JPH10214327A - Method of detecting surface defect using image - Google Patents
Method of detecting surface defect using imageInfo
- Publication number
- JPH10214327A JPH10214327A JP9016038A JP1603897A JPH10214327A JP H10214327 A JPH10214327 A JP H10214327A JP 9016038 A JP9016038 A JP 9016038A JP 1603897 A JP1603897 A JP 1603897A JP H10214327 A JPH10214327 A JP H10214327A
- Authority
- JP
- Japan
- Prior art keywords
- image
- inspection
- residual
- sample
- check
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Length Measuring Devices By Optical Means (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
Description
【０００１】[0001]
【発明の属する技術分野】本発明は，同じ撮像条件で撮
影した規準画像と検査画像とから，工業部品などの被検
査物表面に存在する文字や模様に対しても安定に良品と
不良品を判別する検査方法に関するものである。BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a standard image and an inspection image photographed under the same imaging conditions, and stably detects non-defective products and defective products even for characters and patterns existing on the surface of an inspection object such as industrial parts. It relates to an inspection method for determining.
【０００２】[0002]
【従来の技術】以下に，従来の代表的な画像を用いた表
面検査方式を説明する。まず，濃度の２値化処理による
検査方式の例を図１０に示す。2. Description of the Related Art A conventional typical surface inspection method using images will be described below. First, FIG. 10 shows an example of the inspection method by the binarization processing of the density.
【０００３】ここでは，例えば図１０（ａ）の画像中央
部の矩形領域の欠けや膨らみ具合を判定する問題を考え
る。矩形領域の内部と外部は，それぞれ材質的に一様
で，点線部の濃度断面は理想的な条件下では図１０
（ｃ）の左上に示すような，ステップ状のパターンとな
っていると仮定する。Here, for example, consider a problem of determining the degree of chipping or swelling of a rectangular area at the center of an image in FIG. The inside and outside of the rectangular area are materially uniform, and the concentration cross section of the dotted line under ideal conditions is shown in FIG.
It is assumed that the pattern has a step-like pattern as shown in the upper left of (c).
【０００４】２値化処理を行うためのしきい値を選ぶ方
法として，図１０（ｂ）に示すような濃度ヒストグラム
において最も２つの山が分かれる濃度値を選ぶ方法があ
る。しかし，通常は入力としてそのような理想的なステ
ップ状のパターンは得られないため，実際には照明条件
の変動や汚れに起因する濃度変動が重畳したパターンに
対して，図１０（ｃ）の右側に示すようなしきい値で２
値化を行う。As a method of selecting a threshold value for performing binarization processing, there is a method of selecting a density value at which two peaks are most separated in a density histogram as shown in FIG. However, since such an ideal step-like pattern is not normally obtained as an input, in actuality, a pattern shown in FIG. Threshold 2 as shown on the right
Perform value conversion.
【０００５】ところが，多くの場合，上述のような変動
要因を考慮・予測してしきい値を設定するのは困難であ
るから，その設定を誤ると，図１０（ｄ）のように上記
の濃度変動の影響を直接受けてしまい，斜線部のような
変動成分によって安定した形状検出ができなくなる。よ
って，表面欠陥の検出信頼度が低下する。However, in many cases, it is difficult to set the threshold value while considering and predicting the above-mentioned fluctuation factors. If the setting is erroneous, as shown in FIG. It is directly affected by the density fluctuation, and stable shape detection cannot be performed due to a fluctuation component such as a hatched portion. Therefore, the reliability of detecting a surface defect is reduced.
【０００６】次に，濃度正規化相関による検査方式の例
を図１１に示す。図１１（ａ）の規準画像（大きさ：Ｉ
×Ｊ画素）に対して，図１１（ｂ）に示すような，検査
対象となる規準ウィンドウ（大きさ：Ｋ×Ｌ画素）の濃
度配列Ｓ kl（ｋ＝１，…，Ｋ；ｌ＝１，…，Ｌ）の濃度
パターンを，良品のテンプレートとして登録する。図１
１（ｃ）の検査すべき画像パターンに対して，規準ウィ
ンドウを左上隅から右下隅まで走査し，各走査ポイント
（ｉ，ｊ）に対してすべて，以下の演算で相関係数ｒij
（ｉ＝１，…，Ｉ；ｊ＝１，…，Ｊ；−１≦ｒij≦１）
を算出する。Next, an example of an inspection method using a density normalized correlation
Is shown in FIG. The reference image (size: I) of FIG.
