果作爲檢 查結果而輸出。線缺陷用影像解析部及點缺陷用 影像解析部根據由成形薄片2上之線狀光源像 位置不同的複數之2次元影像資料判定是否有缺陷，所以 比從前的缺陷檢査裝置更能確實檢測出缺陷。 針對線缺陷檢測演算法及點缺陷檢測演算法，將於後 段說明。線缺陷檢測演算法及點缺陷檢測演算法之參數， 亦可爲固定的，亦可藉由使用者進行操作而改變被連接於 線缺陷用影像解析部611〜61„及點缺陷用影像解析部62t 〜62n2資訊處理裝置（未圖示；可省略）。 -20- 201033602 線缺陷用影像解析部611〜61n，在藉由線缺陷檢測演 算法由m個之複數的2次元影像資料之中的L個（LSm )以上檢測出線缺陷的場合’把有線缺陷之結果作爲檢査 結果輸出，其他場合，把沒有線缺陷的結果作爲檢查結果 輸出亦可，或者藉由線缺陷檢測演算法，由m個複數2次 元影像資料之中L個以上檢測出線缺陷的場合，把線缺陷 位置的資訊作爲檢查結果輸出，在其他場合不輸出檢查結 _ 果者亦可。2次元影像資料的個數m爲3以上，L爲2以 上的場合，藉由線缺陷檢測演算法檢測出線缺陷的2次元 影像資料的個數比L個還要少的場合，該線缺陷檢測結果 被視爲誤報（原本沒有缺陷但誤認爲是缺陷而被檢測出者 )而被排除。藉此，可以減低誤報。又，線缺陷位置之資 訊作爲檢查結果輸出的場合，必須使用可以求出線缺陷位 置的線缺陷檢測演算法。 點缺陷用影像解析部62 !〜62n，在藉由點缺陷檢測演 φ 算法由m個之複數的2次元影像資料之中的L個（LSm )以上檢測出點缺陷的場合，把有點缺陷之結果作爲檢査 結果輸出，其他場合，把沒有點缺陷的結果作爲檢查結果 輸出亦可，或者藉由點缺陷檢測演算法，由m個複數2次 元影像資料之中L個以上檢測出點缺陷的場合，把點缺陷 位置的資訊作爲檢査結果輸出，在其他場合不輸出檢査結 果者亦可。2次元影像資料的個數m爲3以上，L爲2以 上的場合，藉由點缺陷檢測演算法檢測出點缺陷的2次元 影像資料的個數比L個還要少的場合，該點缺陷檢測結果 -21 - 201033602 被視爲誤報（原本沒有缺陷但誤認爲是缺陷而被檢測出者 )而被排除。藉此，可以減低誤報。又，點缺陷位置之資 訊作爲檢查結果輸出的場合’必須使用可以求出點缺陷位 置的點缺陷檢測演算法。 控制CPU6 3，統合線缺陷用影像解析部όΗ-όΙη及點 缺陷用影像解析部62 輸出的檢查結果製作對應於 成形薄片2的全區域之檢查結果資訊，使記憶於未圖示的 記憶裝置，同時使顯示於顯示部64。作爲對應於成形薄片 @ 2之全區域的檢查結果資訊，可舉出顯示在成形薄片2之 全區域有還是沒有缺陷之資訊，或成形薄片2的全區域之 缺陷地圖等。製作對應於成形薄片2的全區域之檢查結果 資訊時，以線缺陷用影像解析部及點缺陷用影像 解析部62 i〜62n之至少一方檢測出缺陷的場合，視爲存在 缺陷而製作檢查結果資訊。 控制CPU63，在作爲對應於成形薄片2的全區域之檢 查結果資訊’製作出成形薄片2的全區域之缺陷地圖的場 ❹ 合’線缺陷用影像解析部點缺陷用影像解析部 之各個’把2次元影像資料上的座標位置變換爲 成形薄片2上之座標位置而產生缺陷位置資訊，將此缺陷 位置資訊輸出至控制CPU63。作爲線缺陷用影像解析部 61ι〜6“及點缺陷用影像解析部621〜62n之各個之作標變 換處理’例如可以使用記載於專利文獻1之段落〔〇〇37〕 〜〔0041〕及〔 00 50〕〜〔〇053〕之處理。把缺陷地圖之 資訊輸出至標記（marking)裝置（未圖示）及資訊處理 -22- 201033602 裝置（未圖示），標記裝置根據缺陷地圖在成形薄片2上 標示缺陷位置的作法亦可採用。此標記裝置，例如具有沿 著成形薄片2的寬幅方向設置之臂，及具有筆等之標記頭 ，可以藉由標記頭於臂上在成形薄片2的寬幅方向上往復 移動而在成形薄片2上之任意位置進行標記。此被標記的 缺陷位置之資訊，例如，可以利用於在把成形薄片2裁斷 爲複數特定尺寸之片狀品後，把這些片狀品分爲正常品與 φ 缺陷品的處理等。 又，在前述實施型態，係固定線狀光源4而搬送成形 薄片2，但是改變成形薄片2之線狀光源4之像被投影的 位置亦可。亦即，固定成形薄片2而移動線狀光源4亦可 ，或是使成形薄片2與線狀光源4雙方以不同的方向或不 同的速度移動亦可。固定成形薄片2而移動線狀光源4的 場合’使攝影部Si-Sn與線狀光源4以同方向同速度移動 者較佳。藉此，可以得到包含線狀光源像之複數2次元影 φ 像資料。固定成形薄片2而移動線狀光源4的方法，係以 搬送裝置3拉伸成形薄片2而可以避免線狀光源像扭曲， 但一次可以檢査的成形薄片2的長度受限於線狀光源4之 可移動範圍所對應的長度，所以爲了有效率地檢查長尺寸 的成形薄片2’最好是如前述實施型態那樣搬送成形薄片 2 » 此外，在前述實施型態，線缺陷用影像解析部6h〜 61„及點缺陷用影像解析部62l〜62n，雖係根據從相同之 攝影部5 :〜5n所得到的2次元畫像資料而檢測出缺陷，但 -23- 201033602 是，亦可根據不同的攝影部所得到的2次元影像資料分別 檢測缺陷。藉此，可以把攝影部5 !〜5n之攝影條件（與成 形薄片2之距離、成形薄片2之搬送方向與攝影方向之夾 角等）設爲適於欲檢測出的缺陷之條件。攝影在線缺陷用 影像解析部使用的2次元影像資料之攝影部，與 攝影在點缺陷用影像解析部使用的2次元影像資 料之用的攝影部相比，以與成形薄片之距離較遠，成形薄 片2之搬送方向與攝影方向之夾角爲較窄者較佳。藉此， 參 可以在最適攝影條件攝影點缺陷及線缺陷雙方，所以可進 而精度良好地檢測出點缺陷及線缺陷雙方。 此外，在前述實施型態，以各攝影部51〜5n攝影的2 次元影像藉由線缺陷用影像解析部及點缺陷用影 像解析部621〜62„分散處理，但根據攝影部51〜5n攝影 的2次元影像的相對位置由在以各攝影部5l〜5n攝影的η 枚2次元影像合成包含成形薄片2的寬幅方向之全區域的 1枚全寬幅影像，根據全寬幅影像以1個線缺陷用影像解 @ 析部及1個點缺陷用影像解析部檢測缺陷亦可。作爲由η 枚2次元影像合成i枚全寬幅影像的方法，例如可以使用 記載於專利文獻1之段落〔0050〕之方法》 其次，說明在線缺陷用影像解析部6 1 !〜6 1 n及點缺陷 用影像解析部62 ,〜62n使用之線缺陷檢測演算法及點缺陷 檢測演算法。作爲線缺陷檢測演算法及點缺陷檢測演算法 ’可以使用以下7種缺陷檢測演算法A〜G。又，在以下 的說明，2次元影像資料之亮度値（畫素値），以自然數 -24 - 201033602 來表示。 〔缺陷檢測演算法A〕 針對缺陷檢測演算法A，根據圖6說明如下。圖6 ( a )係以攝影部Si-Sni 1個所產生的多値2次元影像資料 (以下稱爲原影像資料）之例，影像之上側爲搬送方向下 游側，影像之下側爲搬送方向上游側。於圖6 ( a )，延伸 ❹ 於中央之橫方向的帶狀白區域爲線狀光源像，存在於線狀 光源像的內部之暗區域，以及存在於線狀光源像附近的小 的白區域，爲缺陷。 在此缺陷檢測演算法A，針對以攝影部產生的 複數枚原影像資料之各個，進行以下之處理。 首先，把原影像資料，分割爲沿著縱方向（成形薄片 2的搬送方向）之1行行的畫素列之資料（表示亮度値（ 畫素値）以及位置之資料；亮度輪廓圖；1次元影像資料 參 ）。 其次’針對各畫素列之資料，如以下所述進行由一端 (圖6(a)之上端）朝向另一端（圖6(a)之下端）探 索邊緣之第1邊緣判定處理。首先，把畫素列之一端側起 算第2個畫素作爲注目畫素，對注目畫素判定注目畫素亮 度値是否比鄰接於一端側的鄰接畫素亮度値還要小閾値 T1以上。當被判定爲注目畫素之亮度値比鄰接畫素的亮 度値小閾値T1以上的話（亦即，鄰接畫素的亮度値爲La ，注目畫素之亮度値爲Lb的話，La-Lb 2 T1 )，判定爲鄰 -25- 201033602 接畫素爲第1邊緣，記錄第1邊緣的位置（鄰接畫素的位 置），結束處理對象之畫素列的資料處理，其他的場合， 使注目畫素朝向另一端1畫素1畫素地移動，同時反覆前 述判定直到比起鄰接畫素亮度之注目畫素的亮度値被判定 爲比閾値T1以上，在判定爲注目畫素之亮度値比鄰接畫 素之亮度値還要小閾値T 1以上的場合，判定鄰接畫素爲 第1邊緣，記錄第1邊緣的位置（鄰接畫素的位置），結 束處理對象之畫素列的資料之處理。又，閾値T1爲任意 @ 之自然數，亦可爲亮度値之最小單位。閾値T 1爲亮度値 的最小單位的場合，前述判定成爲只判斷注目畫素之亮度 値是否比鄰接畫素之亮度値還要小。 其次，針對.各畫素列之資料，如以下所述進行由另一 端朝向一端探索邊緣之以下之第2邊緣判定處理。首先， 把另一端側起算第2個畫素作爲注目畫素，對注目畫素判 定注目畫素之亮度値是否比鄰接於另一端側的鄰接畫素之 亮度値還要大閾値T2以上。當被判定爲注目畫素之亮度 ❹ 値比鄰接畫素的亮度値大閾値T2以上的話（亦即，鄰接 畫素的亮度値爲La，注目畫素之亮度値爲Lb的話，Lb-La 2T2) ’判定爲注目畫素爲第2邊緣，記錄第2邊緣的 位置（注目畫素的位置），結束處理對象之畫素列的資料 處理’其他的場合，使注目畫素朝向一端1畫素1畫素地 移動’同時反覆前述判定直到注目畫素的亮度値被判定爲 比鄰接畫素之亮度値大閾値T2以上，在判定爲注目畫素 之亮度値比鄰接畫素之亮度値還要大閾値T2以上的場合 -26- 201033602 ，判定注目畫素爲第2邊緣，記錄第2邊緣的位置（注目 畫素的位置），結束處理對象之畫素列的資料之處理。又 ，閾値T2爲任意之自然數，亦可爲亮度値之最小單位。 閾値T2爲亮度値的最小單位的場合，前述判定成爲只判 斷注目畫素之亮度値是否比鄰接畫素之亮度値還要大閾値 以上。 藉由這些第1邊緣判定處理檢測出的第1邊緣之例於 φ 圖6 ( a )以「△」顯示，以及藉由第2邊緣判定處理檢測 出的第2邊緣之例於圖6(a)以「〇」顯示。由圖6(a )可知，在沒有缺陷的區域，於線狀光源像之邊緣以外不 存在邊緣，所以第1邊緣及第2邊緣，一致於線狀光源像 之另一端側的邊緣（在圖6(a)之例爲下側之邊緣），相 互一致。另一方面，由圖6(a)可知，在有缺陷的區域（ 白區域及黑區域），第1邊緣及第2邊緣之至少一方，一 致於缺陷區域的邊緣，比起線狀光源像之另一端側的邊緣 φ 更往邊緣探索開始側偏移，所以第1邊緣與第2邊緣變成 在離開的位置。 此處，接下來針對各畫素列之資料，求出第1邊緣起 至第2邊緣爲止的距離（畫素數）作爲邊緣間距離。把所 求得的邊緣間距離對於畫素列之位置（橫方向的座標）繪 圖之輪廓圖顯示於圖6(b)。接著，存在此邊緣間距離爲 閾値T3以上之畫素列的話，判定爲有缺陷。又，閾値T3 爲任意之自然數’亦可爲1個畫素。閾値T3爲1個畫素 的場合’邊緣間距離非零的畫素列被判定爲包含缺陷的畫 -27- 201033602 素列。閾値T3只要因應於容許的缺陷的尺寸而適當決定 即可，多値之次元影像資料爲256灰階（亮度値0〜255; 8位元）之2次元影像資料的場合，例如爲3者較佳。 又，在圖6之例把上端作爲一端（第1邊緣探索開始 側），但把畫素列之哪個端作爲一端是任意的，把下端當 成一端亦可。在該場合，在沒有缺陷的區域，第1邊緣及 第2邊緣，一致於線狀光源像之上側的邊緣。 此缺陷檢測演算法A，可具有某種程度確實性地檢測 ^ 出種種點缺陷。但是氣泡或魚眼等微小的點缺陷的檢出確 實性，並不高。另一方面，此缺陷檢測演算法A不適合線 缺陷的檢測。以下，把此缺陷檢測演算法A稱爲「邊緣輪 廓圖法1」。 〔缺陷檢測演算法B〕 缺陷檢測演算法B，係以函數曲線套合2次元影像資 料之線狀光源像之邊緣，把線狀光源之像的邊緣與函數曲 @ 線之距離超過閾値T 5 (第1閾値）的處所檢出作爲缺陷 者。 針對缺陷檢測演算法B，根據圖7說明如下。圖7 ( a )係以攝影部5 !〜5n之1個所產生的原影像資料之例’影 像之上側爲搬送方向下游側，影像之下側爲搬送方向上游 側。於圖7 ( a )，延伸於中央之橫方向的帶狀白區域爲線 狀光源像，線狀光源像的下側邊緣之局部扭曲的部分（非 平滑的地方）爲缺陷。 -28- 201033602 在此缺陷檢測演算法B，針對以攝影部5 !〜5n產 複數枚原影像資料之各個，進行以下之處理。 首先，由原影像資料求出線狀光源像的邊緣之至 方。所求得的線狀光源像之邊緣之例於圖7(a)以「 表示。在圖7之例求的是線狀光源像的下側邊緣，但 狀光源像的上側邊緣亦可，求線狀光源像的上側邊緣 側邊緣雙方亦可。 φ 作爲求出線狀光源像的邊緣的方法，可以使用習 邊緣抽出過濾器（例如索貝爾濾波器（Sobel Filter) 出邊緣，把強度強的邊緣作爲線狀光源像的邊緣的方 或把2次元影像資料分割爲1行行的畫素列之資料而 各畫素列之資料求出強的邊緣作爲線狀光源像之邊緣 法，以記載於專利文獻1之〔〇〇57〕之方法（進行2 及標記，被標記的區域之中面積比特定値還要大的區 爲線狀光源像之區域而抽出之方法）等。此處，作爲 Φ ，說明把原影像資料分割爲1行行的畫素列之資料而 各畫素列之資料求取強的邊緣而作爲線狀光源像的邊 方法。首先，把原影像資料，分割爲沿著縱方向（成 片2的搬送方向）之1行行的畫素列之資料。其次， 各畫素列之資料，如以下所述進行由一端（圖7 ( a ) 端）朝向另一端（圖7(a)之下端）探索邊緣處理。 ，把一端側起算第2個畫素作爲注目畫素，對注目畫 定注目畫素之亮度値是否比鄰接於一端側的鄰接畫素 度値小了閾値T4 ( T4爲自然數）以上（亦即，鄰接 生的 少一 〇」 求線 及下 知之 )抽 法， 且由 的方 値化 域作 一例 且由 緣之 形薄 針對 之上 首先 素判 之亮 畫素 -29 - 201033602 的亮度値爲La，注目畫素之亮度値爲Lb的話，La-Lb g T4)。爲了僅檢測出強的邊緣’此時之閾値T4爲比較大 之値。當被判定爲注目畫素之亮度値比鄰接畫素的亮度値 小了閾値Τ4以上的場合，判定鄰接畫素爲線狀光源像之 邊緣，記錄線狀光源像的邊緣的位置（鄰接畫素的位置） ，結束處理對象之畫素列的資料處理，其他的場合，使注 目畫素朝向另一端1畫素1畫素地移動，同時反覆前述判 定直到注目畫素的亮度値被判定爲比鄰接畫素亮度小了閾 値Τ4以上，在判定注目畫素之亮度値比鄰接畫素之亮度 値還要小閾値Τ4以上的場合，判定鄰接畫素爲線狀光源 像之邊緣，記錄線狀光源像的邊緣的位置（鄰接畫素的位 置），結束處理對象之畫素列的資料之處理。 接著，把求得的線狀光源像之邊緣之列，套合於以函 數表現之平滑的曲線（以函數曲線進行套合（fitting )) ，求出套合曲線（函數曲線）。使用於套合的函數，可以 舉出η次函數（η爲2以上），高斯函數、羅倫茲函數， 沃格特（Voigt )函數、這些函數的組合等，其中以η比較 小的η次函數，例如4次函數爲較佳。此外，作爲進行套 合時使用的套合的評估方法，例如可以使用最小平方法。 其次，把2次元影像資料分割爲沿著縱方向（成形薄 片2的搬送方向）之1行行的畫素列之資料，針對各畫素 列之資料，求出由套合曲線至線狀光源像的邊緣爲止的距 離（畫素數）作爲套合度。。把所求得的套合度對於畫素 列之位置（橫方向（直交於成形薄片2的搬送方向，且直 -30- 201033602 交於成形薄片2的厚度方向之方向）的座標）繪圖之輪廓 圖顯示於圖7(b)。接著，存在此套合度爲閩値T5以上 之畫素列的話，判定該畫素列之線狀光學像的邊緣的位置 有缺陷。藉此，可以判定有無缺陷，此外，也可以求出缺 陷的位置。如以上所述進行，可以檢測出作爲線狀光源像 的邊緣之局部扭曲（在邊緣附近之細微的線狀光源像之扭 曲）而顯現之線缺陷。又，使用於前述判定的閾値T5爲 φ 任意之自然數，亦可爲1個畫素。閾値T5爲1個畫素的 場合，若存在套合度不是零的畫素列的話，被判定爲有缺 陷。閾値T5只要因應於容許的缺陷的尺寸而適當決定即 可，多値之次元影像資料爲25 6灰階之2次元影像資料的 場合，以4較佳。 又，在此缺陷檢測演算法B，除了判定缺陷的有無以 外，亦可求缺陷位置。在該場合，抽出套合度爲閾値T5 以上之畫素列，把被抽出的畫素列之線狀光源像的邊緣與 φ 套合曲線之間的畫素的位置作爲缺陷位置求出即可。 此缺陷檢測演算法B，可具有高確實性地檢測出種種 線缺陷。另一方面，此缺陷檢測演算法B不適合點缺陷的 檢測。以下，把此缺陷檢測演算法B稱爲「邊緣輪廓圖法 2 j 〇 〔缺陷檢測演算法C〕 缺陷檢測演算法C，係使2次元影像資料平滑化，把 被平滑化的次元影像資料與原來的2次元影像資料之差分 -31 - 201033602 料之亮度値比閩値 還大的處所，以及 比Τ6Β更小的任意 法C。此缺陷檢測 直交於成形薄片2 度方向之方向）之 的明暗變化空間頻 分，把高頻成分之 下的部分作爲缺陷 以上之整數）之橫 $像資料平滑化（ 的影像資料。藉此 高頻區域，針對橫 方向亮度變化之低 橫方向平滑化過濾 過濾器、平均化過 述被平滑化的影像 僅剩下原影像資料 作爲差分影像資料求出，把差分影像資 T6B (第4閾値；T6B爲任意之正數） 亮度値在閾値T6D (第5閾値；T6D係 正數）以下之處所作爲缺陷檢測出者。 以下，進而詳細說明缺陷檢測演算 演算法C，係與線狀光源像的橫方向（ 的搬送方向，且直交於成形薄片2的厚 明暗變化比較，利用缺陷導致之橫方向 率高的情形，抽出原影像資料之高頻成 亮度値在閾値T6B以上或閩値T6D以 檢測出者。 (1 )首先，使用1行η列（η爲3 方向平滑化過濾器（行列），使原景 smoothing )於橫方向，得到被平滑化| ，除去原影像資料之橫方向亮度變化之 方向亮度變化僅剩下低頻成分（剩下橫 頻成分及縱方向亮度變化）。作爲前述 器，可以使用高斯過濾器等加權平均化 濾器等。又，η最好爲3。 (2 )其次，由原影像資料減去前 資料（減算各畫素之亮度値）。藉此， 之橫方向亮度變化之高頻成分。 (3 )接著對藉由減算得到的影像資料使用3x3畫素 之平滑化過濾器（操作子，operator )進行平滑化。藉由此 201033602 平滑化，除去雜訊，剩下雜訊以外之高頻成分。作爲前述 平滑化過濾器，最好使用雙邊過濾器（bilateral filter)或 正中過濾器（median filter)等那樣進行保存邊緣的平滑 化者。 (4)接著，由原影像資料求出線狀光源像之上側邊 緣（搬送方向下游側之邊緣）及下側邊緣（搬送方向上游 側之邊緣）。求出線狀光源像之邊緣的方法，與關於缺陷 φ 檢測演算法B所說明者相同，所以省略說明。接著，把原 影像資料之成形薄片2的搬送方向爲X軸，構成上側邊緣 的全畫素之X座標値中求出最小値Min，由構成下側邊緣 的全畫素之X軸標値之中求出最大値Max。接著，由最小 値Min減去最大値Max之値視爲線狀光源像之寬幅W， 把X座標値由最大値Max至最小値Min爲止的區域僅往 外側移動寬幅W擴展的區域定義爲檢査區域。亦即，X座 標値在Max- ( Min-Max)以上而Min+ ( Min-Max)以下之 φ 座標定義爲檢查區域。此處理，係把檢查對象區域僅集中 到線狀光源像及其附近區域。此外，使檢査區域比X座標 値由最大値Max至最小値Min爲止的區域更往外側擴展 ，是爲了包含線狀光源像之多少包含扭曲而作爲檢查區域 〇 (5 )其次，由前述（3 )之被施以平滑化後之影像資 料（雜訊以外之高頻成分）之檢查區域內的畫素之亮度値 ，藉由次式，決定明亮（bright )側（亮度高之側）之閩 値T6B以及暗（dark)側（亮度低之側）之閩値T6D。 -33- 201033602 Τ6Β=(檢查區域內之平均亮度値）+ (檢查區域內之 亮度値之標準差）xk T6D =(檢査區域內之平均亮度値）-(檢查區域內之 亮度値之標準差）xk (k表正數之參數） 又，k之値，只要因應於容許的缺陷的尺寸而適當決 定即可，例如爲1.5、3、4.5等。 (6)其次，針對被施以前述（3)之平滑化後的影像 資料之檢査區域內的全畫素，進行判定其亮度値在閾値 T6B以上或閾値T6D以下之處理（閾値處理），把閾値 T6B以上或閾値T6D以下的畫素，抽出作爲缺陷部位。藉 此，可以判定有無缺陷，此外，也可以求出缺陷的位置。 又，在以攝影部51〜5„產生的原影像資料所包含的雜 訊很少的場合，亦可省略（3)之平滑化處理。此外，在 沒有必要把檢査對象區域集中到僅線狀光源像及其附近區 域的場合，亦可以省略（4)之定義檢査區域的處理，而 把（5 ) ( 6 )之處理，對影像資料全體進行。 此缺陷檢測演算法C，可以具有高確實性地檢測出氣 泡或魚眼等包含微小缺陷之各種點缺陷。另一方面，此缺 陷檢測演算法C不適合線缺陷的檢測。但是，針對處理時 間’比起此缺陷檢測演算法C，其他缺陷檢測演算法比較 短（缺陷檢測演算法C的處理時間，例如每1圖框爲 40ms程度）。以下，把此缺陷檢測演算法C稱爲「高通 過濾器（hipass fi 11er )法」。 201033602 〔缺陷檢測演算法D〕 針對缺陷檢測演算法D，根據圖8及圖9說明如下。 圖8係以攝影部1個所產生的原影像資料之例， 影像之上側爲搬送方向下游側，影像之下側爲搬送方向上 游側。於圖8，延伸於中央之橫方向的帶狀白區域爲線狀 光源像，存在於線狀光源像的內部之暗區域，以及存在於 φ 線狀光源像附近的小的白區域，爲缺陷。於圖8，線狀光 源像的上方及下方之曲線，顯示檢查對象區域的上限及下 限。 在此缺陷檢測演算法D，針對以攝影部Si-Sn產生的 複數枚原影像資料之各個，進行以下之處理。 首先，把原影像資料，分割爲沿著縱方向（成形薄片 2的搬送方向）之1行行的畫素列之資料求出表示各畫素 列之依存於位置的亮度値變化之數據點列作爲垂直方向亮 φ 度輪廓圖。所求得的垂直方向亮度輪廓圖之例顯示於圖9 。此例，係關於圖8之箭頭所示的位置之畫素列的垂直方 向亮度輪廓圖，y係以下方向（以圖8之箭頭所示的方向 :與成形薄片2的搬送方向相反的方向）爲y軸時之y座 標。 其次，針對各畫素列之垂直方向亮度輪廓圖，求出谷 部分之深度（參照圖8 )。亦即，首先針對各畫素列之垂 直方向亮度輪廓圖，求出所有的極大點及極小點，針對求 得的所有的極小點，把該極小點之亮度値（極小値）與最 -35- 201033602 靠近該極小點的極大點之亮度値（極大値）之差，作爲谷 部分之深度求出。所求得的谷的部分的深度爲閾値T7以 上（T7爲正數）的話，其谷部分被判定爲有缺陷。閾値 T7只要因應於容許的缺陷的尺寸而適當決定即可，多値 之次元影像資料爲25 6灰階之2次元影像資料的場合，例 如爲0.25x255較佳。 此缺陷檢測演算法D，處理時間比較短。此缺陷檢測 演算法D，可具有某種程度確實性地檢測出種種點缺陷。 特別是適合於線狀光源像的邊緣附近之產生局部的明暗反 轉的點缺陷的檢測。但是點缺陷及包含其附近的影像有必 要是高對比，氣泡或魚眼、胎痕等微小的點缺陷的檢出確 實性並不高。另一方面，此缺陷檢測演算法D不適合線缺 陷的檢測。以下，把此缺陷檢測演算法D稱爲「峰値法」 〔缺陷檢測演算法E〕 φ 缺陷檢測演算法E，係把依存於2次元影像資料之沿 著一直線上的位置之亮度變化表示爲亮度輪廓圖，把亮度 輪廓圖之數據點群假設爲數據點間之移動時間成爲一定的 方式移動之質點，在注目數據點之前2個數據點間之前述 質點的速度向量與前述注目數據點之前3個數據點間之前 述質點的加速度向量來預測前述注目數據點之亮度値’把 預測的亮度値與實際亮度値之差比閾値T8還大（第3閾 値，T8爲自然數）的處所作爲缺陷檢測出者。 -36- 201033602 針對缺陷檢測演算法E，根據圖1 0及圖1 1說明如下 。此缺陷檢測演算法Ε，係提高峰値法的精度者，替代谷 之深度而根據實測値與預測値之差檢測出缺陷。 在此缺陷檢測演算法Ε，針對以攝影部5 !〜5η產生的 複數枚原影像資料之各個，進行以下之處理。 首先，與峰値法同樣，求出各畫素列之垂直方向亮度 輪廓圖。所求得的垂直方向亮度輪廓圖之例以亮度値爲X φ 軸，顯示於圖10。此垂直方向亮度輪廓圖之圓圈部，係以 此缺陷檢測演算法Ε欲檢測出的對應於缺陷的輪廓圖。 其次，針對各畫素間的垂直方向亮度輪廓圖，假設以 使鄰接的數據點間之移動時間不拘於數據點間的距離成爲 一定的方式，由數據點列之一端往另一端移動之質點。而 前述質點，如圖11所示，由數據點c往鄰接的數據點b， 由數據點b往鄰接的數據點a，由數據點a往鄰接的數據 點d移動。此外，數據點d爲對應於注目畫素之數據點。 〇 接著，求出數據點d之前質點所通過的3個數據點a 〜c之質點的速度向量及加速度向量。亦即，根據前述移 動時間、數據點d之前質點通過的2個數據點a與b的座 標（X座標、y座標），求出數據點b至數據點a爲止的 區間之前述質點的速度向量。進而，根據前述移動時間、 數據點d之前質點通過的數據點b及c的座標（X座標、y 座標）’求出數據點c至數據點b爲止的區間之前述質點 的速度向量’根據數據點b至數據點a爲止的區間之前述 質點的速度向量’與數據點c至數據點b爲止的區間之前 -37- 201033602 述質點的速度向量，求出數據點C至數據點a爲止的區間 之前述質點的加速度向量。接著，由數據點b至數據點a 爲止的區間之前述質點的速度向量，與數據點c至數據點 a爲止的區間之前述質點的加速度向量，預測數據點d的 座標（位置）。 求出如此預測的數據點d的X座標（亮度値），與數 據點d之實際（實測）的X座標（亮度値）之差，若這些 之差達閾値T8以上的話把對應於數據點d之畫素作爲缺 陷部位予以抽出。藉此，可以判定有無缺陷，此外，也可 以求出缺陷的位置。閾値T8只要因應於容許的缺陷的尺 寸而適當決定即可，多値之次元影像資料爲25 6灰階之2 次元影像資料的場合，例如以20較佳。 此缺陷檢測演算法E，可具有高確實性地檢測出種種 點缺陷。以下，把此缺陷檢測演算法E稱爲「峰値法2」 〔缺陷檢測演算法F〕 針對缺陷檢測演算法F，根據圖12說明如下。圖12 (a)係以攝影部Si-Sn之1個所產生的原影像資料之例 ，影像之上側爲搬送方向下游側，影像之下側爲搬送方向 上游側。於圖12(a)，延伸於中央之橫方向的帶狀白區 域爲線狀光源像，線狀光源像的下側邊緣之局部扭曲的部 分（對水平線之傾斜很大之處）爲缺陷。 在此缺陷檢測演算法F，針對以攝影部5】〜5n產生的 -38- 201033602 複數枚原影像資料之各個，進行以下之處理。 首先，由原影像資料求出線狀光源像的邊緣之至少一 方。所求得的線狀光源像之邊緣之例於圖12(a)以「〇 」表示。在圖12之例求的是線狀光源像的下側邊緣，但 求線狀光源像的上側邊緣亦可，求線狀光源像的上側邊緣 及下側邊緣雙方亦可。求出線狀光源像之邊緣的方法，與 關於缺陷檢測演算法B所說明者相同，所以省略說明。 φ 其次，以橫方向爲X軸，縱方向爲y軸，二次微分線 狀光源像之邊緣曲線（邊緣輪廓圖）y=f(x)，而求出 二次微分輪廓圖。所求得的二次微分輪廓圖之例顯示於圖 12(b) ° 接著，針對線狀光源像之邊緣的各畫素，判定二次微 分是否在閾値T9 (T9爲正數）以上，把二次微分在閾値 T9以上之畫素（高頻之處）判定爲缺陷部位。藉此，可 以判定有無缺陷，此外，也可以求出缺陷的位置。閾値 φ T9只要因應於容許的缺陷的尺寸而適當決定即可。 此缺陷檢測演算法F，適合線狀光源像的邊緣作爲局 部彎曲而呈現的線缺陷之檢測。此缺陷檢測演算法F，缺 陷檢測能力不太高。以下，把此缺陷檢測演算法F稱爲「 邊緣曲線法1」。 〔缺陷檢測演算法G〕 缺陷檢測演算法G ’係針對2次元影像資料之線狀光 源像之邊緣，求出各畫素之附近區域（附近2N + 1畫素之 -39- 201033602 範圍）之曲率，區域在閾値T10 (第2閾値；T10爲正數 )以上之處所檢出作爲缺陷者。 針對缺陷檢測演算法G’根據圖13(a)〜圖13(c )說明如下。 在此缺陷檢測演算法G，針對以攝影部5,〜5n產生的 複數枚原影像資料之各個，進行以下之處理。 首先，由原影像資料求出線狀光源像的邊緣之至少一 方。所求得的線狀光源像之邊緣之例顯示於圖.1 3 ( a )〜 圖13(c)。求出線狀光源像之邊緣的方法，與關於缺陷 檢測演算法B所說明者枏同，所以省略說明。 其次，針對線狀光源像之邊緣的曲線，求出在各點（ 各畫素）之曲率。求曲率之方法，沒有特別限定，亦可使 用數學上既定的數學式進行計算的方法，在這樣的方法處 理時間會變長，所以用以下之方法近似求出曲率是比較好 的。 (1) 對邊緣上之注目畫素（圖13(a)〜圖13(c) 之黑色點）左右（或者前後）每隔N個畫素（圖13(a) 〜圖13(c)之白色點）與注目畫素所構成的範圍（注目 畫素附近2N+1畫素之範圍）作爲計算對象範圍（N爲自 然數）。N，只要因應於容許的缺陷的尺寸而適當決定即 可’例如最好爲3 0。圖1 3 ( a )〜圖13 ( c )之例，係N 爲3的場合。 (2) 接著以直線連結計算對象範圍的兩端之畫素。 (3) 跨計算對象範圍之全畫素，由該直線求出預測 -40- 201033602 亮度値’求出對預測亮度値之實際亮度値（邊緣曲線上之 亮度値）之增量，積算該增量或者該增量之絕對値。藉由 此處所得到的積算値，可以充分近似注目畫素附近2N+ 1 畫素的範圍之曲率（可得到與使用數學上既定的數學式計 算之曲率幾乎相同的曲率値）。此處，在使用增量之積算 値之構成，係如圖13(c)那樣在計算對象範圍內發生有 時往直線之上行進有時往直線下行進的微小的亮度値變化 φ 的場合’抵銷這些變化故意忽視，而求出曲率的近似値。 另一方面，在使用增分的絕對値之積算値的構成，即使那 樣的變化發生的的場合，也包含那樣的變化求出曲率的近 似値。如果想把圖1 3 ( c )那樣在計算對象範圍內有時往 直線之上行進有時往直線下行進的微小的亮度値變化作爲 缺陷檢測出來的話，只要使用增量的絕對値之積算値構成 即可。相反的，容許這樣的變化而並不作爲缺陷來檢測出 的場合，使用增量之積算値來構成即可。 〇 (4)把注目畫素由線狀光源像之邊緣之端至端1個1 個畫素移動同時針對邊緣上之所有畫素計算前述積算値。 藉此，產生曲率近似値之輪廓圖（曲率輪廓圖）。 其次，針對前述曲率輪廓圖之線狀光源像之邊緣的各 畫素，判定曲率是否在閩値T10以上，把所求得的曲率在 閾値T 1 0以上之畫素判定爲缺陷部位（或者缺陷候補）。 藉此，可以判定有無缺陷，此外，也可以求出缺陷的位置 。成形薄片2多少有些翹曲所以線狀光源像的邊緣也多少 有些彎曲，所以線狀光源像之邊緣的曲率若在某個程度以 -41 - 201033602 內的話’應該容許視爲不是缺陷。亦即，閾値T10應該取 比較大之値。閾値T10只要因應於容許的缺陷的尺寸而適 當決定即可，多値之次元影像資料爲25 6灰階之2次元影 像資料的場合，例如以11 0較佳。 此缺陷檢測演算法G，可具有高確實性地檢測出種種 線缺陷。以下，把此缺陷檢測演算法G稱爲「邊緣曲線法 2」° 於本實施型態，在線缺陷用影像解析部6 1 ,〜6 1 n及點 0 缺陷用影像解析部62,〜62n分別使用之線缺陷檢測演算法 及點缺陷檢測演算法之組合，爲以下之某一個。 (A) 在線缺陷用影像解析部όΐ-όΐη使用的線缺陷 檢測演算法係邊緣輪廓圖法2或邊緣曲線法2，在點缺陷 用影像解析部621〜62n使用的點缺陷檢測演算法係高通過 濾器法或峰値法2。 (B) 在線缺陷用影像解析部使用的線缺陷 檢測演算法係邊緣輪廓圖法2或邊緣曲線法2，在點缺陷 ❹ 用影像解析部62,〜62η使用的點缺陷檢測演算法係高通過 濾器法或峰値法2以外之缺陷檢測演算法。 (C) 在線缺陷用影像解析部使用的線缺陷 檢測演算法係邊緣輪廓圖法2或邊緣曲線法2以外之缺陷 檢測演算法，在點缺陷用影像解析部621〜62η使用的點缺 陷檢測演算法係高通過濾器法或峰値法2。 (A )〜（C )之組合之中，以（A )之組合最佳。( A )之組合的場合，可以確實檢測出線缺陷及點缺陷雙方 • 42- 201033602 。（B )之組合的場合，可以確實檢測出線缺陷。（C )之 組合的場合，可以確實檢測出點缺陷。 〔實施型態2〕 本發明之其他實施型態根據圖14說明如下。又，爲 了說明上的方便，對在前述實施型態1所示之各構件具有 相同機能之構件，賦予相同符號，省略其說明。 φ 相關於本實施型態之缺陷檢查裝置，取代圖1所示之 解析裝置6而具備圖14所示之解析裝置6A以外，具備相 關於實施型態1的缺陷檢査裝置1之相同的構成。解析裝 置6A，如圖14所示，係由圖1所示之解析裝置6省略點 缺陷用影像解析部 於本實施型態，在線缺陷用影像解析部6 1 ,〜6 1 n使用 的線缺陷檢測演算法，爲邊緣輪廓圖法2或邊緣曲線法2 。在本實施型態，可以確實檢測出線缺陷。 0 又，本實施型態之缺陷檢查裝置，可以單獨使用，但 以與可檢測出點缺陷的缺陷檢查裝置組合使用較佳。藉此 ，不僅線缺陷也可以檢測出點缺陷。與本實施型態之缺陷 檢查裝置組合之可檢測出點缺陷的缺陷檢查裝置，亦可爲 週知之種種缺陷檢查裝置，但最好是後述之實施型態3之 缺陷檢査裝置。亦即’可以確實檢測出線缺陷及點缺陷雙 方。 〔實施型態3〕 -43- 201033602 本發明之其他實施型態根據圖15說明如下。又，爲 了說明上的方便，對在前述實施型態1所示之各構件具有 相同機能之構件，賦予相同符號，省略其說明。 相關於本實施型態之缺陷檢查裝置，取代圖1所示之 解析裝置6而具備圖15所不之解析裝置6B以外，具備相 關於實施型態1的缺陷檢查裝置1之相同的構成。解析裝 置6B，如圖15所示，係由圖1所示之解析裝置6省略線 缺陷用影像解析部 於本實施型態，在點缺陷用影像解析部62 ,-62。使用 的點缺陷檢測演算法，爲高通過濾器法或峰値法2。在本 實施型態，可以確實檢測出點缺陷。 又，本實施型態之缺陷檢查裝置，可以單獨使用，但 以與可檢測出線缺陷的缺陷檢查裝置組合使用較佳。藉此 ，不僅點缺陷也可以檢測出線缺陷。與本實施型態之缺陷 檢査裝置組合之可檢測出線缺陷的缺陷檢查裝置，亦可爲 週知之種種缺陷檢查裝置，但最好是實施型態2之缺陷檢 査裝置。亦即，可以確實檢測出線缺陷及點缺陷雙方。 〔實驗例〕 接著，爲了確認本發明之效果，使用類似於相關前述 實施型態之缺陷檢查裝置之14種實驗用缺陷檢查裝置顯 示進行實驗的結果。 第1〜第7實驗用缺陷檢查裝置，係由相關於前述實 施型態3的缺陷檢査裝置省略攝影部52〜5η，作爲搬送裝 -44 - 201033602 置3’替代搬送輥，而使用將成形薄片2載置於其表面進 行搬送之輸送帶。第1〜7實驗用缺陷檢查裝置，係由包 含點缺陷之樣品檢測出點缺陷之用者。 第1實驗用缺陷檢査裝置具備使用邊緣輪廓圖法1之 點缺陷用影像解析部62 1〜62n，第2實驗用缺陷檢査裝置 具備使用邊緣輪廓圖法2之點缺陷用影像解析部62,〜62n ，第3實驗用缺陷檢查裝置具備使用高通過濾器法之點缺 φ 陷用影像解析部62 ^62«»，第4實驗用缺陷檢查裝置具備 使用峰値法之點缺陷用影像解析部62 !〜6%，第5實驗用 缺陷檢查裝置具備使用峰値法2之點缺陷用影像解析部 ’第6實驗用缺陷檢查裝置具備使用邊緣曲線法 1之點缺陷用影像解析部62〗〜62n，第7實驗用缺陷檢查 裝置具備使用邊緣曲線法2 (使用增量之積算値之方法） 之點缺陷用影像解析部 第8〜第14實驗用缺陷檢查裝置，係由相關於前述實 φ 施型態2的缺陷檢查裝置省略攝影部52〜5n，作爲搬送裝 置3,替代搬送輥，而使用將成形薄片2載置於其表面進 行搬送之輸送帶。第8〜14實驗用缺陷檢查裝置，係由包 含線缺陷之樣品檢測出線缺陷之用者。 第8實驗用缺陷檢查裝置具備使用邊緣輪廓圖法1之 線缺陷用影像解析部6 1 !〜6 1 n，第9實驗用缺陷檢查裝置 具備使用邊緣輪廓圖法2之線缺陷用影像解析部όΐ,-όΐη ，第10實驗用缺陷檢查裝置具備使用高通過濾器法之線 缺陷用影像解析部όΐ,κόΐη，第11實驗用缺陷檢查裝置 -45- 201033602 具備使用峰値法之線缺陷用影像解析部611〜61„，第12 實驗用缺陷檢查裝置具備使用峰値法2之線缺陷用影像解 析部，第13實驗用缺陷檢查裝置具備使用邊緣 曲線法1之線缺陷用影像解析部όΐ,-όΐη，第14實驗用 缺陷檢查裝置具備使用邊緣曲線法2 (使用增量之積算値 之方法）之線缺陷用影像解析部6 1 !〜6 1 η。 在本實驗例’作爲成形薄片2，使用包含不同種類之 點缺陷10種類之偏光片之樣品，與包含不同種類線缺陷6 種類之偏光片之樣品。包含線缺陷之1 〇種樣品，係包含 氣泡之樣品01、包含魚眼的樣品02、包含第1異物之樣 品03、包含與第1異物不同的第2異物之樣品〇4、包含 第1胎痕之樣品06、包含與第1胎痕不同的第2胎痕之樣 品07、包含第1打痕之樣品08、包含與第打痕不同的第2 打痕之樣品09、包含第1傷痕之樣品1 1、及包含與第1 傷痕不同的第2傷痕之樣品1 2。