1272688 九、發明說明： 【發明所屬之技術領域】 本發明是有關—種光學影像檢測之技術，尤其是利用 胃_換之原理’將具有複數個週期 列圖案之灰階影像進行處理，進而應时瑕紐測之= 頻域遮罩與其實財法以及使用該賴料 瑕疵之檢驗方法。 $1272688 IX. Description of the invention: [Technical field of invention] The present invention relates to a technique for optical image detection, in particular to processing a gray scale image having a plurality of periodic column patterns by using the principle of gastric _changing, and further Time 瑕 New Test = Frequency Domain Mask and Actual Finance Method and the test method used. $
【先前技術】 隨著科技的進步，半導體工業以及液晶顯示器產 已經變成我國的主流ι#，由於半導體以及液晶顯 器產業之生產其主要就是以批次式生產（b时比 production)在同一片晶圓上或者是基板上形呈複數個 期性排列的圖案，於製程完畢之後再切割封裝或者是 晶製程階段。然而在生產的過程當中良率的掌握相當重 要:以液晶面板製造產業為例，在陣列工程階段中就要進 行溥膜、頁光以及蝕刻等加工製程，而在製程階段難免合 因為製程環境或者是灰塵（Paniele)造成產品有缺陷ς 疵，引此檢測的工作就相當重要，幾乎在每一道製程下就 會有對應之檢測流程。 。而在檢測的的方式中，有打光進行人工目視檢測缺陷 或者是自動檢測缺陷，由於人工的不確定性極高，因此自 的檢測在半導體工業或者是液晶面板製造業中就顯得 』田重要。而在自動化的檢測過程中，擷取影像來進行判 °貝的方式為現在此領域之技術中的主流。 5 1272688[Prior Art] With the advancement of technology, the semiconductor industry and liquid crystal display production have become the mainstream of China, because the production of semiconductor and liquid crystal display industry is mainly in batch production (b is more than production) in the same film. A plurality of periodic patterns are formed on the wafer or on the substrate, and the package is cut or crystallized after the process is completed. However, in the process of production, the mastery of yield is very important: taking the liquid crystal panel manufacturing industry as an example, in the array engineering stage, processing processes such as ruthenium film, page light, and etching are required, and in the process stage, it is inevitable because of the process environment or It is the dust (Paniele) that causes the product to be defective. The work that leads to this test is very important, and there will be a corresponding inspection process in almost every process. . In the method of detection, there is a light for manual visual inspection of defects or automatic detection of defects. Since the artificial uncertainty is extremely high, the self-detection appears to be important in the semiconductor industry or the liquid crystal panel manufacturing industry. . In the automated detection process, the method of capturing images to perform the judgment is the mainstream in the current technology in this field. 5 1272688
而習用之技術當中，比較常用的為先以一無瑕症之影 像為基準影像’然後將需要被判定之影像與該基準影像進 行比對，不過由於此種方式需要考慮到定位補正之問題， 如果定位不夠精準，則會導致誤判狀況產生。 又有Orbetech所提出之轉移膜（Die t〇 die)檢驗法， 其係比較週邊®案相同位置下的像素灰階值，從而決定瑕 疵所在的位置。然而其方法僅適用於小尺度瑕疵 (micro-defect)，當週邊圖案楚於大尺度瑕疵時 (Macro-defect)時，可能無法察覺大尺度瑕疵的存在因 產生誤判。 學術期刊上另有一種方法是直接利用數位化傅利業 換（Discrete Fourier Transform)及其逆轉換來進行 疵檢驗。這個方法的缺點是計算時間太長，而且目前 已發表的頻域遮罩形狀不恰當，會造成瑕疵影像的模1 化，因而無法精確判斷瑕疮的尺寸與形狀。 、, 綜合上述，因此亟需一種頻域遮罩與其實現方法以 使用該頻域遮罩檢測重複圖案瑕疵之檢驗方法來解決^ 技術所產生之問題。 ' ^ 【發明内容】 本發明的主要目的是提供一種頻域遮罩與其實現方去 以及使用該頻域遮罩檢測重複圖案瑕疵之檢驗方法，其係 利用一可以過濾頻率強度之頻域遮罩來實現一影像濾、= 器，達到重建不含瑕疵之灰階樣本影像進而檢測匈斷該瑕 疯之位置’達到瑕疵自動檢測判斷之目的。 ^ 6In the conventional technique, it is more common to use an image of a flawless image as a reference image and then compare the image to be determined with the reference image, but since this method needs to take into account the problem of positioning correction, Inaccurate positioning can lead to misjudgment. There is also a Die t〇 die test by Orbetech, which compares the pixel grayscale values at the same position in the perimeter case to determine where the 瑕 is located. However, the method is only applicable to small-scale micro-defects. When the peripheral