× J pixels), as shown in FIG.
Darkness of the target reference window (size: K × L pixels)
Degree array S kl(K = 1,..., K; l = 1,..., L)
The pattern is registered as a good template. FIG.
1 (c) for the image pattern to be inspected,
Scans the window from the upper left corner to the lower right corner, at each scan point
For (i, j), the correlation coefficient rij
(I = 1, ..., I; j = 1, ..., J; -1≤rij≦ 1)
Is calculated.
【０００７】[0007]
【数１】 (Equation 1)
【０００８】ただし，Ｔijは検査画像パターンの濃度配
列，上に線の付いたＳは規準ウィンドウ内の濃度平均
値，上に線の付いたＴは検査画像パターン中の規準ウィ
ンドウ内の濃度平均値である。Where T ij is the density array of the inspection image pattern, S with a line above is the density average value in the reference window, and T with the line above is the density average in the reference window in the inspection image pattern. Value.
【０００９】こうして得られた各走査ポイント（ｉ，
ｊ）の相関係数ｒijのうち，景大値を探索し，それが１
に近いほど良品であると判定する。この相関係数ｒ
ijは，濃度のオフセット変動に対しては不変な値となる
が，照明条件の変動など不均一な濃度変動に対しては，
図１１（ｄ）の規準濃度パターン（太線）と検査画像濃
度パターン（細線）で囲まれた変動成分（斜線部）のた
めに，良品に対しても相関係数ｒijにバラツキが生じて
しまい，誤判定の原因となる。Each of the scanning points (i,
j) is searched for a scenic maximum value, and if it is 1
It is determined that the nearer the product is, the better the product is. This correlation coefficient r
ij is an invariable value with respect to the density offset fluctuation, but it is a non-uniform density fluctuation such as the fluctuation of the lighting condition.
Due to the fluctuation component (hatched portion) surrounded by the reference density pattern (thick line) and the inspection image density pattern (thin line) in FIG. 11D, the correlation coefficient r ij varies even for non-defective products. , Which may cause erroneous determination.
【００１０】[0010]
【発明が解決しようとする課題】図１２に被検査物の画
像パターンの一例を示す。図１２（ａ）は理想的なパタ
ーン例であるが，実際に得られる良品の画像は，図１２
（ｂ）〜（ｄ）のように，ノイズや文字や被検査物固有
の模様といった良品自体のバラツキに起因した濃度変動
の影響を受けたものとなり，図１２（ｅ）に例示したよ
うな不良品の画像に類似してくる。FIG. 12 shows an example of an image pattern of an object to be inspected. FIG. 12 (a) shows an example of an ideal pattern.
As shown in FIGS. 12 (b) to (d), the density is affected by the density fluctuation due to the variation of the non-defective product itself such as noise, characters, and patterns unique to the inspection object. It resembles a good image.
【００１１】このような状況下で良品と不良品を間違い
なく判別する必要があるが，従来の技術では欠陥の検出
が困難であるため，こうした濃度変動に対して欠陥の検
出信頼度が低下するという問題点があった。In such a situation, it is necessary to correctly discriminate a good product from a defective product. However, it is difficult to detect a defect with the conventional technology, and thus the reliability of detecting a defect is reduced due to such a density fluctuation. There was a problem.
【００１２】本発明は上記の問題点の解決を図り，図１
２に示すような状況において，良品間のバラツキや被検
査物固有の模様が存在しても，良品／不良品を高信頼度
で判別できる検査方法の提供を目的としている。The present invention has been made to solve the above problems, and FIG.
It is an object of the present invention to provide an inspection method capable of discriminating a non-defective / defective product with high reliability even in a situation as shown in FIG.