包含線缺陷的6種樣品是 包含折曲（線缺陷）之樣品10、包含與第1折曲不同的第 2折曲（線缺陷）的樣品1 3、包含沿著成形薄片2的搬送 方向之折線的樣品51、包含與成形薄片2的搬送方向直交 之強的折線的樣品52、包含與成形薄片2的搬送方向直交 之弱的折線的樣品53、及包含對成形薄片2的搬送方向爲 斜向之折線的樣品5 4。） 此外，在第1〜14之實驗用缺陷檢查裝置，作爲攝影 部5,，使用攝影二次元影像之，可以產生256灰階之橫 512畫素X縱480畫素之二次元影像資料的，使用CCD元 201033602 件之主動掃描區域感測器。在第1〜14之實驗用缺陷檢査 裝置，作爲線狀光源4，使用貼附銳利邊緣罩蓋（sharp edge hood)(使線狀光源像的邊緣銳利化之罩蓋）之高頻 螢光燈。在第1〜14之實驗用缺陷檢査裝置，使根據輸送 帶之成形薄片2的搬送速度爲20mm/sec(=1.2m/m i η)。此外，係使由輸送帶之端（圖2之前方之端）起算 位於145mm的距離之位置成爲攝影部攝影區域的中 0 心。此外，計算此145mm使用金屬尺，使由金屬尺到缺 陷爲止的距離爲55mm。 在供檢測出點缺陷之用的第1〜7實驗用缺陷檢査裝 置，以使成形薄片2上之攝影區域（視野）成爲橫（直交 於成形薄片2的搬送方向，且直交於成形薄片2的厚度方 向之方向）51.2mmx縱（成形薄片2的搬送方向）48mm 的方式，調整攝影部5,的位置及角度。但是，以前述主 動掃描區域感測器之512畫素X4 80畫素之中，由上起算第 φ 240個畫素的位置（上下方向之中央位置）之橫方向上排 列的512畫素攝影成形薄片2表面之橫向51.2mm之區域 的方式調整攝影部之位置及角度。亦即，以從成形薄片2 至攝影部5!爲止之距離成爲190mm的方式調整攝影部5! 之位置，此外以攝影部5!之攝影方向（由攝影部5,之聚 光透鏡的中心起，朝向藉由攝影部5 ,攝影的區域的中心 之方向）與成形薄片2表面之夾角角度成爲40度的方式 調整攝影部5,之角度。在此場合，攝影部之解析度爲 100 A m/畫素。此外，在供檢測出點缺陷之用的第1〜7 -47- 201033602 實驗用缺陷檢查裝置，作爲主動掃描區域感測器’焦點距 離爲25mm，最小F値爲1.4，透鏡先端工作距離爲 270mm之C固定件（C mount)之透鏡安裝於主動掃描區 域感測器本體者，光圈調整爲約11。 此外，在供檢測出點缺陷之用的第1〜7實驗用缺陷 檢査裝置，線狀光源4之長邊方向與成形薄片2之搬送方 向直交，由成形薄片2至線狀光源4爲止的距離爲2 4 0mm ，且連結成形薄片2上的攝影區域的中心與線狀光源4的 中心之直線對成形薄片2表面成爲夾37度角的方式’配 置線狀光源4。 在供檢測出線缺陷的第8〜14實驗用缺陷檢查裝置， 以成形薄片2上之攝影區域成爲橫204.8mmx縱192mm的 方式調整攝影部的位置與角度。但是，以前述主動掃 描區域感測器之512畫素χ4 80畫素之中，由上起算第24〇 個畫素的位置（上下方向之中央位置）之橫方向上排列的 512畫素攝影成形薄片2表面之橫向204.8mm之區域的方 式調整攝影部5!之位置及角度。亦即，以從成形薄片2 至攝影部51爲止之距離成爲400mm的方式調整攝影部 之位置，此外以攝影部5!之攝影方向與成形薄片2表面 之夾角角度成爲15度的方式調整攝影部5,之角度。在此 場合，攝影部5,之解析度爲2 00 ;zm/畫素。此外，在供 檢測出線缺陷之用的第8〜1 4實驗用缺陷檢查裝置，作爲 主動掃描區域感測器，焦點距離爲25mm，最小F値爲I·4 ，透鏡先端工作距離爲4 90mm之C固定件（C mount )之 201033602 透鏡安裝於主動掃描區域感測器本體者，光圈調整爲約 5 · 6 〜8 〇 此外，在供檢測出線缺陷的第8〜14實驗用缺陷檢查 裝置，線狀光源4之長邊方向對成形薄片2之搬送方向成 爲25度角度的方式配置線狀光源4，使線狀光源4之工作 距離爲900mm。 此處，以確實可以檢測出直徑〇.5mm的點缺陷的方式 φ 設定缺陷檢測演算法之參數。在第1及第8實驗用缺陷檢 查裝置，把邊緣輪廓圖法1之閾値T3設定爲3。在第2 及第9實驗用缺陷檢查裝置，把邊緣輪廓圖法2之閩値 T5設定爲4。在第3及第10實驗用缺陷檢查裝置，把邊 緣輪廓圖法2之k設定爲4.5,使用邊緣輪廓圖法2之橫 方向平滑化過濾器作爲1行3列之平滑化過濾器。在第4 及第11實驗用缺陷檢查裝置，把峰値法之閾値T7設定爲 最大亮度値之25% (255χ〇.25)。在第5及第12實驗用 φ 缺陷檢査裝置’把峰値法2之閬値Τ8設定爲20。在第6 及第13實驗用缺陷檢查裝置，把邊緣曲線法1使用的參 數之距離設定爲15'把k設定爲5。在第7及第14實驗 用缺陷檢査裝置，作爲邊緣曲線法2使用求出前述之近似 的曲率的方法’把計算對象範圍定爲對注目畫素前後30 畫素之範圍（亦即使N爲IS)，使閾値T10爲110。 接著’使用第1〜7實驗用缺陷檢查裝置，調查是否 可由包含1 〇種點缺陷的樣品檢測出點缺陷，使用第8〜1 4 之實驗用缺陷檢查裝置，調査是否可由包含6種線缺陷的 -49- 201033602 樣品檢測出線缺陷，所得結果顯示於表1。[Technical Field] The present invention relates to a defect of forming a formed sheet such as an optical film such as a polarizing film or a retardation film (especially a long-length optical film that is wound and transported in a roll shape) Defect inspection device. [Prior Art] φ The defect inspection device of the former formed sheet uses a linear sensor called a line sensor to illuminate the formed sheet with a linear light source such as a fluorescent tube, along the length of the formed sheet. One side of the longitudinal direction is scanned from one end of the longitudinal direction to the other end to obtain a single still image by scanning the surface of the formed sheet with a one-dimensional camera, and the defects of the formed sheet are inspected based on the one still image data. The still image data here usually includes a linear light source image. The linear light source image is a linear light source and an image in which a sheet is placed between the camera and the reflecting surface, and the linear light source is emitted by the linear light source and is reflected by the forming sheet to the camera, the linear light source and the camera. When the formed sheet is disposed between them, it is an image of light emitted from the linear light source and transmitted through the formed sheet to the camera. In the defect inspection apparatus, when the width of the formed sheet is wide, a plurality of linear sensors are used in parallel in the width direction so that the entire width direction of the formed sheet can be inspected. However, in the conventional defect inspection apparatus, the defect of the formed sheet is inspected based on one still image data (hereinafter referred to as "image data J" in the entire area of the formed sheet, so that the image of the inspection object and the linear light source image of the image data are detected. The positional relationship becomes a determined positional relationship. Defect -5 - 201033602 will only appear on the image data when the positional relationship between the object pixel (the pixel of interest) and the linear light source image is in a specific positional relationship. For example, a type of bubble of a defect often appears on the image data only at or near the periphery of the line-like source image. Therefore, the defect is often not detected due to its position. That is, the aforementioned defect inspection is performed. The device has only a limited defect detection capability. Therefore, the applicant of the present application has applied for a defect inspection device for a formed sheet which can improve the defect recognition ability of the aforementioned defect inspection device (refer to Patent Document 1). A linear light source such as a fluorescent tube illuminates the formed sheet, and the formed sheet is continuously conveyed in a specific direction while The animation data (a plurality of image data having different photographing positions on the formed sheet) is obtained by using a two-dimensional camera called an area sensor, and the defective person of the formed sheet is inspected based on the animation data. The device can determine whether or not there is a defect based on a plurality of pieces of image data having different positional relationships between the inspection target pixel and the linear light source image, so that the defect can be detected more reliably than the previous defect inspection device. That is, the defect inspection The device is more capable of detecting the defect than the previous defect inspection device. Further, with the animation data, the appearance of the defect on the illumination image can be seen. [Prior Art Document] [Patent Document] [Patent Document 1] Japan [Patent Publication No. 2007-218 62 9 (published on August 30, 2007).] 201033602 [Disclosure] [Problems to be Solved by the Invention] However, according to the review by the inventor of the present invention, it is known that The defect inspection device of Patent Document 1 also has room for improving the defect detection capability. In the defect inspection apparatus described in Patent Document 1, the defect is detected by the following image processing from each of a plurality of image data (multiple images) photographed by the area sensor (refer to Patent Document 1). [φ 0032] ~ [0035]). First, the image data of multiple images is degenerated, and the white area and the black area are marked as detection objects. Next, a white area having an area (number of pixels) exceeding a specific 値 (a relatively large area (for example, 2,500 pixels) that fits the area of the linear light source image) is regarded as a line by the white area of the detection target. The light source is excluded from the image. Similarly, a black area having an area exceeding a specific 値 (a relatively large 配合 area of the background area) is regarded as a background area (a shadow φ image of a region having no defects of the formed sheet) ) and excluded. Further, the white area and the black area of the area which is less than a specific 値 (a relatively small 接近 of one pixel; for example, 9 pixels) are excluded from the white area and the 黒 area of the detection target, and are excluded as noise. . Next, the remaining area that has not been excluded from the white area and the black area of the detection target is detected as a defect. However, the defect inspection device described in Patent Document 1 detects a defect based on the inversion of light and darkness, so that a defect with low contrast on the image data cannot be detected. Here, various defects of the present invention as detection targets will be described. According to Japanese Laid-Open Patent Publication No. 201033602, it is mainly to detect defects (appearance defects) accompanying minute irregularities (especially irregularities having a high degree of m) from the surface of the formed sheet. Examples of the defects accompanying minute irregularities include minute irregularities generated on the surface of the formed sheet due to bubbles or foreign matter; and marks (indentations due to point pressing): marks that are bent (referred to as "folding"曲 ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( These defects, which are accompanied by minute irregularities, are very difficult to detect with the defect inspection device of the former linear sensor. The main object of the present invention is to detect defects of these types. In the present specification, for the sake of convenience, the small unevenness is locally concentrated (the diameter of the convex portion is not more than 1 mm (the decomposition energy of the imaging device is several pixels or less in the case of 200 / pixel)), for example, bubbles, foreign matter, The scratches and the like are referred to as point defects, and those in which the minute irregularities are connected in a line shape exceeding 1 mm are called line defects. A typical line defect such as a broken line has a width of more than 10 mm (the resolution of the photographic device is more than tens of pixels in the case of 200 // m/picture), typically tens of cm The degree, sometimes with a width of more than tens of cm. Flexibility, which has a width of less than 10 mm (the resolution of the photographic device is less than tens of pixels in the case of 200 # m/pixel), and the width is typically a few mm, with point defects and typical The intermediate nature between line defects. An example of a defective image (movie) photographed by the photographing unit of the defect inspection device described in Patent Document 1 is shown in Figs. 14 to 15 of Patent Document 1. In Fig. 14 to Fig. 15 of Patent Document 1, five consecutive frames of -8-201033602 frames are arranged in order of time (a) to (e). If the animation is a normal TV animation, the time interval between frames (frame rate) is 1/30 second. The time interval between frames depends on the characteristics of the photography department. In the defect inspection apparatus of the patent document 1, the imaging unit is oriented in the direction of the center of the image forming unit toward the image forming area of the molded sheet (the rectangular shape indicated by the broken line on the surface of the formed sheet of Fig. 1 of Patent Document 1) and the conveying direction of the formed sheet. The acute angle is included, and an illumination image (an inverse φ image of the linear light source) is included in one of the imaging regions, and is disposed such that there is a region (background region) having no illumination image on both sides of the illumination image of the imaging region. That is, the direction of the upper side of the animation shown in Figs. 14 to 15 of Patent Document 1 coincides with the direction in which the formed sheets are conveyed. In the photographing area of the formed sheet, the defect moves in the transport direction of the formed sheet. Therefore, in the animation shown in FIGS. 14 to 15 of Patent Document 1, the defect moves from the bottom to the top (the illumination image is a white strip). FIG. 14 of Patent Document 1 is an example of an animation including a photographing area (a size of a direction orthogonal to the transport direction: 5 mm) of a bubble (dot defect) photographed by the photographing unit, and the bubble is visible to the light and dark. The reverse part. In this animation 'the bubble cannot be seen in the first frame (a), the dot defect starts to see the bubble in the second frame (b) close to the illumination image, and the bubble is located at the edge of the illumination image. The bubbles in the frame (c) can be clearly seen, and the fourth and fifth frames (d)(e) of the bubbles entering the illumination image are invisible because the bubbles are buried in the illumination light. Fig. 15 of Patent Document 1 is an example of an animation-9" 201033602 which includes a photographed region (a size orthogonal to the transport direction: 5 mm) which is photographed by the photographing unit, and is not folded by the photographing unit. The light and dark reversal caused by the song' As the flexion passes, the illumination image that should have been taken out of the rectangular white area is distorted over time. Fig. 15 of Patent Document 1 is an example of an animation including an imaging region (a dimension in a direction orthogonal to the transport direction: 200 mm) of a broken line (line defect) photographed by the photographing unit. In the first to third frames (a) to (c), the illumination image is somewhat twisted into a bow shape, but this degree of bow-like distortion is not caused by the defect but by the pulling of the formed sheet. On the other hand, the illumination image in the @4th frame is S-shaped, and the distortion is relatively large. The illumination image is distorted beyond this level because of the polyline at the twisted position. In other words, it is expected that the distortion portion of the illumination image that is distorted beyond this is detected as a defect. However, the defect inspection device of Patent Document 1 knows that the polarization region is distinguished by the bright region (linear light source image and linear light source). Like the internal defect area) and the dark area (the background area and the defect area inside the background area), the contrast between the image data is very low and the lack of the trap can not be detected. In other words, in the defect inspection apparatus of Patent Document 1, when the contrast on the defect image (the brightness change due to the defect) is relatively high, as shown in the vertical direction (formed sheet conveyance direction) of the brightness profile of Fig. 3, the defect The resulting change in brightness is observed in a manner that spans the threshold of the 値 process. More specifically, the luminance 値 of the minimum point (the valley portion of Fig. 3) corresponding to the defect observed as the dark region in the linear light source image becomes larger, and the luminance of the maximum point on both sides of the minimum point 値Become smaller. In addition, 2 -10- 201033602 The threshold of the deuteration process, the brightness of the maximum point (the mountain part of Fig. 3) corresponding to the defect observed as a bright area other than the linear light source image becomes larger, and its maximum point The brightness of the very small dots on both sides becomes larger. Further, a defect observed as a dark region in the linear light source image or a defect observed as a bright region outside the linear light source image is also detected. That is, the defect inspection device of Patent Document 1 can surely detect a defect having a relatively high contrast. φ On the other hand, in the defect inspection device of Patent Document 1, when the contrast of the defect in the image data is low, the relationship between the vertical brightness profile of the defect portion and the threshold of the 値 process becomes the vertical brightness of FIG. The relationship shown in the contour map, that is, the change in luminance