pattern is large-scale, it may not be able to detect the existence of large-scale defects due to misjudgment. Another method in academic journals is to use the Digital Fourier Transform and its inverse transformation to perform the 疵 test. The disadvantage of this method is that the calculation time is too long, and the current published frequency domain mask shape is not appropriate, which will cause the 瑕疵 image to be molded, so that the size and shape of the acne cannot be accurately determined. In view of the above, there is a need for a frequency domain mask and its implementation method to solve the problem caused by the technique by using the frequency domain mask to detect the repeated pattern 瑕疵. The main object of the present invention is to provide a frequency domain mask and its implementation, and a method for detecting a repeating pattern using the frequency domain mask, which utilizes a frequency domain mask that can filter the frequency intensity. To achieve an image filter, =, to achieve the reconstruction of the gray-scale sample image without sputum and to detect the location of the smashing madness 'to achieve the purpose of automatic detection and judgment. ^ 6
I272688 、本發明的次要目的是提供一種頻域遮罩與其實 /及使用該頻域遮罩檢测重複圖案瑕疵之檢驗方法，方去 該頻域遮罩係具有複數個增益值為丨之頻率窗口以其中 週期性重複排列之灰階影像之正常影像通㉟，相 ^有I272688, a secondary object of the present invention is to provide a frequency domain mask and a test method for detecting a repeating pattern using the frequency domain mask, wherein the frequency domain mask has a plurality of gain values. The frequency window is normal image 35 of the gray-scale image periodically arranged in a periodic manner.
之頻率區域之增錄為〇以阻卻職影像通過 自D 重建影像失真率之目的。細降低 、本發明的另一目的是提供一種頻域遮罩與其實現方 以及使用該頻域遮罩檢測重複圖案瑕疵之檢驗方法，其^ =用—可以過濾頻率強度之頻域遮罩來實現一影像濾波 二，達到重建不含瑕疵之灰階樣本影像進而檢測判斷該瑕 =之位置，其中該影像濾波器係以兩空間矩陣相減而實 見以達到減少不必要之運算而增加檢測速度之目的。 ^為了達到上述之目的，本發明提供一種頻域遮罩，其 係形成於一二維頻率平面上，該頻域遮罩可對具有複數個 週期性排列圖案之一灰階影像的頻域影像進行頻率強度過 Ί °亥頻域遮罩包括·一第一頻區窗口，其係設置於該二 維頻率平面之中心，該第一頻區窗口具有一第一增益值； 複數個第二頻區窗口，其係以該第一頻區窗口中心為中心 依序排列使得相鄰之該第一頻區窗口中心與該第>頻區窗 口中心以及相鄰兩第二頻區窗口之中心於一頻率第一袖向 上相距一固定寬度以及於一頻率第二轴向上相距/固定長 度，該複數個第二頻區窗口其係具有一第二增益侬；以及 第三頻區’其係設置於該第一頻區窗口與該複數個第^一 頻區窗口以及相鄰之該複數個第二頻區窗口之間，該第三 頻區係具有一第三增益值。 7 1272688 該i诚、疮一增益值之一第—頻區窗口且設置於 ;第二，之中心；複數個第二頻區窗口，其係以 二’貝區窗口中心為中心依序排列使得相鄰之 區窗口中心與該第二頻區窗口中心以及 頻區窗口之中心於—頻率第一轴向上 固疋寬度以及於一頻率第二軸向上相距一 二t度’該複數個第二頻區窗口其係具有-第二The addition of the frequency region is to prevent the job image from reconstructing the image distortion rate from D. Fine reduction, another object of the present invention is to provide a frequency domain mask and its implementation side and a detection method for detecting a repeating pattern using the frequency domain mask, which can be implemented by filtering a frequency domain mask of frequency intensity. An image filtering method 2 is used to reconstruct a grayscale sample image without 瑕疵 and to detect and determine the position of the 瑕=, wherein the image filter is subtracted by two spatial matrices to reduce the unnecessary operation and increase the detection speed. The purpose. In order to achieve the above object, the present invention provides a frequency domain mask formed on a two-dimensional frequency plane, and the frequency domain mask can be used for a frequency domain image having a gray scale image of a plurality of periodically arranged patterns. Performing a frequency intensity over Ί 亥 频 频 频 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 第一 第一 第一 第一 第一 第一 第一 第一 第一 第一 第一 第一 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥 亥a zone