【００１３】[0013]
【課題を解決するための手段】上記目的を達成するため
に，第１の発明は，被検査物のうち規準となる代表的な
良品サンプルを選び，その画像（規準画像という）を撮
影する第１の処理と，前記第１の処理と同じ撮像条件下
で，検査するサンプルの画像（検査画像という）を取得
する第２の処理と，前記規準画像と前記検査画像との濃
度差分画像を求める第３の処理と，前記規準画像のエッ
ジ抽出により領域分割を行う第４の処理と，前記第４の
処理で得られた各領域毎に残差の各しきい値に対する残
差分散の変化量を計算し，その最大値を求める第５の処
理と，前記第５の処理で求めた変化量の最大値をもと
に，当該検査サンプルの良否判定を行う第６の処理とを
有することを特徴とする。In order to achieve the above object, a first aspect of the present invention is to select a representative non-defective sample as a reference from among inspection objects, and take an image of the sample (referred to as a reference image). 1, a second process of acquiring an image of a sample to be inspected (referred to as an inspection image) under the same imaging conditions as the first process, and obtaining a density difference image between the reference image and the inspection image. A third process, a fourth process of performing area division by edge extraction of the reference image, and a change amount of a residual variance with respect to each threshold value of a residual for each region obtained in the fourth process. And a sixth process of determining the acceptability of the test sample based on the maximum value of the change amount obtained in the fifth process. Features.
【００１４】また，第２の発明は，前記第１の発明にお
いて，第５の処理が，第３の処理で得られた差分画像を
用いて，第４の処理で求められた各領域毎にしきい値を
変えた場合の当該差分画像の分散の変化量を計算して行
うことを特徴とする。[0014] In a second aspect based on the first aspect, the fifth processing uses the difference image obtained in the third processing for each area determined in the fourth processing. The method is characterized in that the change amount of the variance of the difference image when the threshold value is changed is calculated and performed.
【００１５】本発明では，まず被検査物の規準画像を良
品のテンプレートとして，それと検査画像との差分画像
を求めることで，正常な部分の文字や模様が及ぼす濃度
変動の影響を抑制し，欠陥のある部分との差異を強調す
る。そして，各しきい値に対する残差の分散の振る舞い
に着目し，残差に関する最適なしきい値を規準画像のエ
ッジに基づき分割された各領域毎に決定することによっ
て，検査の高信頼化を図る。According to the present invention, first, a reference image of the inspection object is used as a non-defective template, and a difference image between the reference image and the inspection image is obtained. To emphasize the difference from the part with. Focusing on the behavior of the variance of the residual with respect to each threshold, the optimum threshold for the residual is determined for each region divided based on the edge of the reference image, thereby improving the reliability of the inspection. .
【００１６】[0016]
【発明の実施の形態】本発明の実施の一形態について，
図面に基づいて詳細に説明する。図１に，本発明の実施
の形態に係る一連の基本的な処理手順の一例を示す。本
実施の形態は，第１から第６までの処理段階を有する。DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described.
This will be described in detail with reference to the drawings. FIG. 1 shows an example of a series of basic processing procedures according to the embodiment of the present invention. This embodiment has first to sixth processing steps.
【００１７】第１の処理１０１では，被検査物のサンプ
ルから良品規準となりうる代表的なサンプルを選択し
て，ＣＣＤカメラ等で撮影し規準画像を取得する。第２
の処理１０２では，前記第１の処理１０１において規準
画像を取得した系と同じ照明系と撮像系で，検査したい
サンプルを撮影し，検査サンプルの画像を取得する。In a first process 101, a representative sample which can be a non-defective standard is selected from samples of a test object, and a standard image is obtained by photographing with a CCD camera or the like. Second
In the process 102, a sample to be inspected is photographed by the same illumination system and imaging system as the system that acquired the reference image in the first process 101, and an image of the inspection sample is acquired.
【００１８】第３の処理１０３では，濃度正規化相関マ
ッチングにより前記規準画像と前記検査画像の位置合わ
せ処理を行い，各画素毎に両者の濃度差の絶対値を計算
し，差分画像を求める。In a third process 103, the reference image and the inspection image are aligned by density normalized correlation matching, and the absolute value of the density difference between the two is calculated for each pixel to obtain a difference image.
【００１９】第４の処理１０４では，規準画像の微分強
度をしきい値処理してエッジを抽出し，その結果に基づ
いて領域分割を行う。第５の処理１０５では，前記第４
の処理１０４で得られた各領域毎に，残差のしきい値ｔ
ｈを所定の計算により算出した初期値から順に上げてい
き，残差がしきい値ｔｈ以上の最大連結領域の分散をそ
の都度計算する。前しきい値（ｔｈ−１）との分散の減
少量を算出し，その最大値および該最大値を与えるしき
い値ｔｈを求める。In a fourth process 104, an edge is extracted by performing threshold processing on the differential intensity of the reference image, and a region is divided based on the result. In the fifth processing 105, the fourth processing
Threshold t of the residual for each region obtained in the process 104
h is sequentially increased from an initial value calculated by a predetermined calculation, and the variance of the largest connected region whose residual is equal to or larger than the threshold th is calculated each time. The amount of decrease in variance with the previous threshold value (th-1) is calculated, and the maximum value and the threshold value th that gives the maximum value are obtained.