caused by the defect becomes a relationship that does not cross the threshold of the 値 processing. As a general guess, when the subtype is satisfied, (the amount of change in brightness caused by the defect) <{(luminance level of the linear light source image area) - (luminance level of the background area)} / 2 (1) becomes the relationship shown in Fig. 4 . When you become the relationship shown in Figure 4, you will see the _ defect. However, the contrast of the defects on the image data is very low. Even if the formula (1) is satisfied, the brightness change of the defect shown in the vertical brightness profile of Fig. 5 is observed if it crosses the threshold of the 値 process. In this case, defects can be detected. In other words, in the defect inspection apparatus of Patent Document 1, even when the formula (1) is satisfied, the defect can be detected by the relationship between the luminance change due to the defect and the threshold of the enthalpy treatment. Further, in the defect inspection device of Patent Document 1, a moving image is used, and in the moving image, the defect image is observed in one frame and is close to the linear light source -11 - 201033602 image, and passes through the linear light source image. The light source is like moving away from it. Also, even if the contrast of the defects in the image data is low, in the case where the formula (1) is satisfied, even if one of the plurality of images constituting the moving image is lacking, the luminance change and the subtraction processing are lacking. When the relationship of the threshold 成为 is the image of the _ system shown in Fig. 5, the defect can be detected. As described above, in the defect inspection device of Patent Document 1, the amount of change in luminance with a defect is reduced by the equation (1), and the possibility of seeing a defect is high. In other words, the defect inspection device of Patent Document 1 has room for improvement in the reliability of detection of defects having a low _ ratio. The present invention has been made in view of the above problems, and an object thereof is to provide a defect inspection apparatus which can provide various types of defects more reliably and which can provide a formed sheet. [Means for Solving the Problem] In order to solve the above-described problems, the defect inspection device according to the present invention is a defect inspection device that detects a defect of a formed sheet, and is characterized in that it includes @: multiple generation of the second-order image generation of the formed sheet. a photographing means for multiplexing a plurality of dimensional image data for illuminating a linear light source for forming the formed sheet to project an image of the linear light source to a portion of the photographed region of the formed sheet to change the formed sheet At least one of the formed sheet and the linear light source is moved so as to intersect at least one of the linear light source in the longitudinal direction of the linear light source, and is orthogonal to the thickness direction of the formed sheet, in such a manner that the image of the linear light source is projected. The moving means on the upper side, and the line defect detecting means for detecting the defect of the line-12-201033602 from the plurality of binary image data generated by the above-mentioned photographing means; the line defect detecting means is by fitting with a function curve The edge of the image of the linear light source of the aforementioned 2-dimensional image data, the edge of the image of the linear light source and the function curve A line defect detection algorithm as a line defect is detected at a distance above the first threshold, or an edge of an image of a linear light source of the second-order image data is obtained, and a curvature of a vicinity of each pixel is obtained. A line defect detection algorithm as a line defect is detected at a position above the second threshold 来 to detect a φ line defect. According to the above configuration, the line defect can be more reliably detected. Preferably, the defect inspection device further includes a dot defect detecting means for detecting a point defect from a plurality of binary image data generated by the image capturing means. Thereby, not only the line defects but also the point defects can be detected. Preferably, in the defect inspection device, the point defect detecting means displays the luminance change of the second-order image data depending on the position along the straight line as a luminance contour map, and the data point group of the luminance contour map is false. a particle that moves in such a manner that the movement time between data points becomes constant, and the velocity vector between the velocity points of the two data points before the attention data point and the data points before the data points of the preceding attention data point The acceleration vector is used to predict the brightness 値 of the above-mentioned attention data point, and the point defect detection algorithm as a point defect is detected by the difference between the predicted brightness 値 and the actual brightness 在 above the third threshold ,, or the above 2 The dimensional image data is smoothed, and the difference between the smoothed 2nd dimensional image data and the original 2nd dimensional image material is obtained as differential image data, and the brightness of the differential image data is greater than the fourth threshold, and the brightness is present. The fifth threshold 値 ( -13- 201033602 The fifth threshold 値 is smaller than the fourth threshold )) is detected as the point defect detected as a point defect Algorithm to detect point defects. Thereby, compared with the technique of Patent Document 1, it is possible to more reliably detect a point defect having a low contrast (for example, a defect showing a small change in luminance as shown in Fig. 4). That is, both the point defect and the line defect can be reliably detected. In order to solve the above-described problems, the defect inspection device according to the present invention is a defect inspection device that detects a defect of a formed sheet, and is characterized in that the second-order image of the formed sheet is photographed in plural times to generate a plurality of second-order images of the image data. And means for illuminating the linear light source for forming the formed sheet so as to change an image of the linear light source to a portion of the imaged region of the formed sheet to change an image of the linear light source of the formed sheet At least one of the formed sheet and the linear light source is moved in a direction intersecting the longitudinal direction of the linear light source and orthogonal to the direction of the thickness direction of the formed sheet, and a point defect detecting means for detecting a point defect by using the plurality of second-order image data generated by the photographing means; wherein the point defect detecting means indicates that the second-order image data is dependent on a brightness change along a position along the straight line For the luminance profile, the data point group of the luminance profile is assumed to be between the data points. The moving time becomes a particle of a certain way of moving, and the above-mentioned attention data is predicted from the velocity vector of the mass point between the two data points before the data point of interest and the acceleration vector of the mass point between the data points of the three preceding data points. The brightness of the point 检测, by detecting the difference between the predicted brightness 实际 and the actual brightness 在 above the third threshold 检测, detecting the point defect detection algorithm as a point defect, or smoothing the aforementioned 2nd image data-14 - 201033602 The difference between the smoothed 2nd-dimensional image data and the original 2nd-dimensional image material is obtained as the difference image data, and the brightness of the difference image data is above the 4th threshold, and the brightness is at the 5th threshold ( The fifth threshold 値 is smaller than the fourth threshold ). The point defect detection algorithm as a point defect is detected at the following point to detect the point defect. As a result, compared with the technique of Patent Document 1, it is possible to more reliably detect a point defect having a low contrast (for example, a small luminance change is shown in Fig. 4). Preferably, the defect inspection device further includes a line defect detecting means for detecting a line defect from a plurality of binary image data generated by the image capturing means. Thereby, not only the point defects but also the line defects can be detected. [Effects of the Invention] As described above, the present invention has an effect of providing a defect inspection device capable of more reliably detecting various defects and providing a formed sheet. [Embodiment] [Embodiment 1] An embodiment of the present invention will be described below with reference to the drawings. The defect inspection apparatus according to this embodiment detects a defect of a formed sheet. The defect inspection apparatus according to this embodiment is suitable for inspection of a light-transmissive formed sheet, particularly a molded sheet made of a resin such as a thermoplastic resin. The molded sheet made of a resin is, for example, a thermoplastic resin extruded from an extruder, which is provided with a smooth -15-201033602 or gloss on the surface by a gap between the rolls, and is cooled on the conveying roller while being pulled by a pulling roll. Pull and shape. Suitable for the thermoplastic resin of this embodiment, such as methacrylic resin, methyl methacrylate-styrene copolymer, polyolefin such as polyethylene or polypropylene, polycarbonate 'polyvinyl chloride, polystyrene, Polyvinyl alcohol, cellulose triacetate resin, and the like. The formed sheet may be composed of only one of these thermoplastic resins, or may be a laminate of plural types of these thermoplastic resins (laminated sheets). Further, the defect inspection apparatus according to the present embodiment is suitable for inspecting an optical film such as a polarizing film or a retardation film, in particular, a long-length optical film which is wound and stored in a roll shape. Further, the formed sheet may be of any thickness, generally referred to as the relatively thin thickness of the film, or generally referred to as a relatively thick sheet. Examples of the defects of the formed sheet include point defects such as bubbles (produced during molding), fish eyes, foreign matter, wheel marks, marks, and scratches; bending and folding lines (due to differences in thickness) The generator, etc., etc., will be described below with reference to FIGS. 1 and 2 for the configuration of the defect inspection apparatus 1 according to the present embodiment. Fig. 1 is a functional block diagram showing the main part of the defect inspection device 1. Fig. 2 is a schematic view showing the appearance of the defect inspection device 1. Further, in Fig. 2, the light and dark of the surface of the formed sheet are reversed in such a manner that the member superposed on the formed sheet can be easily recognized. That is, the black area of the surface of the formed sheet of Fig. 2 is actually a bright (bright) area, and the white area of the surface of the formed sheet of Fig. 2 is actually a dark area. In the defect inspection device 1, the rectangular formed sheet 2 is conveyed in a certain direction by the transport device (moving means) 3, and the formed sheet 2 illuminated by the linear light source 4-16-201033602 is n by n (n is 2) The imaging unit (photographing means) 51 to 5 of the above-mentioned integer) generates a plurality of sub-dimensional image data for each of the imaging units 51 to 5n, and causes the analysis device 6 to detect the formed sheet based on the generated binary image data. The defect detecting device 1 includes a conveying device (moving means) 3 for conveying the formed sheet 2, an image capturing region for the formed sheet 2 (a region photographed by the photographing portions 51 to 5n, and a forming sheet 2 of Fig. 2). The linear light source 4 for forming the sheet 2 is illuminated by one of the φ rectangles whose surface is indicated by a broken line, and the reflected image of the linear light source 4 is photographed as a line from the linear light source 4 The direct light is reflected by the formed sheet 2 and reaches the image of the linear light source 4 formed as a result of the imaging portions 5! to 5n) and the reflected image of the formed sheet 2 (as scattering from the linear light source 4). The image forming units 51 to 5n of the image of the formed sheet 2 formed by the reflection of the formed sheet 2 and reaching the photographing portions 51 to 511, and the image processing algorithm (defect detection) based on the plurality of binary image data Algorithm) An analysis device 6 that detects defects of the formed ruthenium sheet 2. The conveying device 3 conveys the formed sheet 2 in a direction orthogonal to the thickness direction thereof, particularly in the longitudinal direction thereof, so as to change the position at which the image of the linear light source 4 of the formed sheet 2 is projected. The conveying device 3 includes, for example, a feeding roller and a receiving roller that convey the formed sheet 2 in a predetermined direction, and measures the conveying speed by a rotary encoder or the like. The transport speed is set to, for example, 2 m to 12 m/min. The transport speed of the transport device 3 is set and controlled by an information processing device (not shown) or the like. The linear light source 4 has a direction in which the longitudinal direction thereof intersects with the feeding direction of the forming sheet 2 - 201033602 (for example, a direction orthogonal to the conveying direction of the formed sheet 2), and the reflection of the linear light source 4 The image forming area of the cross-cut sheet 2 is disposed such that there is a region (background area) having no linear light source image on both sides of the reflection image of the linear light source 4 in the image capturing area. The linear light source 4 is not particularly limited as long as it emits light that does not affect the composition and properties of the formed sheet 2, and is, for example, a fluorescent lamp (especially a high-frequency fluorescent lamp), a metal halide lamp, or a halogen. Transfer lights, etc. Further, the linear light source 4 is placed at a position facing the imaging units 5! to 5n φ with the formed sheet 2 interposed therebetween, and the transmission image including the linear light source 4 is captured by the imaging unit (direct light from the linear light source 4) The image of the linear light source 4 formed by the formed sheet 2 reaching the imaging portions 51 to 511 and the transmission image of the molded sheet 2 (the formation of the scattered light from the linear light source 4 through the molded sheet 2 to the imaging portions 51 to 511) The 2nd image of the image of the sheet 2 can also be used. Each of the photographing units includes a two-dimensional