window, which is arranged in the order of the center of the first frequency zone window so that the center of the first frequency zone window and the center of the first frequency zone window and the center of the adjacent two frequency zone windows are a frequency first sleeve is spaced apart from a fixed width and a second axial distance/fixed length at a frequency, the plurality of second frequency region windows having a second gain 侬; and a third frequency region ′ The third frequency region has a third gain value between the first frequency region window and the plurality of first frequency region windows and the plurality of adjacent second frequency region windows. 7 1272688 The first-frequency region window of one of the i-sense and sore-gain values is set at; the second, the center; and the plurality of second frequency-region windows are sequentially arranged centering on the center of the window of the two's bay area The center of the adjacent zone window and the center of the second frequency zone window and the center of the frequency zone window are at a frequency first axial fixed width and at a frequency second axial distance of one or two t degrees 'the plurality of second Frequency zone window has a - second
與該複數置於該第一頻區窗口 一 ’員區_ 口以及相鄰之該複數個第 :頻區窗口之間，該第三頻區係具有一第 值，以及 、e)使該第 增盈值為〇、該第二頻率增益值為1以及 該弟三頻率增益值為〇; m::亥頻域遮罩以形成-影像濾波器； g 像濾波器過濾該灰階影像，以形成不含瑕疵 之一樣本影像； ⑻設定至少—灰階差異門檻值；以及 ⑴將該樣本5,上之像素與該灰階影像之像素逐-比對’獲得每-個像素之灰階差異值而與該至少一 灰1¾差異門播值進行比較以判對該灰階影像對應 於該樣本影像之像素是否為瑕疵。 【實施方式】 為使貝審查委員能對本發明之特徵、目的及功能有更 進〆少的認知與瞭解，下文特將本發明之裝置的相關細部 1272688 • 結構以及設計的理念原由進行說明，以使得審查委員可以 • 了解本發明之特點，詳細說明陳述如下·· 請參閱圖五c所示，該圖係為本發明頻域遮罩之較佳實 • 施例示意圖。本發明係提供一種頻域遮罩7，其係形成於 ‘ 一二維頻率平面70上，該頻域遮罩7可對具有週期性複數 個圖案之/灰階影像（如圖二A所示）的頻域影像（如圖二B - 所不）進行頻率強度過濾，該頻域遮罩7包括··一第一頻區 窗口 71、複數個第二頻區窗口 72以及〆第三頻區73。該 • 第一頻區窗口 71，其係設置於該二維頻率平面70之中心， 該第一頻區窗口 71具有一第一增益值；該複數個第二頻區 窗口 72 ,其係以該第一頻區窗口 71中心為中心依序排列 使得相鄰之該第一頻區窗口 71中心與該第二頻區窗口 72a、72c中心以及相鄰該第二頻區窗口 72a、72b以及72b、 72c之中心於頻率第一軸91向上相距一固定寬度w以及於 頻率第二軸92向上相距一固定長度L，該複數個第二頻區 窗口 72其係具有一第二增益值；該一第三頻區73，其係 • 設置於該第一頻區窗口 7丨與該複數個第二頻區窗口 72以 及相鄰之該複數個第二頻區窗口之間，該第三頻區73係具 有一第三增益值。該第一增益值、該第二增益值以及該第 三增益值係介於大於等於〇以及小於等於1之間。 在本較佳實施例中，該第一增益值係為零、該第二增益 值係為1以及該第二增益值係為零。而該固定寬度w係為 ωχ=2 7Γ /Tx ’其中該Τχ係為該灰階影像投射於該第二軸（縱 軸）向上之週期；該固定長度L係為0y=2;r/Ty，其中該Ty係 為該灰階影像投射於該第一軸（橫軸）向上之週期。該第一 !272688 頻區窗α 71以及該第二頻區窗口 以及一牲料Ζ係^、有一特徵長度h 特徵見度w，其中該特徵寬 係分別為W2以及〇y/2,其中仏心/τ 3特徵長以 該灰階影傻於身+於正够 x且該Tx係為 之以像技射於平面第二軸（縱軸）向上之頻 = ’ 0y=2;r/Ty ’且該Ty係為該灰階影像投射^ 第:轴向（橫軸）上之頻域圖像而得之週期。讀於千面 請參閱圖二A以及圖所示，其中圖二 個:，性排列圖案之灰階影像圖；圖二β係為氣 生排列之灰階影像圖20可以是半導體日曰二:圖 二像:黃光製程中所使用之光罩影像或者是薄 次曰曰顯示面板上電路影像等。當該灰階影像2〇經由3曰# 傅利葉轉換（Discrete Fourier Transform)之後會；^ 圖二^所示之頻域圖像21。其中圖二B中較亮的 代表該灰階影像能量較強之頻譜訊號位置。藉由圖可 以了解該灰階影像20轉換至頻域時，能量強之頰:亮點 12並非只集中於特定之區域例如通過該頻域中心之^二頻 率轴10以及第二頻率轴11上（如圖一所示），反而是分布 於整個頻域平面上等間格之次中心點位置附近區域疋^該 灰階圖像出現若干個瑕疵影像時，該瑕疵影像之賴域象 通常會集中於該頻域中心區域。 由前述之結果，本發明提供一種影像濾波器之實現方 法，用以過濾含有瑕疵之複數個週期性排列圖案之灰階影 像中的瑕疵影像，其流程步驟如圖三A所示，該影像^ = 器之實現方法是包括有： 彳心彳 12 ^272688 以及有限脈波長濾波器（Finite Impulse Response 正如^ j原理而設計。其中’該ωχ=2 τγ /Τχ，ωy=2 7Γ /Ty 〇 圖案…述圖二β所示，由於含有瑕疵具有週期性排列重複 域圖階影像其瑕疵影像經由數位傅利葉轉換之後其頻 量則备t集中於頻域之中心，而相關於正常灰階影像之能 此如㈢^布於頻域平面上複數個中心點附近區域之中，因 果此夠设計如圖五C所示之頻域遮罩將屬於瑕疵之訊 =強度過濾，而保留屬於影像之訊號強度，這樣在原始影 、重建的時候就會得到解析度較高之樣本，以提高缺陷檢 測之檢出率’以及保留較精確之瑕疵形狀，從而提高 瑕疲型態分類之可靠度。 、只 請參閱圖三B所示，該圖係為本發明影像濾波器之實現 方法中形成該頻域遮罩之較佳實施例流程示意圖。