image of the reflected image of the linear light source 4 and the reflected image of the formed sheet 2, and generates and outputs a plurality of binary image data. The photographing unit 5^5«« is photographed 2 A region sensor consisting of a CCD (Charge Coupled Device) or a CMOS (Complementary Metal-Oxide Semiconductor) photographic element. Since the size of the defect detected by the defect inspection device 1 depends on the decomposition energy of the imaging units 51 to 5, the decomposition energy of the imaging portions 51 to 5 can be selected in accordance with the size of the defect to be detected. Further, since the three-dimensional shape (ratio of the width to the height) of the defect detected by the defect inspection device 1 does not substantially depend on the decomposition energy of the imaging unit Si-Sn, there is no difference in the type of the defect to be detected. Select camera -18- 201033602 to resolve the need. The photographing units 51 to 5n are disposed such that the direction from the photographing portion SixSn toward the center of the photographing region of the formed sheet 2 and the transport direction of the formed sheet 2 are acute. The photographing units 5 i to 5 n are at least one of the photographing portions Si-Sn in the entire width direction of the molded sheet 2 (straight in the transport direction of the formed sheet 2 and in the direction perpendicular to the thickness direction of the formed sheet 2). The method of photographing 'φ along the width direction of the formed sheet 2 is juxtaposed. By photographing the entire area in the width direction of the formed sheet 2 by the photographing portions 5! to 5n, it is possible to inspect the defects of the entire region of the formed sheet 2. The photographing interval (frame rate) of the photographing units 51 to 5n may be fixed, may be changed as the user operates the photographing unit itself, or may be changed to be connected to the photographing units 5i to 5n by the user's operation. The information processing device (not shown; may be omitted) may also be used. Further, the photographing interval of the photographing sections 5i to 5n may be a fraction of a second of the time interval of continuous photographing of the digital camera, etc., in order to improve the efficiency of the inspection, it is preferable to use a short time interval, for example, a general animation. The frame rate of the data is 1/3 of 0 seconds. Here, the distance (transport distance) in which the formed sheet 2 is transported during the period from the time when one of the two-dimensional images is captured to the next-to-two-dimensional image is set to be the photographing along the transport direction of the formed sheet 2 The length of the region is at least l/m (m is 2 or more). Thereby, the 2-dimensional image including the same place of the formed sheet 2 is photographed m times. m is preferably larger than 2. By increasing the number of times of photographing of the same place of the formed sheet 2, the defect can be inspected with high precision. As shown in FIG. 1, the analysis device 6 receives a plurality of binary image data output from each of the imaging units 5! to 5 „ -19 -19 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 。 The line-defective image analysis unit (line defect detecting means) 6 1 1 to ό 1 n and the point defect image analyzing unit (point defect detecting means) 02, 62n' and the display unit 64 for displaying the detection result (inspection result) The control CPU 63 that controls the respective units is collectively input to the line defect image analysis units 61 to 61n and the point defect image analysis units 62 to 62n, respectively. Each of the line defect image analysis unit defect analysis units 62 to 62n has a complex number (complex frame) of two dimensions which are different in position of the linear light source image on the formed sheet 2 by the line defect detection algorithm. The image data is detected as a defect, and the result is output as an inspection result. Each of the point defect image analysis units 62 62n is formed by the point defect detection algorithm 'the linear light source image on the formed sheet 2 Defects are detected in the two-dimensional image data of the plural (complex frame), and the result is output as the inspection result. The line defect image analysis unit and the point defect image analysis unit are based on the linear light source on the formed sheet 2. The second-order image data with different positions is determined to be defective, so the defect can be detected more reliably than the previous defect inspection device. The line defect detection algorithm and the point defect detection algorithm will be described later. The parameters of the algorithm and the point defect detection algorithm may be fixed, and may be connected to the line defect image analysis unit 611 to 61 „ and the point defect image analysis unit 62t to 62n2 by the user's operation. Information processing device (not shown; can be omitted). -20- 201033602 The line defect image analysis units 611 to 61n use a line defect detection algorithm to detect a line defect from L (LSm) or more of m plural pieces of binary image data. The result of the defect is output as the inspection result. In other cases, the result without the line defect may be output as the inspection result, or the line defect may be detected by more than L of the m plural binary image data by the line defect detection algorithm. In the case of the line defect position, the information of the line defect position is output as the inspection result, and the inspection result may not be output in other cases. When the number m of the two-dimensional image data is 3 or more, and L is 2 or more, when the number of the binary image data of the line defect is detected by the line defect detection algorithm is smaller than L, the line defect The test results are considered as false positives (originally undefected but mistakenly identified as defects) and excluded. In this way, false positives can be reduced. Further, when the information of the line defect position is output as the inspection result, it is necessary to use a line defect detection algorithm that can obtain the position of the line defect. The point defect image analysis units 62 to 62n detect a point defect by L (LSm) or more of m plural binary image data by the point defect detection φ algorithm, and have a somewhat defective defect. The result is output as the inspection result. In other cases, the result of not having the point defect may be output as the inspection result, or the point defect detection algorithm may be used to detect the point defect from more than L of the m plural binary image data. The information of the point defect position is output as the inspection result, and the inspection result may not be output in other occasions. When the number m of the two-dimensional image data is 3 or more, and L is 2 or more, when the number of the binary image data of the point defect is detected by the point defect detection algorithm is smaller than L, the point defect Test results -21 - 201033602 were excluded as a false positive (originally undetected but mistakenly identified as a defect). In this way, false positives can be reduced. Further, when the information of the point defect position is output as the inspection result, it is necessary to use a point defect detection algorithm that can obtain the point defect position. The control CPU 6 3, the inspection result output by the image analysis unit όΗ-όΙη and the point defect image analysis unit 62 of the integrated line defect is used to create inspection result information corresponding to the entire area of the formed sheet 2, and is stored in a memory device (not shown). At the same time, it is displayed on the display unit 64. As the inspection result information corresponding to the entire area of the formed sheet @ 2, information showing whether or not there is no defect in the entire area of the formed sheet 2, or a defect map of the entire area of the formed sheet 2, or the like can be cited. When the inspection result information corresponding to the entire area of the formed sheet 2 is produced, when at least one of the line defect image analysis unit and the point defect image analysis unit 62 i to 62n detects a defect, the defect is determined to be a result of the inspection. News. The control CPU 63 creates a field map of the defect map of the entire area of the formed sheet 2 as the inspection result information 'corresponding to the inspection result information of the entire area of the formed sheet 2'. The coordinate position on the 2nd-dimensional image data is converted into the coordinate position on the formed sheet 2 to generate defect position information, and the defect position information is output to the control CPU 63. For example, paragraphs [〇〇37] to [0041] and [described in Patent Document 1] can be used as the line defect image analysis units 61 1-6 to 6 and the "mark conversion processing for each of the point defect image analysis units 621 to 62n". Processing of 00 50]~[〇053]. Outputting the information of the defect map to a marking device (not shown) and information processing -22-201033602 device (not shown), the marking device is forming the sheet according to the defect map The method of marking the position of the defect can also be employed. The marking device, for example, has an arm disposed along the width direction of the formed sheet 2, and a marking head having a pen or the like, which can be formed on the arm by the marking head 2 The reciprocating movement in the width direction is performed at any position on the formed sheet 2. The information of the marked defect position can be used, for example, after cutting the formed sheet 2 into a plurality of sheets of a specific size. These sheets are classified into a normal product and a φ defective product, etc. Further, in the above embodiment, the linear light source 4 is fixed and the formed sheet 2 is conveyed, but the line of the formed sheet 2 is changed. The position of the image of the light source 4 may be projected. That is, the fixed sheet 2 may be fixed to move the linear light source 4, or the formed sheet 2 and the linear light source 4 may be moved in different directions or at different speeds. When the linear light source 4 is fixed and the moving light source 4 is fixed, it is preferable to move the imaging unit Si-Sn and the linear light source 4 at the same speed in the same direction. Thereby, a plurality of 2nd dimensional images including the linear light source image can be obtained. φ image data. The method of fixing the formed sheet 2 and moving the linear light source 4 is to stretch the formed sheet 2 by the conveying device 3 to avoid distortion of the linear light source image, but the length of the formed sheet 2 that can be inspected at one time is limited by the line. In order to efficiently inspect the long-sized formed sheet 2', it is preferable to convey the formed sheet 2 as in the above-described embodiment. Further, in the above embodiment, the line defect is used for the length corresponding to the movable range of the light source 4. The image analysis units 6h to 61' and the point defect image analysis units 62l to 62n detect defects based on the binary image data obtained from the same imaging unit 5: 5n, but -23-201033602 is also can Defects were detected based on the 2-dimensional image data obtained by different photography departments. Thereby, the photographing conditions (the distance from the formed sheet 2, the angle between the transport direction of the formed sheet 2 and the photographing direction, and the like) of the photographing portions 5 to 5n can be set as conditions suitable for the defect to be detected. The imaging unit of the 2nd-dimensional image data used by the imaging defect imaging analysis unit is formed so as to be farther from the molding sheet than the imaging unit for capturing the 2D image data used by the point defect image analysis unit. It is preferable that the angle between the conveying direction of the sheet 2 and the photographing direction is narrow. As a result, both the defect and the line defect can be detected in the optimum imaging conditions, so that both the point defect and the line defect can be accurately detected. Further, in the above-described embodiment, the two-dimensional image captured by each of the image capturing units 51 to 5n is distributed by the line defect image analyzing unit and the point defect image analyzing unit 621 to 62, but is photographed by the image capturing units 51 to 5n. The relative position of the 2nd-order image is composed of one full-width image including the entire area in the wide direction of the formed sheet 2 in the η-element image captured by each of the image capturing units 51 to 5n, and is 1 according to the full-width image. For the line defect, the image analysis unit and the one point defect image analysis unit may detect the defect. As a method of synthesizing the i full width image from the η two-dimensional image, for example, the paragraph described in Patent Document 1 can be used. [0050] Next, the line defect detection algorithm and the point defect detection algorithm used by the line defect image analysis units 6 1 to 6 1 n and the point defect image analysis units 62 and 62n are described as line defects. The detection algorithm and the point defect detection algorithm can use the following seven defect detection algorithms A to G. In addition, in the following description, the brightness of the 2 dimensional image data (pixels) is taken as a natural number -24 - 201033602 Come [Defect Detection Algorithm A] The defect detection algorithm A is described below with reference to Fig. 6. Fig. 6 (a) is a multi-dimensional 2 image data generated by the imaging unit Si-Sni (hereinafter referred to as the original image) In the case of the data, the upper side of the image is the downstream side of the transport direction, and the lower side of the image is the upstream side of the transport direction. In Fig. 6 (a), the strip-shaped white area extending in the lateral direction of the center is a linear light source image. The dark area inside the linear light source image and the small white area existing in the vicinity of the linear light source image are defects. In this defect detection algorithm A, for each of the plurality of original image data generated by the photographing unit, The following processing is performed: First, the original image data is divided into data of a pixel row along the vertical direction (the transport direction of the formed sheet 2) (indicating brightness 値 (pixel 値) and position data; Brightness contour map; 1 dimensional image data reference). Secondly, the data for each pixel sequence is explored from one end (upper end of Fig. 6(a)) toward the other end (Fig. 6(a) below). The first side of the edge First, the second pixel is counted as the pixel of interest on one end of the pixel column, and whether the brightness of the pixel of the eye is smaller than the brightness of the adjacent pixel adjacent to the one end side is smaller than the threshold 値T1. In the above, when it is determined that the luminance 値 of the pixel of interest