形成該 頻域遮罩包括有步驟341以該角頻率以及決定具 一特徵寬度以及特徵長度之該第一頻區窗口與該第二頻區 窗口；決定了窗π大小係為了決定可以通過之頻率幅 圍，接下來參閱圖五Α所示，進行該步驟342於具有= 率第一軸91以及-頻率第二軸92之-二維頻率區域5〇: 該第一頻區窗口 71置於該二維頻率區域50之中心，俅 第一頻區窗口 71外緣與該二維頻率區域50之間的區域;; 其增益值為零，以形成一第一頻域遮罩5 ;然後， 步驟343於該第-頻域遮罩5上以複數個該第 ^/ 72以該第-頻域遮罩5之中心為中心依序排列，使：： 該第一頻區窗〇 71中心與該第二頻區窗π 72a、72c中 以及相鄰兩第二頻區窗口 721)、7此與72 心 ί乙D又中心相 14 1272688 及該灰階影像投射於該平面第一軸上之週期 ‘ Ty ； 步驟43-將該週期Tx、Ty轉換成角頻率ωχ、; ^ 步驟44-以該ωχ以及〇¥形成一頻域遮罩，該頻域遮罩 係包括具有一第一增益值之一第一頻區窗口 且設置於該頻域遮罩之中心；複數個第二頻區 窗口，其係以該第一頻區窗口中心為中心依序 排列使得相鄰之該第一頻區窗口中心與該第 籲二頻區窗口中心以及相鄰該第二頻區窗口之 中心於頻率第一轴向上相距一固定寬度以及 於頻率第二軸向上相距一固定長度，該複數個 第二頻區窗口其係具有一第二增益值；一第三 頻區，其係設置於該第一頻區窗口與該複數個 第二頻區窗口以及相鄰之該複數個第二頻區 窗口之間，該第三頻區係具有一第三增益值； 以及 • 步驟45-使該第一增益值為0或接近0、該第二頻率增 益值為1或接近1以及該第三頻率增益值為〇 或接近0 ; 步驟46-實現該頻域遮罩以形成一影像濾波器； - 步驟47-以該影像濾波器過濾該灰階影像，以形成不含 瑕疫之一樣本影像， 步驟48-設定至少一灰階差異門檻值；以及 步驟49-將該樣本影像上之像素與該灰階影像之像素逐 一比對，獲得每一個像素之灰階差異值而與該 18 1272688 ι 至少一灰階差異門檻值進行比較以判對該灰 階影像對應於該樣本影像之像素是否為瑕疵。 利用上述之步驟可以應用於很多領域，例如LCD的 array段、偏光板貼附前、偏光板貼附後的瑕疵自動光學 仏驗、半導體廢晶元每一道姓刻程序之後的瑕庇自動光學 松驗、半導體廠晶元蝕刻之光罩的瑕疵自動光學檢驗、半 、導體製程中任何具有重複圖案之光罩、半成品或成品的瑕 - 疵自動光學檢驗以及任何非半導體製程中具有重複圖案 . 之半成品或成品的瑕疵自動光學檢驗等只要是具有週期 陵重複排列之灰階影像都可以應用本方法建立出不含瑕 疵影像之樣本，然後再利用該步驟49判斷出瑕疵所在之 位置。由於利用該步驟46所產生之影像濾波器在該步驟 47中所產生之該樣本影像解晰度相當高，因此在利用該步 驟49進行像素（pixei t〇 pixel)的逐步比對時，可以得 到相當精準之瑕疵判斷。 。亥衫像滤波器的數學形式是一個矩陣，其中非零元素 I 愈少則該步驟47之序所花的時間愈少。此外，這個遮罩 渡波器的非零元素為同一個整數時，該步驟47之程序將 只涉及整數的加法，而不涉及其它運算，因而使得該步驟 47所花的時間極小化。 唯以上所述者，僅為本發明之較佳實施例，當不能以之 限制本發明範圍。即大凡依本發明申請專利範圍所做之均 等變化及修飾，仍將不失本發明之要義所在，亦不脫離本 發明之精神和範圍，故都應視為本發明的進一步實施狀 況。例如在步驟45中之增益值選擇僅為一實施例之方式， 19 l272688 .由於增錢之選擇係取決於想要域之峨強度，因此當 =情況而有所變更。又如本發明偏好-個所有非零元素皆 ^同-個整數的遮罩濾波器，但是在相同原理上可以產生 ^種計算效率接近此理想形式的其它遮罩濾波器。這些 〜匕^式的遮罩m皆在本專利聲請保護的範圍之列。 、’”丁、σ上述’本發明由於具有可發揮檢測速度快以及精準 ^憂點所以可以滿足業界之需求，進而提高該產業之競爭 .，誠已符合發明專利法所規定申請發明所需具備之要 牛’故爰依法呈提發明專利之巾請，謹請貴審查委員允 揆時間惠予審視，並賜準專利為禱。 【圖式簡單說明】 圖一係為習用技術使用之頻域圖像示意圖。 圖,Α係為具有複數個週期性排㈣案之灰階影像圖。 圖二B係為具有複數個週期性排·案之灰階影像圖轉換 成頻域影像圖。And the complex number is placed between the first frequency region window and the adjacent plurality of first frequency region windows, the third frequency region has a first value, and e) The gain value is 〇, the second frequency gain value is 1 and the third frequency gain value is 〇; m:: the sea frequency domain mask is formed to form an image filter; the g image filter filters the gray scale image to Forming a sample image that does not contain 瑕疵; (8) setting at least the grayscale difference threshold value; and (1) aligning the pixel on the sample 5 with the pixel of the grayscale image to obtain a grayscale difference of each pixel The value is compared with the at least one gray sigma difference value to determine whether the grayscale image corresponds to a pixel of the sample image. [Embodiment] In order to enable the Beck Review Committee to have a lesser understanding and understanding of the features, objects and functions of the present invention, the following details of the structure and design concept of the device of the present invention are described below. The review board member can understand the characteristics of the present invention, and the detailed description is as follows. Please refer to FIG. 5c, which is a schematic diagram of a preferred embodiment of the frequency domain mask of the present invention. The present invention provides a frequency domain mask 7 formed on a two-dimensional frequency plane 70, which can be used to image a grayscale image having a plurality of periodic patterns (as shown in FIG. 2A). The frequency domain image (as shown in FIG. 2B - no) performs frequency intensity filtering, and the frequency domain mask 7 includes a first frequency domain window 71, a plurality of second frequency domain windows 72, and a third frequency region. 73. The first frequency region window 71 is disposed at a center of the two-dimensional frequency plane 70, the first frequency region window 71 has a first gain value; and the plurality of second frequency region windows 72 are The center of the first frequency zone window 71 is centered in order such that the center of the first frequency zone window 71 and the center of the second frequency zone window 72a, 72c and the adjacent second frequency zone windows 72a, 72b and 72b, The center of 72c is spaced apart from the first axis 91 by a fixed width w and the second axis 92 is upwardly spaced apart by a fixed length L. The plurality of second frequency bin windows 72 have a second gain value; a tri-band region 73, which is disposed between the first frequency region window 7 and the plurality of second frequency region windows 72 and the adjacent plurality of second frequency region windows, the third frequency region 73 Has a third gain value. The first gain value, the second gain value, and the third gain value are between greater than or equal to 〇 and less than or equal to one. In the preferred embodiment, the first gain value is zero, the second gain value is 1 and the second gain value is zero. The fixed width w is ω χ=2 7 Γ /Tx ' where the Τχ is the period in which the gray scale image is projected upward on the second axis (vertical axis); the fixed length L is 0 y=2; r/Ty The Ty is a period in which the grayscale image is projected on the first axis (horizontal axis) upward. The first !272688 frequency zone window α 71 and the second frequency zone window and a living system have a feature length h characteristic visibility w, wherein the feature width is W2 and 〇y/2, respectively. The /τ 3 feature is long in that the gray-scale shadow is stupid in the body + is just enough x and the Tx is such that the image is shot on the plane second axis (vertical axis) upward frequency = ' 0y = 2; r / Ty ' And the Ty is a period obtained by projecting the frequency domain image on the first axial direction (horizontal axis) of the grayscale image. Read Fig. 2A and the figure shown in Fig. 2, where Fig. 2: Grayscale image of the sexual arrangement pattern; Fig. 2: Grayscale image of the β system is an aerial arrangement. Figure 20 can be a semiconductor diary 2: Figure 2: The image of the mask used in the process of yellow light or the circuit image on the thin display panel. After the grayscale image 2〇 is transmitted through the 3曰# Discrete Fourier Transform; the frequency domain image 21 shown in Fig. 2^. The brighter one in Figure 2B represents the position of the spectral signal with stronger energy of the grayscale image. It can be seen from the figure that the grayscale image 20 is converted to the frequency domain, and the energy is strong: the bright spot 12 is not concentrated only on a specific region, for example, through the frequency axis center of the frequency domain and the second frequency axis 11 ( As shown in FIG. 1 ), instead, it is distributed in the vicinity of the sub-center position of the equal-interval plane on the entire frequency domain plane. When the gray-scale image has several 瑕疵 images, the 赖 image of the 瑕疵 image is usually concentrated. In the center of the frequency domain. According to the foregoing results, the present invention provides a method for implementing an image filter for filtering a 瑕疵 image in a grayscale image of a plurality of periodically arranged patterns containing 瑕疵, the flow of which is shown in FIG. 3A. The implementation method of the = device includes: 彳 彳 12 ^ 272 688 and finite pulse wavelength filter (Finite Impulse Response is