is smaller than the threshold 値T1 of the luminance of the adjacent pixel (that is, the luminance 値 of the adjacent pixel is La, and the luminance 値 of the pixel of interest is Lb, La-Lb 2 T1 ), judged to be neighbor -25 - 201033602 The pixel is the first edge, the position of the first edge (the position of the adjacent pixel) is recorded, and the data processing of the pixel column of the processing target is ended. The pixel moves toward the other end 1 pixel 1 while repeating the above-described determination until the luminance 値 of the pixel of the neighboring pixel luminance is determined to be greater than the threshold 値T1, and is determined to be adjacent to the luminance of the pixel of interest. When the brightness of the pixel is smaller than the threshold 値T 1 or more, it is determined that the adjacent pixel is the first edge, the position of the first edge (the position of the adjacent pixel) is recorded, and the processing of the data of the pixel column to be processed is ended. Further, the threshold 値T1 is a natural number of any @, and may be a minimum unit of luminance 値. When the threshold 値T 1 is the minimum unit of the luminance 値, the above determination is to determine whether or not the luminance 値 of the pixel of interest is smaller than the luminance 値 of the adjacent pixel. Second, aim. The data of each pixel sequence is subjected to a second edge determination process from the other end toward the end of the search edge as described below. First, the second pixel is counted as the pixel of interest on the other end side, and it is determined whether the luminance 値 of the pixel of interest is larger than the luminance 値 of the adjacent pixel adjacent to the other end side by a threshold 値T2 or more. When it is determined that the luminance ❹ 注 of the pixel of interest is larger than the luminance 邻接T2 of the adjacent pixel (that is, the luminance 値 of the adjacent pixel is La, and the luminance 値 of the pixel of interest is Lb, Lb-La 2T2 "It is determined that the pixel of interest is the second edge, the position of the second edge is recorded (the position of the pixel of interest), and the data processing of the pixel column of the processing target is finished. In other cases, the pixel of interest is oriented toward one end. The first step is repeated until the brightness of the pixel of interest is determined to be greater than the brightness of the adjacent pixel by a threshold T2 or more, and it is determined that the brightness of the pixel of interest is larger than the brightness of the adjacent pixel. When the threshold 値T2 or more is -26-201033602, it is determined that the pixel of interest is the second edge, the position of the second edge (the position of the pixel of interest) is recorded, and the processing of the data of the pixel column to be processed is ended. Moreover, the threshold 値T2 is an arbitrary natural number, and may be a minimum unit of brightness 値. When the threshold 値T2 is the minimum unit of the luminance 値, the above determination determines whether the luminance 値 of the pixel of interest is larger than the luminance 値 of the adjacent pixel by a threshold 値 or more. An example of the first edge detected by the first edge determination processing is shown as "Δ" in φFig. 6 (a), and an example of the second edge detected by the second edge determination processing is shown in Fig. 6 (a). ) is displayed as "〇". As can be seen from Fig. 6(a), in the region where there is no defect, there is no edge other than the edge of the linear light source image, so the first edge and the second edge coincide with the edge of the other end side of the linear light source image (in the figure). The case of 6(a) is the edge of the lower side), which is consistent with each other. On the other hand, as can be seen from Fig. 6(a), in the defective region (white region and black region), at least one of the first edge and the second edge coincides with the edge of the defective region, compared to the linear light source image. The edge φ on the other end side is further shifted toward the edge search start side, so the first edge and the second edge become the separated position. Here, next, the distance (pixel number) from the first edge to the second edge is obtained as the inter-edge distance for each pixel sequence data. The outline of the obtained inter-edge distance for the position of the pixel column (the coordinate in the horizontal direction) is shown in Fig. 6(b). Then, if there is a pixel sequence whose distance between the edges is equal to or greater than the threshold 値T3, it is determined to be defective. Further, the threshold 値T3 is an arbitrary natural number' or may be one pixel. When the threshold 値T3 is 1 pixel, the pixel column whose distance between edges is non-zero is determined to be a defective picture -27- 201033602. The threshold 値T3 may be appropriately determined according to the size of the allowable defect, and the multi-dimensional image data is 256-gray (brightness 値0 to 255; 8-bit) 2-dimensional image data, for example, three good. Further, in the example of Fig. 6, the upper end is referred to as one end (the first edge search start side), but which end of the pixel column is arbitrary as one end, and the lower end may be regarded as one end. In this case, in the region where there is no defect, the first edge and the second edge coincide with the edge on the upper side of the linear light source image. This defect detection algorithm A can detect a variety of point defects with a certain degree of certainty. However, the detection of minute point defects such as bubbles or fish eyes is not true. On the other hand, this defect detection algorithm A is not suitable for the detection of line defects. Hereinafter, this defect detection algorithm A is referred to as "edge contour method 1". [Defect Detection Algorithm B] The defect detection algorithm B is a function curve that fits the edge of the linear light source image of the 2-dimensional image data, and the distance between the edge of the image of the linear light source and the function curve @ line exceeds the threshold 値T 5 The location of the (first threshold 检) is detected as a defect. The defect detection algorithm B will be described below with reference to FIG. Fig. 7 (a) shows an example of the original image data generated by one of the image capturing units 5! to 5n. The upper side of the image is the downstream side in the transport direction, and the lower side of the image is the upstream side in the transport direction. In Fig. 7(a), the strip-shaped white area extending in the lateral direction of the center is a linear light source image, and the partially twisted portion (non-smooth place) of the lower side edge of the linear light source image is a defect. -28- 201033602 In the defect detection algorithm B, the following processing is performed for each of the original image data produced by the imaging unit 5 ! 5 5n. First, the edge of the linear light source image is found from the original image data. An example of the edge of the linear light source image obtained is indicated by "() in Fig. 7(a). In the example of Fig. 7, the lower edge of the linear light source image is obtained, but the upper edge of the image of the light source may be obtained. Both sides of the upper edge side of the linear light source image may be used. φ As a method of finding the edge of the linear light source image, a margin extraction filter (for example, a Sobel filter) may be used to make the edge strong. The edge is used as the edge of the line image of the linear light source or the data of the pixel data is divided into the pixel columns of one row and the data of each pixel column is obtained as the edge method of the linear light source image to record In the method of [57] of Patent Document 1, (the method of performing 2 and marking, the area of the marked area larger than the specific area is extracted by the area of the linear light source image), etc. As Φ, it is explained that the original image data is divided into the data of the pixel column of one row and the data of each pixel is obtained as the edge method of the linear light source image. First, the original image data is divided into In the longitudinal direction (transport direction of sheet 2) The data of the pixel rows of the 1 row line. Secondly, the data of each pixel column is explored by one end (Fig. 7 (a) end) toward the other end (Fig. 7 (a) lower end) as described below. The second pixel on the one end side is regarded as the pixel of interest, and whether the brightness 値 of the attention pixel is smaller than the adjacent pixel on the one end side is smaller than the threshold 値T4 (T4 is a natural number) or more ( That is to say, the neighboring students are less than one 〇" seeking line and the next knowledge), and the square 値 domain is taken as an example and the brightness of the edge is firstly determined by the brightness of the bright pixel -29 - 201033602値 is La, and if the brightness of the pixel is Lb, La-Lb g T4). In order to detect only strong edges, the threshold 値T4 at this time is relatively large. When it is determined that the luminance 値 of the pixel of interest is smaller than the threshold 値Τ4 or more, the adjacent pixel is determined as the edge of the linear light source image, and the position of the edge of the linear light source image is recorded (adjacent pixel) Position), the data processing of the pixel column of the processing target is ended, and in other cases, the pixel of interest is moved toward the other pixel 1 pixel, and the above determination is repeated until the brightness of the pixel of interest is determined to be adjacent. The brightness of the pixel is smaller than the threshold 値Τ4 or more. When it is determined that the brightness of the pixel of interest is smaller than the brightness 邻接4 of the adjacent pixel, the adjacent pixel is determined as the edge of the linear light source image, and the linear light source image is recorded. The position of the edge (the position of the adjacent pixel) ends the processing of the data of the pixel column of the processing object. Next, the obtained edge of the linear light source image is fitted to a smooth curve represented by a function (fitting with a function curve) to obtain a fit curve (function curve). The function used for the nesting may be an n-th order function (η is 2 or more), a Gaussian function, a Lorentz function, a Voigt function, a combination of these functions, etc., wherein n is relatively small n times A function such as a quadratic function is preferred. Further, as the evaluation method of the nesting used in the tying, for example, the least square method can be used. Next, the two-dimensional image data is divided into pieces of pixel rows along the vertical direction (the transport direction of the formed sheet 2), and the data of each pixel is obtained from the fit curve to the linear light source. The distance from the edge of the image (the number of pixels) is used as the fit. . The outline of the obtained fit is plotted on the position of the pixel column (the coordinate in the horizontal direction (the direction perpendicular to the transport direction of the formed sheet 2, and the direction perpendicular to the thickness direction of the formed sheet 2 is straight -30-201033602)) Shown in Figure 7(b). Then, when there is a pixel sequence in which the fitting degree is 闽値T5 or more, it is determined that the position of the edge of the linear optical image of the pixel column is defective. Thereby, it is possible to determine whether or not there is a defect, and it is also possible to determine the position of the defect. As described above, it is possible to detect a line defect which appears as a local distortion of the edge of the linear light source image (twisting of a fine linear light source image near the edge). Further, the threshold 値T5 used for the above determination is an arbitrary natural number of φ, and may be one pixel. When the threshold 値T5 is one pixel, if there is a pixel sequence whose fitting degree is not zero, it is determined that there is a defect. The threshold 値T5 is appropriately determined depending on the size of the allowable defect, and the multi-dimensional image data is a 2nd-order image data of 256 gray scales, and 4 is preferable. Further, in this defect detection algorithm B, in addition to determining the presence or absence of a defect, the defect position can also be obtained. In this case, the pixel sequence of the threshold 値T5 or more is extracted, and the position of the pixel between the edge of the linear light source image and the φ splicing curve of the extracted pixel sequence may be obtained as the defect position. This defect detection algorithm B can detect various line defects with high certainty. On the other hand, this defect detection algorithm B is not suitable for the detection of point defects. Hereinafter, the defect detection algorithm B is referred to as "edge contour map method 2 j 〇 [defect detection algorithm C] defect detection algorithm C, which smoothes the 2 dimensional image data, and smoothes the smoothed image data and The difference between the original 2nd dimensional image data -31 - 201033602 The brightness of the material is larger than that of 闽値, and any method C smaller than Τ6Β. This defect detects the light and dark in the direction of the 2nd direction of the formed sheet) The spatial frequency division is changed, and the portion below the high-frequency component is used as the integer of the defect.) The image data is smoothed (by the high-frequency region, the low-direction smoothing filter for the lateral brightness change) The averaging of the smoothed image only leaves the original image data as the difference image data, and the difference image T6B (the fourth threshold T; T6B is an arbitrary positive number) brightness 値 at the threshold 値T6D (the fifth threshold 値; The T6D system is a positive number. The defect detection is performed as follows. Hereinafter, the defect detection calculation algorithm C will be described in detail in the horizontal direction of the linear light source image. And the thickness of the formed sheet 2 is compared with the change of the thickness of the formed sheet 2, and the high-frequency brightness of the original image data is extracted by the threshold 値T6B or above 闽値T6D to detect the situation. (1) First, Use one row of η columns (η is a 3-direction smoothing filter (row and row), and the original scene is smoothed) in the horizontal direction to obtain smoothed |, and the brightness change in the direction of the horizontal direction of the original image data is removed. Only the low frequency is left. The component (remaining lateral frequency component and vertical brightness change). As the above-mentioned device, a weighted averaging filter such as a Gaussian filter or the like can be used. Further, η is preferably 3. (2) Secondly, before subtracting from the original image data The data (the brightness of each pixel is subtracted), thereby the high-frequency component of the luminance change in the horizontal direction. (3) Then, the smoothing filter of the 3x3 pixel is used for the image data obtained by subtraction (operator, operator) Smoothing is performed by smoothing 201033602 to remove noise and leaving high frequency components other than noise. As the smoothing filter, it is preferable to use a bilateral filter. Or the smoothing of the edge is saved as in the median filter, etc. (4) Next, the upper edge of the linear light source image (the edge on the downstream side in the transport direction) and the lower edge are obtained from the original image data ( The method of obtaining the edge of the linear light source image is the same as that described for the defect φ detection algorithm B, and therefore the description thereof is omitted. Next, the transport direction of the formed sheet 2 of the original image data is On the X-axis, the minimum 値Min is found in the X coordinate 全 of the full-pixel constituting the upper edge, and the maximum 値Max is obtained from the X-axis standard of the full-pixel constituting the lower edge. Then, the minimum 値Min is subtracted. The maximum 値Max is regarded as the width W of the linear light source image, and the area where the X coordinate mark is extended from the maximum 値Max to the minimum 値Min is only the outer side moving width W extended area is defined as the inspection area. That is, the φ coordinate of the X coordinate mark above Max- (Min-Max) and Min+ (Min-Max) is defined as the inspection area. This processing concentrates the inspection target area only on the linear light source image and its vicinity. Further, the inspection area is expanded outward from the area where the X coordinate mark is from the maximum 値Max to the minimum 値Min, in order to include the distortion of the linear light source image as the inspection area 5(5), and the above (3) The brightness of the pixels in the inspection area of the smoothed image data (high-frequency components other than noise) is determined by the sub-pattern to determine the bright side (the side with the high brightness).値T6B and the dark side (low side of the brightness) 闽値T6D. -33- 201033602 Τ6Β=(Average brightness in inspection area値)+ (Standard deviation of brightness 检查 in inspection area)xk T6D =(Average brightness in inspection area値)-(Standard deviation of brightness 检查 in inspection area) )xk (parameter of k-positive number) Further, k may be appropriately determined depending on the size of the allowable defect, for example, 1. 5, 3, 4. 5 and so on. (6) Next, the total pixel in the inspection region to which the smoothed image data is applied (3) is subjected to a process of determining the brightness 値 above the threshold 値T6B or the threshold 値T6D (threshold 値 processing). A pixel having a threshold 値T6B or higher or a threshold 値T6D or lower is extracted as a defective portion. Thereby, it is possible to determine whether or not there is a defect, and it is also possible to determine the position of the defect. Further, when there is little noise included in the original image data generated by the imaging units 51 to 5, the smoothing processing of (3) may be omitted. Further, it is not necessary to concentrate the inspection target area only in the line shape. In the case of the light source image and its vicinity, the processing of defining the inspection area in (4) may be omitted, and the processing of (5) (6) may be performed on the entire image data. This defect detection algorithm C may have a high Various point defects including small defects such as bubbles or fish eyes are detected sexually. On the other hand, this defect detection algorithm C is not suitable for the detection of line defects. However, for the processing time 'other defects than the defect detection algorithm C, The detection algorithm is relatively short (the processing time of the defect detection algorithm C is, for example, about 40 ms per frame). Hereinafter, the defect detection algorithm C is referred to as a "hipass fi 11er method". 201033602 [Defect Detection Algorithm D] The defect detection algorithm D will be described below with reference to FIGS. 8 and 9. Fig. 8 shows an example of the original image data generated by one of the image capturing units. The upper side of the image is the downstream side in the transport direction, and the lower side of the image is the upstream side in the transport direction. In Fig. 8, the strip-shaped white region extending in the lateral direction of the center is a linear light source image, a dark region existing inside the linear light source image, and a small white region existing in the vicinity of the φ linear light source image, which is a defect. . In Fig. 8, the upper and lower curves of the linear light source image show the upper and lower limits of the inspection target area. In the defect detection algorithm D, the following processing is performed for each of the plurality of original image data generated by the imaging unit Si-Sn. First, the original image data is divided into data of a pixel sequence of one line along the vertical direction (the conveyance direction of the formed sheet 2), and a data point column indicating the brightness-dependent change of the position of each pixel column is obtained. The φ degree profile is illuminated as a vertical direction. An example of the obtained vertical brightness profile is shown in Fig. 9. In this example, the vertical direction luminance profile of the pixel sequence at the position indicated by the arrow in FIG. 8 is the following direction (the direction indicated by the arrow of FIG. 8: the direction opposite to the transport direction of the formed sheet 2). The y coordinate for the y-axis. Next, the depth of the valley portion is obtained for the vertical direction luminance profile of each pixel sequence (see Fig. 8). That is to say, firstly, for each vertical pixel brightness profile, all the maximum points and minimum points are obtained, and for all the minimum points obtained, the brightness of the minimum point is 値 (minimum 値) and most -35 - 201033602 The difference between the brightness 値 (maximum 値) of the maximum point close to the minimum point is obtained as the depth of the valley portion. When the depth of the portion of the valley determined is above the threshold 値T7 (T7 is a positive number), the valley portion is determined to be defective. The threshold 値 T7 may be appropriately determined according to the size of the allowable defect, and the multi-dimensional image data is a 2nd-order image data of 25 6 gray scale, for example, 0. 25x255 is preferred. This defect detection algorithm D has a relatively short processing time. This defect detection algorithm D can detect various point defects with certain degree of certainty. In particular, it is suitable for the detection of point defects in the vicinity of the edge of the linear light source image which cause local light and dark reversal. However, the point defect and the image including the vicinity thereof are necessarily high contrast, and the detection of minute point defects such as bubbles or fish eyes and tire marks is not high. On the other hand, this defect detection algorithm D is not suitable for the detection of line defects. Hereinafter, this defect detection algorithm D is referred to as a "peak-and-shoot method" [defect detection algorithm E] φ a defect detection algorithm E, which expresses a change in luminance depending on a position along a straight line of 2-dimensional image data as The brightness contour map, the data point group of the brightness contour map is assumed to be a moving point of the moving time between the data points, and the velocity vector of the foregoing mass point between the two data points before the attention data point and the foregoing attention data point The acceleration vector of the above-mentioned particle point between the three data points is used to predict the brightness of the above-mentioned pixel data point 値 'the difference between the predicted brightness 値 and the actual brightness 比 is larger than the threshold 値T8 (the third threshold 値, T8 is a natural number) Defect detection. -36- 201033602 The defect detection algorithm E is described below with reference to Fig. 10 and Fig. 1 . This defect detection algorithm Ε is to improve the accuracy of the peak 値 method, and replace the valley depth to detect the defect based on the difference between the measured 値 and the predicted 値. In this defect detection algorithm, the following processing is performed for each of the plurality of original image data generated by the imaging units 5 to 5n. First, in the same manner as the peak 値 method, the vertical brightness profile of each pixel column is obtained. An example of the obtained vertical luminance profile is shown in Fig. 10 with luminance 値 as the X φ axis. The circle portion of the vertical direction luminance profile is a contour map corresponding to the defect that is to be detected by the defect detection algorithm. Next, for the vertical-direction luminance contour map between the pixels, it is assumed that the moving time between adjacent data points is constant, and the distance from one end of the data dot column to the other end is determined so that the distance between the adjacent data points is constant. As shown in Fig. 11, the data point b from the data point c to the adjacent data point b moves from the data point b to the adjacent data point a from the data point a to the adjacent data point d. In addition, the data point d is a data point corresponding to the pixel of interest. 〇 Next, find the velocity vector and acceleration vector of the mass points of the three data points a to c that the mass point passes before the data point d. That is, the velocity vector of the mass point in the section from the data point b to the data point a is obtained from the coordinates (X coordinate, y coordinate) of the two data points a and b through which the mass point passes before the data point d. . Further, based on the movement time and the coordinates (X coordinate, y coordinate) of the data points b and c through which the mass point passes before the data point d, the velocity vector of the mass point in the section from the data point c to the data point b is obtained. The speed vector of the mass point from the point b to the data point a and the section from the data point c to the data point b -37-201033602 The speed vector of the mass point is determined, and the interval from the data point C to the data point a is obtained. The acceleration vector of the aforementioned particle. Next, the velocity vector of the mass point in the section from the data point b to the data point a, and the acceleration vector of the mass point in the section from the data point c to the data point a, predict the coordinates (position) of the data point d. The difference between the X coordinate (brightness 値) of the data point d thus predicted and the actual (measured) X coordinate (brightness 値) of the data point d is obtained, and if the difference is equal to or greater than the threshold 値T8, it corresponds to the data point d. The pixels are extracted as defective parts. Thereby, it is possible to determine whether or not there is a defect, and it is also possible to determine the position of the defect. The threshold 8T8 may be appropriately determined in accordance with the size of the allowable defect, and the multi-dimensional image data is a 2nd-order image data of 25 6 gray scales, for example, 20 is preferable. This defect detection algorithm E can detect various point defects with high certainty. Hereinafter, this defect detection algorithm E will be referred to as "peak detection method 2" [defect detection algorithm F] The defect detection algorithm F will be described below with reference to FIG. Fig. 12 (a) shows an example of original image data generated by one of the imaging units Si-Sn. The upper side of the image is on the downstream side in the transport direction, and the lower side of the image is on the upstream side in the transport direction. In Fig. 12(a), the strip-shaped white region extending in the lateral direction of the center is a linear light source image, and the partially twisted portion of the lower edge of the linear light source image (where the inclination to the horizontal line is large) is a defect. In the defect detection algorithm F, the following processing is performed for each of the plurality of original image data of -38 to 201033602 generated by the imaging units 5] to 5n. First, at least one of the edges of the linear light source image is obtained from the original image data. An example of the edge of the obtained linear light source image is indicated by "〇" in Fig. 12(a). In the example of Fig. 12, the lower edge of the linear light source image is obtained. However, the upper edge of the linear light source image may be obtained, and both the upper edge and the lower edge of the linear light source image may be obtained. The method of obtaining the edge of the linear light source image is the same as that described for the defect detection algorithm B, and therefore the description thereof will be omitted. φ Next, the second differential contour map is obtained by taking the horizontal direction as the X-axis and the vertical direction as the y-axis and the edge curve (edge contour map) of the second differential linear light source image as y=f(x). An example of the obtained second differential contour map is shown in Fig. 12(b). Then, for each pixel at the edge of the linear light source image, it is determined whether the second derivative is above the threshold 値T9 (T9 is a positive number). The subdifferential is determined as a defective portion by a pixel (at a high frequency) equal to or higher than the threshold 値T9. Thereby, it is possible to determine whether or not there is a defect, and it is also possible to determine the position of the defect. The threshold φ φ T9 may be appropriately determined in accordance with the size of the allowable defect. This defect detection algorithm F is suitable for detecting the line defects presented by the edge of the linear light source image as a local curvature. This defect detection algorithm F has a low defect detection capability. Hereinafter, this defect detection algorithm F is referred to as "edge curve method 1". [Defect Detection Algorithm G] The defect detection algorithm G' is for the edge of the linear light source image of the 2nd-dimensional image data, and finds the vicinity of each pixel (the range of -NCO 1 -39-201033602 in the vicinity) The curvature is detected as a defect at a position above the threshold 値T10 (the second threshold 値; T10 is a positive number). The defect detection algorithm G' will be described below with reference to Figs. 13(a) to 13(c). In the defect detection algorithm G, the following processing is performed for each of the plurality of original image data generated by the imaging units 5, 5n. First, at least one of the edges of the linear light source image is obtained from the original image data. An example of the edge of the linear light source image obtained is shown in the figure. 