designed according to the principle of ^ j. where 'ω χ = 2 τ γ / Τχ, ω y = 2 7 Γ / Ty 〇 pattern As shown in Fig. 2β, since the 瑕疵 image has a periodic arrangement of repeated domain map images, the 瑕疵 image is converted to the center of the frequency domain by digital Fourier transform, and the energy associated with the normal gray scale image If (3) ^ is placed in the vicinity of a plurality of center points on the plane of the frequency domain, the result is that the frequency domain mask shown in Figure 5C is designed to belong to the signal = intensity filtering, and the signal belonging to the image is retained. Intensity, so that in the original shadow, reconstruction will get a higher resolution sample, in order to improve the detection rate of defect detection 'and retain a more accurate shape, thus improving the fatigue pattern For the reliability of the class, please refer to FIG. 3B, which is a schematic flowchart of a preferred embodiment of forming the frequency domain mask in the implementation method of the image filter of the present invention. Step 341: determining the first frequency region window and the second frequency region window having a feature width and a feature length by using the angular frequency; determining a window π size to determine a frequency range that can be passed, and then referring to FIG. Α, the step 342 is performed on the second-axis frequency region 5 having the = first axis 91 and the second second frequency 92: the first frequency region window 71 is placed at the center of the two-dimensional frequency region 50, a region between the outer edge of the first frequency region window 71 and the two-dimensional frequency region 50; its gain value is zero to form a first frequency domain mask 5; then, step 343 is masked in the first frequency domain The cover 5 is sequentially arranged with a plurality of the first/the 72th centers centered on the first frequency domain mask 5, such that: the center of the first frequency domain window 71 and the second frequency domain window π 72a, 72c and adjacent two second frequency zone windows 721), 7 and 72, and the center phase 14 1272688 and a period in which the gray scale image is projected on the first axis of the plane 'T; Step 43 - Converting the period Tx, Ty into an angular frequency ω χ; ^ Step 44 - forming a frequency domain mask with the ω χ and 〇 ¥, The frequency domain mask includes a first frequency region window having a first gain value and is disposed at a center of the frequency domain mask; and a plurality of second frequency region windows centered on the center of the first frequency region window Arranging sequentially such that the center of the adjacent first frequency zone window and the center of the second frequency zone window and the center of the adjacent second frequency zone window are at a fixed width in the first axial direction of the frequency and second in frequency A plurality of second frequency domain windows have a second gain value; a third frequency region is disposed in the first frequency domain window and the plurality of second frequency domain windows and Between adjacent plurality of second frequency domain windows, the third frequency zone has a third gain value; and • step 45 - making the first gain value 0 or close to 0, the second frequency gain value 1 or close to 1 and the third frequency gain value is 〇 or close to 0; Step 46 - implementing the frequency domain mask to form an image filter; - Step 47 - filtering the gray scale image by the image filter to form a sample image without a plague, and step 48 - setting at least one gray level a threshold value of the difference; and step 49 - comparing the pixel on the sample image with the pixel of the grayscale image to obtain a grayscale difference value of each pixel and comparing with the at least one grayscale difference threshold value of the 18 1272688 ι It is determined whether the grayscale image corresponds to whether the pixel of the sample image is 瑕疵. The above steps can be applied to many fields, such as the array segment of the LCD, the pre-attachment of the polarizing plate, the automatic optical inspection after the polarizing plate is attached, and the auto-optical loosening after each of the semiconductor waste wafers. Inspection, semiconductor wafer etched reticle 瑕疵 automatic optical inspection, semi-conductor process, any repeating pattern of reticle, semi-finished or finished product, 瑕-疵 automatic optical inspection and repeating pattern in any non-semiconductor process. The automatic optical inspection of the semi-finished or finished product, etc., can be applied to the gray-scale image with the periodic arrangement of the repeating arrangement, and the method can be used to establish the sample without the flawed image, and then the step 49 is used to determine the position of the flaw. Since the resolution of the sample image generated in the step 47 by the image filter generated by the step 46 is relatively high, when the stepwise comparison of pixels (pixei t〇pixel) is performed by using the step 49, Quite accurate judgment. . The mathematical form of the jersey image filter is a matrix in which the less the non-zero element I, the less time it takes for the sequence of step 47. Moreover, when the non-zero elements of the masked waver are of the same integer, the procedure of step 47 will involve only the addition of integers, and no other operations are involved, thus minimizing the time taken by step 47. The above is only the preferred embodiment of the present invention, and the scope of the present invention is not limited thereto. It is to be understood that the scope of the present invention is not limited by the spirit and scope of the present invention, and should be considered as a further embodiment of the present invention. For example, the gain value selection in step 45 is only one embodiment, 19 l272688. Since the choice of the money increase depends on the strength of the desired domain, it is changed when the situation is changed. In another aspect, the present invention prefers a mask filter in which all non-zero elements are the same integer, but on the same principle, other mask filters having computational efficiencies close to this ideal form can be produced. These 匕^-type masks m are all covered by the scope of this patent. ''Ding, σ' above] The invention can meet the needs of the industry by virtue of its fast detection speed and accuracy, and thus enhance the competition in the industry. It has met the requirements of the invention patent law. If you want to raise a patent for the invention, please ask the review committee to allow the time to review and grant the patent as a prayer. [Simplified illustration] Figure 1 is the frequency domain used by the conventional technology. The image is a gray-scale image with multiple periodic rows (4). Figure 2B is a gray-scale image with multiple periodic rows converted into a frequency domain image.