1 3 ( a ) ~ Figure 13 (c). The method of obtaining the edge of the linear light source image is the same as that described for the defect detection algorithm B, and therefore the description thereof will be omitted. Next, the curvature at each point (each pixel) is obtained for the curve of the edge of the linear light source image. The method of obtaining the curvature is not particularly limited, and a method of calculating using a mathematically determined mathematical formula may be employed. The processing time in such a method becomes long, so it is preferable to approximate the curvature by the following method. (1) Every N pixels (Fig. 13(a) to Fig. 13(c)) are left and right (or before and after) on the edge of the pixel (the black dots in Fig. 13(a) to Fig. 13(c)). The range of the white point and the eye-catching pixel (the range of 2N+1 pixels near the pixel of interest) is the calculation target range (N is a natural number). N is appropriately determined depending on the size of the allowable defect, for example, preferably 30. Fig. 13 (a) to Fig. 13 (c), where N is 3. (2) Next, calculate the pixels at both ends of the target range by a straight line. (3) The total pixel across the range of the calculation object, from which the prediction is calculated by the straight line -40-201033602 brightness 値', the actual brightness 预测 of the predicted brightness 値 (the brightness 边缘 on the edge curve) is calculated, and the increase is calculated. The amount or the absolute value of the increment. By the integrated enthalpy obtained here, the curvature of the range of 2N + 1 pixels near the pixel of interest can be sufficiently approximated (a curvature 几乎 which is almost the same as the curvature calculated using a mathematically determined mathematical formula can be obtained). Here, in the case of using the integrated calculation of the increments, as shown in FIG. 13(c), in the case where the calculation is performed, a slight luminance 値 change φ which may travel straight below the straight line may occur. Offsetting these changes is deliberately ignored, and the approximation of the curvature is obtained. On the other hand, in the case where the total enthalpy of the increase is used, even if such a change occurs, the similarity of the curvature is obtained by including such a change. If you want to move a small brightness 値 change that sometimes travels straight below the straight line in the range of the calculation target as shown in Fig. 13 (c), as long as the absolute 値 is used as the defect, It can be constructed. Conversely, in the case where such a change is allowed and is not detected as a defect, it may be configured by using the integral of the increment. 〇 (4) Move the pixel of interest from the end of the edge of the line source image to the end of one pixel and calculate the aforementioned enthalpy for all the pixels on the edge. Thereby, a contour map (curvature contour map) whose curvature approximates 値 is generated. Next, for each pixel of the edge of the linear light source image of the curvature profile, it is determined whether the curvature is equal to or greater than 闽値T10, and the pixel whose curvature is above the threshold 値T 1 0 is determined as a defective portion (or defect) Alternate). Thereby, it is possible to determine whether or not there is a defect, and it is also possible to determine the position of the defect. The formed sheet 2 is somewhat warped, so that the edge of the linear light source image is somewhat curved, so that the curvature of the edge of the linear light source image should be allowed to be regarded as a defect if it is within a certain range of -41 - 201033602. That is, the threshold 10T10 should be relatively large. The threshold 10T10 may be appropriately determined in accordance with the size of the allowable defect, and when the multi-dimensional image data is a 2nd-order image data of 25 6 gray scale, for example, 110 is preferable. This defect detection algorithm G can detect various line defects with high certainty. Hereinafter, the defect detection algorithm G is referred to as "edge curve method 2". In the present embodiment, the line defect image analysis units 6 1 to 6 1 n and the point 0 defect image analysis units 62 to 62n are respectively The combination of the line defect detection algorithm and the point defect detection algorithm used is one of the following. (A) The line defect detection algorithm used by the line defect image analysis unit όΐ-όΐη is the edge contour map method 2 or the edge curve method 2, and the point defect detection algorithm used by the point defect image analysis units 621 to 62n is Qualcomm. Filter method or peak method 2. (B) The line defect detection algorithm used by the online defect image analysis unit is the edge contour map method 2 or the edge curve method 2, and the point defect detection algorithm used by the image analysis units 62 and 62n is high-pass filter. Defect detection algorithm other than the method or the peak method. (C) The line defect detection algorithm used by the line defect image analysis unit is a point defect detection algorithm other than the edge contour map method 2 or the edge curve method 2, and the point defect detection algorithm used by the point defect image analyzing units 621 to 62n The law is high through the filter method or the peak method 2 . Among the combinations of (A) to (C), the combination of (A) is the best. In the case of a combination of (A), both line defects and point defects can be reliably detected. 42- 201033602 In the case of the combination of (B), the line defect can be surely detected. In the case of a combination of (C), point defects can be surely detected. [Embodiment 2] Another embodiment of the present invention will be described below with reference to Fig. 14 . For the convenience of the description, members having the same functions as those of the members shown in the first embodiment are denoted by the same reference numerals and will not be described. φ The defect inspection device according to the present embodiment has the same configuration as the defect inspection device 1 of the first embodiment, except for the analysis device 6 shown in Fig. 1 instead of the analysis device 6 shown in Fig. 1 . As shown in FIG. 14, the analysis device 6A omits the line defect used by the line defect image analysis unit 6 1 to 6 1 n in the present embodiment by the analysis device 6 shown in FIG. 1 . The detection algorithm is edge contour method 2 or edge curve method 2 . In the present embodiment, line defects can be surely detected. Further, the defect inspection device of this embodiment can be used alone, but it is preferably used in combination with a defect inspection device that can detect a point defect. Thereby, not only line defects but also point defects can be detected. The defect inspection device capable of detecting a point defect in combination with the defect inspection device of the present embodiment may be a known defect inspection device, but is preferably a defect inspection device of the third embodiment described later. That is, both line defects and point defects can be detected. [Embodiment 3] -43- 201033602 Another embodiment of the present invention will be described below with reference to Fig. 15. For the convenience of the description, members having the same functions as those of the members shown in the first embodiment are denoted by the same reference numerals and will not be described. The defect inspection device according to the present embodiment has the same configuration as that of the defect inspection device 1 of the first embodiment, except for the analysis device 6 shown in Fig. 1 and the analysis device 6B shown in Fig. 15 . In the analysis device 6 shown in Fig. 1, the analysis device 6 shown in Fig. 1 omits the line defect image analysis unit in the present embodiment, and the point defect image analysis units 62 and -62. The point defect detection algorithm used is a high pass filter method or a peak pass method 2 . In this embodiment, point defects can be surely detected. Further, the defect inspection device of the present embodiment can be used alone, but it is preferably used in combination with a defect inspection device capable of detecting a line defect. Thereby, not only the point defects but also the line defects can be detected. The defect inspecting device capable of detecting a line defect in combination with the defect inspecting device of the present embodiment may be a well-known defect inspecting device, but it is preferable to implement the defect detecting device of the type 2. That is, both the line defect and the point defect can be surely detected. [Experimental Example] Next, in order to confirm the effect of the present invention, 14 experimental defect inspection apparatuses similar to the defect inspection apparatus of the above-described embodiment were used to display the results of experiments. In the first to seventh experimental defect inspection devices, the image detecting units 52 to 5n are omitted from the defect inspecting device according to the third embodiment, and the transfer sheet is replaced by the transporting device - 44 - 201033602. 2 A conveyor belt placed on the surface for transport. The first to seventh experimental defect inspection devices are those in which a point defect is detected from a sample containing a point defect. The first experimental defect inspection device includes the point defect image analysis units 62 1 to 62 n using the edge contour map method 1, and the second experimental defect inspection device includes the point defect image analysis unit 62 using the edge contour map method 2, 62n, the third experimental defect inspection device includes a point-defining image analysis unit 62^62«» using a high-pass filter method, and the fourth experimental defect inspection device includes a point-defect image analysis unit 62 using a peak-and-shoot method. ～6%, the fifth experimental defect inspection device includes the image analysis unit for the defect using the peak 値 method 2, and the sixth experimental defect inspection device includes the image analysis unit 62 for the point defect using the edge curve method 1 to 62n In the seventh experimental defect inspection apparatus, the eighth-to-fourth experimental defect inspection apparatus for the point defect image analysis unit using the edge curve method 2 (the method of using the incremental total calculation method) is related to the above-described real φ application. The defect inspection device of the type 2 omits the imaging units 52 to 5n, and the transfer device 3 uses a conveyor belt on which the formed sheet 2 is placed on the surface instead of the conveyance roller. The 8th to 14th experimental defect inspection device is a user who detects a line defect from a sample containing a line defect. The eighth experimental defect inspection device includes the line defect image analysis unit 6 1 to 6 1 n using the edge contour map method 1, and the ninth experimental defect inspection device includes the line defect image analysis unit using the edge contour map method 2 Όΐ, -όΐη, the 10th experimental defect inspection device includes an image analysis unit for line defects using a high-pass filter method, κόΐη, the 11th experimental defect inspection device-45-201033602, and the image for line defect using the peak 値 method The analysis unit 611 to 61 „, the 12th experimental defect inspection device includes a line defect image analysis unit using the peak 値 method 2, and the 13th experimental defect inspection device includes a line defect image analysis unit 使用 using the edge curve method 1 - όΐη, the 14th experimental defect inspection apparatus includes the line defect image analysis unit 6 1 ! 6 6 η using the edge curve method 2 (the method of integrating the enthalpy). In the present experimental example 'as the formed sheet 2 Use a sample containing 10 types of polarizers of different types of point defects, and a sample containing 6 types of polarizers of different types of line defects. 1 sample containing line defects, including Sample 01 of bubble, sample 02 containing fisheye, sample 03 containing first foreign matter, sample 包含4 containing second foreign matter different from first foreign matter, sample 06 containing first fetal trace, and inclusion with first fetal trace Sample 07 of different second tread marks, sample 08 including the first mark, sample 09 containing the second mark different from the first mark, sample 1 1 including the first flaw, and containing the same as the first flaw The sample of the second flaw is 12. The six samples including the line defect are the sample 10 including the bent (line defect), and the sample 1 including the second bending (line defect) different from the first bending, and the sample a sample 51 along a fold line in the conveyance direction of the formed sheet 2, a sample 52 including a line which is orthogonal to the conveyance direction of the formed sheet 2, a sample 53 including a broken line which is orthogonal to the conveyance direction of the formed sheet 2, and a sample 53 including The sample 5 of the oblique direction of the conveyance direction of the formed sheet 2 is a sample 5.4. In addition, the experimental defect inspection apparatus of the 1st to 14th is used as the imaging unit 5, and the photographic second-order image is used to generate 256 gray scales. The horizontal 512 pixels X vertical 480 pixels of the secondary image data An active scanning area sensor using a CCD element 201033602. The experimental defect inspection apparatus of the first to the 14th, as the linear light source 4, is attached with a sharp edge hood (to make a linear light source image) The high-frequency fluorescent lamp of the edge-sharpening cover. In the experimental defect inspection apparatus of the first to the 14th, the conveying speed of the formed sheet 2 according to the conveyor belt was 20 mm/sec (=1. 2m/m i η). Further, the position at the distance of 145 mm from the end of the conveyor belt (the end of the front side in Fig. 2) is the center of the photographing area of the photographing unit. In addition, the 145 mm metal ruler was calculated so that the distance from the metal ruler to the defect was 55 mm. In the first to seventh experimental defect inspection apparatuses for detecting a point defect, the imaging area (field of view) on the molded sheet 2 is made horizontal (straight to the conveyance direction of the formed sheet 2 and orthogonal to the formed sheet 2) The direction of the thickness direction) 51. The position and angle of the imaging unit 5 are adjusted so that the longitudinal direction of 2 mmx (the conveyance direction of the formed sheet 2) is 48 mm. However, among the 512 pixel X4 80 pixels of the active scanning area sensor, 512 pixels are formed in the horizontal direction of the position of the φ 240 pixels (the center position in the vertical direction) from the top. The lateral direction of the surface of the sheet 2 51. The position and angle of the photography unit are adjusted in the area of 2 mm. In other words, the position of the photographing unit 5! is adjusted so that the distance from the formed sheet 2 to the photographing unit 5! is 190 mm, and the photographing direction of the photographing unit 5! (from the center of the collecting lens of the photographing unit 5) The angle of the imaging unit 5 is adjusted so that the angle between the center of the imaged portion 5 and the surface of the formed sheet 2 is 40 degrees. In this case, the resolution of the photographing unit is 100 A m/pixel. Further, in the first to 7-47 to 201033602 for detecting a point defect, the experimental defect inspection device as the active scanning area sensor has a focal length of 25 mm and a minimum F 値 of 1. 4. The lens of the C mount lens with a working distance of 270 mm at the tip end of the lens is mounted on the active scanning area sensor body, and the aperture is adjusted to about 11. Further, in the first to seventh experimental defect inspection apparatuses for detecting a point defect, the longitudinal direction of the linear light source 4 is orthogonal to the conveyance direction of the formed sheet 2, and the distance from the formed sheet 2 to the linear light source 4 is obtained. The linear light source 4 is disposed in such a manner that the straight line connecting the center of the image capturing area on the formed sheet 2 and the center of the linear light source 4 is formed at a 37 degree angle to the surface of the formed sheet 2 is 240 mm. In the 8th to 14th experimental defect inspection apparatus for detecting a line defect, the photographing area on the formed sheet 2 becomes a horizontal direction 204. The position and angle of the photographic unit are adjusted by 8 mm x 192 mm. However, among the 512 pixels of the active scanning area sensor, 512 pixels are arranged in the horizontal direction from the position of the 24th pixel (the center position in the vertical direction) from the top. The lateral direction of the surface of the sheet 2 204. The position and angle of the photographing unit 5! are adjusted in the area of 8 mm. In other words, the position of the imaging unit is adjusted so that the distance from the molded sheet 2 to the imaging unit 51 is 400 mm, and the imaging unit is adjusted so that the angle between the imaging direction of the imaging unit 5 and the surface of the molded sheet 2 is 15 degrees. 5, the angle. In this case, the resolution of the photographing unit 5 is 200; zm/pixel. In addition, in the 8th to 14th experimental defect inspection devices for detecting line defects, as the active scanning area sensor, the focal length is 25 mm, the minimum F値 is I·4, and the lens tip working distance is 4 90 mm. 201033602 lens mount C (C mount) lens mounted on the active scanning area sensor body, the aperture is adjusted to about 5 · 6 ~ 8 〇 In addition, in the 8th to 14th experimental defect inspection device for detecting line defects In the longitudinal direction of the linear light source 4, the linear light source 4 is disposed such that the transport direction of the formed sheet 2 is at an angle of 25 degrees, and the working distance of the linear light source 4 is 900 mm. Here, it is indeed possible to detect the diameter 〇. 5mm point defect mode φ Set the parameters of the defect detection algorithm. In the first and eighth experimental defect inspection devices, the threshold 値T3 of the edge contour map method 1 was set to three. In the second and ninth experimental defect inspection devices, the edge contour method 2 闽値 T5 was set to 4. In the third and tenth experimental defect inspection devices, the edge contour method 2 k is set to 4. 5. Use the edge contour method 2 to smooth the filter in the horizontal direction as a smoothing filter of 1 row and 3 columns. In the 4th and 11th experimental defect inspection devices, the threshold 値T7 of the peak 値 method is set to 25% of the maximum brightness χ〇 (255χ〇. 25). In the fifth and twelfth experimental φ defect inspection devices, 阆値Τ8 of the peak 値 method 2 was set to 20. In the sixth and thirteenth experimental defect inspection apparatuses, the distance of the parameter used by the edge curve method 1 was set to 15' and k was set to 5. In the 7th and 14th experimental defect inspection apparatuses, the method of obtaining the approximate curvature described above is used as the edge curve method 2, and the calculation target range is set to the range of 30 pixels before and after the attention pixel (even if N is IS). ), the threshold 値T10 is set to 110. Then, using the first to seventh experimental defect inspection devices, it is investigated whether or not a point defect can be detected from a sample containing one type of point defect, and the experimental defect inspection device of the eighth to the fourth is used to investigate whether or not six kinds of line defects can be included. -49- 201033602 The sample was tested for line defects and the results are shown in Table 1.
-50- 201033602-50- 201033602
邊緣曲線法2 判定 X X X X X X X X ◎ X ◎ ◎ ◎ ◎ 4A.I r 恨ffi置 格數 〇 Ο 〇 Ο ο Ο ο ΙΟ τ-· ο ια ID Ο ：·〇： ΓΟ 200 1 1 200 1 200 1 邊緣曲線法1 判定 0 < <1 X X < <1 κ ◎ X <3 X X o to 檢出畫 格數 CO 〇 Ο CM eg Ο Τ"- Ο τ·· 4^. § d o 寸 峰値法2 判定 ◎ ◎ ◎ ⑬ ◎ ◎ ◎ 窃 ◎ ◎ X X ◎ X X 00> 檢出畫 格數 卜 ί〇 to 卜 CSI •y·» CO ο <Μ Ο Ο s 〇 〇 峰値法 判定 X <3 ◎ ◎ < * ◎ Ο @ 0 X X X ◎ X X to 傲m量 格數 〇 csi o o CM Ο ο CO CO ττ— to Ο ο Ο 〇 〇 高通過濾器法 判定 • ◎ ◎ © ◎ ◎ ◎ ◎ ◎ ◎ <ι X X X X CO 給山童 恨ffi置 _數 卜 r^·· <£> ο 卜 <SI 0D ο ο ο Ο CSI • τ·· Ο 〇 〇 纖輪廓 圖法2 判定 X < X <1 X X < X ◎ X d @ ◎ ◎ © o 檢出畫 格數 〇 〇 Ο Ο Ο s ο φ 00 S c〇 I 1 s CM 邊緣輪廓 圖法1 判定 < < 0 ◎ @ 0 © 0 < X X ◎ K X CM 恨ω置 格數 IO 卜 Ο ο CM τ-· vt> ιχ> C4 Ο Ο g 04 〇 〇 忉駿 棚 嫉 f* 盛 疵 3異物 4異物 1 ©胎痕 7胎痕 Β 〇0 链 〇> 11傷痕 12傷痕 10折曲 I ffi Η γ*·»] 1 鹱 i m N 10 /—s #· I C*3 EO 藤 I 寸 10 誤報(1800畫格中） mmm 罐谨塑 -51 - 201033602Edge curve method 2 judge XXXXXXXX ◎ X ◎ ◎ ◎ ◎ 4A.I r hate ffi grid number 〇Ο ο Ο ο ΙΟ τ · ο ι ID 〇 〇 〇 〇 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Method 1 determines 0 <<1 XX <<1 κ ◎ X <3 XX o to Detects the number of frames CO 〇Ο CM eg Ο Τ"- Ο τ·· 4^. § do 寸峰値Method 2 Judgment ◎ ◎ ◎ 13 ◎ ◎ ◎ 偷 ◎ ◎ XX ◎ XX 00> Detecting the number of frames 〇 〇 C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C C 3 ◎ ◎ lt lt * 0 0 0 0 0 <ι XXXX CO To the mountain boy hate ffi _ number 卜r ^·· <£> ο 卜 <SI 0D ο ο ο Ο CSI • τ·· 〇〇 〇〇 轮廓 轮廓 轮廓 2 2 judgment X < X <1 XX < X ◎ X d @ ◎ ◎ © o Detecting the number of frames 〇〇Ο Ο Ο s ο φ 00 S c〇I 1 s CM Edge contour method 1 Determination << 0 ◎ @ 0 © 0 < XX ◎ KX CM hate ω grid number IO divination ο CM τ-· vt>ιχ> C4 Ο Ο g 04 〇〇忉 嫉 嫉 嫉 f* 疵 疵 3 foreign bodies 4 foreign objects 1 © Tire marks 7 Fetal marks 〇0 Chain 〇> 11 scars 12 scars 10 folds I ffi Η γ*·»] 1 鹱im N 10 /-s #· IC*3 EO vine I inch 10 false positives (1800 frames Medium) mmm tank plastic-51 - 201033602
在第1〜7實驗用缺陷檢査裝置，被攝影的成形薄片2 之動畫之全圖框數（全擷取枚數）爲1 50枚’ 1個缺陷檢 測對象點或點缺陷進入攝影部5!的攝影區域而由攝影部 之攝影區域離開爲止的期間內’30圖框之動畫以攝影 部5,攝影的。亦即，若缺陷經常可見到’以攝影部5 1攝 影的2次元影像資料之中’前述期間內攝影的30枚（30 圖框）之2次元影像資料內會包含著缺陷的影像。在表1 之點缺陷之各行，於前述期間內攝影的30圖框之2次元 影像資料之中，藉由各實驗用缺陷檢查裝置檢測出缺陷的 圖框（畫格）數目（被檢測出缺陷的次數）記載於「檢出 畫格數」之欄。In the first to seventh experimental defect inspection devices, the number of full frames (the total number of images) of the imaged formed sheet 2 is 1,500. One defect detection target point or point defect enters the imaging unit 5! In the period from the photographing area of the photographing unit, the animation of the '30 frame is photographed by the photographing unit 5. In other words, if the defect is often seen in the "two-dimensional image data captured by the photographing unit 51", the defective image is contained in the 30-dimensional (30-frame) 2-dimensional image data photographed during the period. In each row of the defect of Table 1, the number of frames (frames) in which defects were detected by each of the experimental defect inspection devices among the two-dimensional image data of the 30 frames photographed in the above period (defects were detected) The number of times is listed in the column of "Checking the number of frames".
在第8〜1 4實驗用缺陷檢査裝置，1個缺陷檢測對象 點進入攝影部5,的攝影區域而由攝影部5,之攝影區域離 開爲止的期間內，3 00圖框之動畫被攝影部5!攝影的。在 表1之線缺陷之各行，於前述期間內攝影的300圖框之2 次元影像資料之中，藉由各實驗用缺陷檢査裝置檢測出缺 陷的圖框（畫格）數目記載於「檢出畫格數」之欄。 於表1，檢出畫格數爲〇代表無法檢測出缺陷，檢出 畫格數的大小，顯示缺陷檢出的精度或是確實性。例如檢 出畫格數爲1〜2畫格那樣很少的數目的場合，認爲缺陷 檢出的精度很低。在表1，藉由本案申請人獨自定下的步|」 斷基準，把缺陷檢出之精度分類爲「X」（檢出畫格數爲〇 畫格）、「△」（檢出畫格數爲1〜2畫格）、「〇」（ 檢出畫格數爲3〜6畫格）、「◎」（檢出畫格數爲7畫 -52- 201033602 格以上）之4個階段，記載於「判定」之欄位。 此外，於表1，「誤報」之行，顯示於1 800圖框（畫 格）中有幾個圖框發生誤報。 由表1之結果，可知藉由使用高通過濾器法以及峰値 法2之缺陷檢査裝置（第3及第5實驗用缺陷檢查裝置） ，可以精度佳地檢測出所有種類之點缺陷。亦即，於相關 於前述實施型態之缺陷檢査裝置，使用點缺陷用影像解析 ❿ 部62 之點缺陷檢測演算法若爲高通過濾器法或峰値 法2的話，可以精度佳地檢測出所有種類之點缺陷。 此外，由表1之結果，可知藉由使用邊緣輪廓圖法2 以及邊緣曲線法2之缺陷檢查裝置（第2及第7實驗用缺 陷檢查裝置），可以精度佳地檢測出所有種類之線缺陷。 亦即，於相關於前述實施型態之缺陷檢査裝置，使用線缺 陷用影像解析部6 1 i〜6 1 η之線缺陷檢測演算法若爲邊緣輪 廓圖法2或邊緣曲線法2的話，可以精度佳地檢測出所有 φ 種類之線缺陷。 此外，於表1之幾乎所有的場合，檢出畫格比最大値 (點缺陷的場合爲3 0，線缺陷的場合爲3 0 0 )還小，所以 在本實驗例使用於缺陷的檢測之複數枚之2次元影像資料 (動畫資料）之中僅根據1枚之2次元影像資料（靜止影 像資料）而檢測缺陷的方法，有無法檢測出缺陷的可能性 。例如，在本實驗例使用於缺陷的檢測之複數枚2次元影 像資料之中隨機選擇1枚2次元影像資料，僅根據該丨枚 2次元影像資料要檢測出包含於樣品13的折曲的場合，即 -53- 201033602 使使用可以確實檢測出折曲的缺陷檢測演算法（邊緣曲線 法2 )也只能夠以8/300之低的機率檢測到折曲。 本發明並不被限定於前述各實施型態，申請專利範圍 所示的範圍內可以進行種種變更，針對適當組合不同的實 施型態所分別揭示的技術手段而得到的實施型態也被包含 於本發明之技術範圍。 【圖式簡單說明】 @ 圖1係顯示相關於本發明的一實施型態之缺陷檢查裝 置的主要部構成的功能方塊圖。 圖2係顯示前述缺陷檢查裝置的槪觀之模式圖。 圖3係供說明從前技術的問題點之圖。 圖4係供說明從前技術的問題點之圖。 圖5係供說明從前技術的問題點之圖。 圖6係供說明缺陷檢測演算法之一例（邊緣輪廓圖法 1 )之圖。 @ 圖7係供說明缺陷檢測演算法之其他例（邊緣輪廓圖 法2 )之圖。 圖8係供說明缺陷檢測演算法之其他例（峰値法）之 圖。 圖9係供說明缺陷檢測演算法之其他例（峰値法）之 圖。 圖1 0係供說明缺陷檢測演算法之其他例（峰値法2 ) 之圖 -54- 201033602 圖11係供說明缺陷檢測演算法之其他例（峰値法2 ) 之圖。 圖12係供說明缺陷檢測演算法之其他例（邊緣曲線 法1 )之圖。 圖1 3 ( a )係供說明缺陷檢測演算法之其他例（邊緣 曲線法2)之圖。 圖1 3 ( b)係供說明缺陷檢測演算法之其他例（邊緣 φ 曲線法2 )之圖。 圖1 3 ( c )係供說明缺陷檢測演算法之其他例（邊緣 曲線法2 )之圖。 圖14係顯示相關於本發明的其他實施型態之缺陷檢 査裝置的主要部構成的功能方塊圖。 圖15係顯示相關於本發明的進而其他之實施型態之 缺陷檢查裝置的主要部構成的功能方塊圖。 φ 【主要元件符號說明】 1 :缺陷檢查裝置 2 :成形薄片 3:搬送裝置（移動手段） 4 :線狀光源 5n :攝影部（攝影手段） 6 :解析裝置 6A :解析裝置 6B :解析裝置 -55- 201033602 6 1 !〜6 1 n :線缺陷用影像解析部（線缺陷檢測手段） 62 !〜62n :點缺陷用影像解析部（點缺陷檢測手段）In the 8th to 14th experimental defect inspection apparatus, the imaging unit of the 300-frame frame is in the period from when the one defect detection target point enters the imaging area of the imaging unit 5 and the imaging area 5 is separated from the imaging area. 5! Photography. In each row of the line defect of Table 1, among the 2nd-dimensional image data of the 300 frame photographed in the above period, the number of frames (frames) detected by each experimental defect inspection device is described in "Detection" The column of the number of frames. In Table 1, the number of detected frames is 〇, the defect cannot be detected, the size of the number of frames is detected, and the accuracy or authenticity of the defect detection is displayed. For example, when it is found that the number of frames is a small number of 1 to 2 frames, the accuracy of the defect detection is considered to be low. In Table 1, the accuracy of the defect detection is classified as "X" (the number of detected frames is 〇), and "△" (detected frame) by the step taken by the applicant in this case. The number is 1 to 2 frames), "〇" (the number of frames detected is 3 to 6 frames), and "◎" (the number of frames detected is 7 to -52 - 201033602). It is recorded in the field of "judgment". In addition, in Table 1, the "false positives" line shows that several frames in the 1 800 frame (picture) have false positives. As a result of Table 1, it is understood that all types of point defects can be accurately detected by using the high-pass filter method and the defect inspection device (the third and fifth experimental defect inspection devices) of the peak enthalpy method 2. In other words, in the defect inspection apparatus relating to the above-described embodiment, the point defect detection algorithm using the point defect image analysis unit 62 can accurately detect all of the point defect detection algorithms if the high pass filter method or the peak 値 method 2 is used. A defect in the type. Further, as a result of Table 1, it is understood that all types of line defects can be accurately detected by using the edge contour map method 2 and the edge inspection method 2 (the second and seventh experimental defect inspection devices). . In other words, in the defect inspection apparatus according to the above-described embodiment, the line defect detection algorithm using the line defect image analysis unit 6 1 i to 6 1 η may be the edge contour method 2 or the edge curve method 2 Line defects of all φ types are detected with high precision. In addition, in almost all cases of Table 1, the detected frame is smaller than the maximum 値 (30 for point defects and 300 for line defects), so it is used for the detection of defects in this experimental example. Among the plurality of binary image data (animation data), the defect is detected based on only one of the two-dimensional image data (still image data), and there is a possibility that the defect cannot be detected. For example, in the experimental example, a plurality of 2-dimensional image data are randomly selected from a plurality of binary image data used for detecting defects, and only the bending of the sample 13 is detected based on the two-dimensional image data. , that is, -53- 201033602 The defect detection algorithm (edge curve method 2) which can detect the bending accurately can also detect the bending only at a low probability of 8/300. The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the embodiments obtained by appropriately combining the technical means disclosed in the different embodiments are also included. The technical scope of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a functional block diagram showing the configuration of a main part of a defect inspection apparatus according to an embodiment of the present invention. Fig. 2 is a schematic view showing the appearance of the aforementioned defect inspection device. Fig. 3 is a diagram for explaining the problem of the prior art. Fig. 4 is a view for explaining the problem of the prior art. Fig. 5 is a diagram for explaining the problem of the prior art. Fig. 6 is a diagram for explaining an example of the defect detection algorithm (edge contour map method 1). @ Figure 7 is a diagram illustrating another example of the defect detection algorithm (edge contour method 2). Fig. 8 is a view for explaining another example of the defect detection algorithm (peak method). Fig. 9 is a view for explaining another example of the defect detection algorithm (peak method). Fig. 10 is a diagram for explaining another example of the defect detection algorithm (peak method 2) -54 - 201033602 Fig. 11 is a diagram for explaining another example of the defect detection algorithm (peak method 2). Fig. 12 is a view for explaining another example of the defect detection algorithm (edge curve method 1). Figure 1 3 (a) is a diagram illustrating another example of the defect detection algorithm (edge curve method 2). Figure 1 3 (b) is a diagram illustrating another example of the defect detection algorithm (edge φ curve method 2). Figure 1 3 (c) is a diagram illustrating another example of the defect detection algorithm (edge curve method 2). Fig. 14 is a functional block diagram showing the main configuration of a defect inspecting apparatus according to another embodiment of the present invention. Fig. 15 is a functional block diagram showing the configuration of main parts of a defect inspection apparatus according to still another embodiment of the present invention. φ [Description of main component symbols] 1 : Defect inspection device 2 : Molded sheet 3 : Transport device (moving means) 4 : Linear light source 5 n : Photographing unit (photographing means) 6 : Analysis device 6A : Analysis device 6B : Analysis device - 55- 201033602 6 1 !~6 1 n : Image analysis unit for line defect (line defect detection means) 62 !~62n : Image analysis unit for point defect (point defect detection means)
-56--56-