圖三A係為本發㈣㈣波器之實現方法録實施例流程 示意圖。 頻域遮 圖二B係為本發明影像濾波器之實現方法中形成誃 罩之較佳實施例流程示意圖。 μ 圖二C、D、Ε係為本發明影像遽波器 頻域遮罩之較佳實施例流程示意圖。 之 實現方法中實現該 方法較佳實 圖四係為本發明具有重複圖案影像之瑕錄測 施例流程示意圖。 圖五Α係為本發明第一頻域遮罩之較佳實施例示意 20 1272688 圖五B係為本發明第二頻域遮罩之較佳實施例示意圖。 ^ 圖五C係為本發明頻域遮罩之較佳實施例示意圖。 【主要元件符號說明】 、 1 -頻域圖像 10 -第一頻率轴 1卜第二頻率軸 12-頻域亮點 _ 20-灰階影像 21_頻域圖像 211-頻域亮點 3- 流程 31〜35_步驟 341〜344-步驟 351〜353 -步驟 3511〜3513-步驟 φ 3521 〜3523-步驟 4- 流程 41〜49 -步驟 5- 第一頻域遮罩 ' 50-二維頻率區域 ‘ 51-增益值為零之區域 6- 第二頻域遮罩 7- 頻域遮罩 70-二維頻率區域 21 1272688 71-第一頻區窗口 ’ 72、72a、72b、72c-第二頻區窗口 73-第三頻區 ^ 91 -頻率第一轴 . 92-頻率第二軸 h-特徵長度 „ w-特徵寬度 L-固定長度 • W-固定寬度Figure 3A is a schematic diagram of the flow of an embodiment of the method for implementing the (four) (four) wave device of the present invention. Frequency Domain Masking FIG. 2B is a schematic flow chart of a preferred embodiment of forming a mask in the implementation method of the image filter of the present invention. μ Figure 2 C, D, and Ε are the flow diagrams of a preferred embodiment of the frequency domain mask of the image chopper of the present invention. The implementation of the method in the implementation method is better. The fourth embodiment is a schematic diagram of the process of recording a repeated pattern image. Figure 5 is a schematic diagram of a preferred embodiment of the first frequency domain mask of the present invention. 20 1272688 Figure 5B is a schematic diagram of a preferred embodiment of the second frequency domain mask of the present invention. ^ Figure 5C is a schematic diagram of a preferred embodiment of the frequency domain mask of the present invention. [Main component symbol description], 1 - frequency domain image 10 - first frequency axis 1 second frequency axis 12 - frequency domain bright spot _ 20 - grayscale image 21_ frequency domain image 211 - frequency domain bright spot 3 - flow 31~35_Steps 341~344-Steps 351~353 - Steps 3511~3513-Step φ 3521~3523-Step 4 - Flow 41~49 - Step 5 - First Frequency Domain Mask '50-Two-Dimensional Frequency Area' 51 - region where the gain value is zero 6 - second frequency domain mask 7 - frequency domain mask 70 - two-dimensional frequency region 21 1272688 71 - first frequency region window '72, 72a, 72b, 72c - second frequency region Window 73 - Third frequency zone ^ 91 - Frequency first axis. 92 - Frequency second axis h - Feature length „ w - Feature width L - Fixed length • W - Fixed width
22twenty two