【００２０】第６の処理１０６では，前記第５の処理１
０５で求めた最大値が所定の値よりも大きければ，しき
い値ｔｈを２値化しきい値として採用し，しきい値処理
した２値画像を検出結果として表示する。逆に，求めた
値が小さく該しきい値が設定できなければ良品と判定す
る。In the sixth processing 106, the fifth processing 1
If the maximum value obtained in step 05 is larger than a predetermined value, the threshold value th is adopted as a binarization threshold value, and the binary image subjected to the threshold value processing is displayed as a detection result. Conversely, if the obtained value is small and the threshold value cannot be set, it is determined to be good.
【００２１】以下では，本実施の一形態による上記の各
段階での処理例を具体的に説明する。図２は本実施の形
態で用いた画像入力系の概略図である。図中の２０１は
円環型蛍光灯照明，２０２はＣＣＤカメラ（６４０×４
８０画素），２０３は試料台，２０４は検査用サンプル
を表す。ここでは，切りキズや打痕といったある程度の
長さや面積を有する面的な広がりを有するような表面欠
陥を検出する場合を取り上げた。Hereinafter, an example of the processing in each of the above steps according to this embodiment will be described in detail. FIG. 2 is a schematic diagram of the image input system used in the present embodiment. In the drawing, reference numeral 201 denotes an annular fluorescent lamp illumination, and 202 denotes a CCD camera (640 × 4).
Reference numeral 203 denotes a sample stage, and 204 denotes an inspection sample. Here, a case is described in which a surface defect such as a cut scratch or a dent is detected, which has a certain length and area and has a planar spread.
【００２２】まず，図２に示すような画像の入力系で被
検査物の規準画像を取得する。次に，検査したいサンプ
ルの画像（検査画像）を撮影する。以下説明のため，規
準画像をＩR （ｉ，ｊ）（ｉ＝１，…，６４０；ｊ＝
１，…，４８０），検査画像をＩS （ｉ，ｊ）（ｉ＝
１，…，６４０；ｊ＝１，…，４８０）とし，その中で
検査対象となる規準ウィンドウ（大きさ：Ｋ×Ｌ画素）
の濃度パターンをＷR （ｋ，ｌ）（ｋ＝１，…，Ｋ；ｌ
＝１，…，Ｌ）とおく。First, a reference image of an object to be inspected is acquired by an image input system as shown in FIG. Next, an image (inspection image) of the sample to be inspected is taken. For the following description, the reference image is represented by I R (i, j) (i = 1,..., 640; j =
, 480), the inspection image is represented by I S (i, j) (i =
, 640; j = 1,..., 480), in which a reference window to be inspected (size: K × L pixels)
The density pattern W R (k, l) ( k = 1, ..., K; l
= 1,..., L).
【００２３】濃度差分画像を求めるために，良品の規準
ウィンドウに対応する検査画像中の濃度パターン領域を
濃度正規化相関マッチングにより探索する。すなわち，
規準ウィンドウを左上隅から右下隅まで走査し，各走査
点（ｉ，ｊ）に対して，すべて，以下の演算で相関係数
ｒij（ｉ＝１，…，Ｉ；ｊ＝１，…，Ｊ）を算出する。In order to obtain a density difference image, a density pattern area in the inspection image corresponding to the non-defective reference window is searched by density normalized correlation matching. That is,
The reference window is scanned from the upper left corner to the lower right corner, and for each scanning point (i, j), the correlation coefficient r ij (i = 1,..., I; j = 1,. J) is calculated.
【００２４】[0024]
【数２】 (Equation 2)
【００２５】ただし，上に線の付いたＷR は規準ウィン
ドウ内の濃度平均値，上に線の付いたＩS は検査画像パ
ターン中の規準ウィンドウ内の濃度平均値である。相関
値ｒ（ｉ，ｊ）の最大値を与える点を（ｉ* ，ｊ* ）と
おくと，この走査点において最も位置合わせの精度が出
ているとみなし，差分画像ＷD （ｋ，ｌ）[0025] However, W R marked with a line above the average density in the reference window, the I S marked with a line above an average density value of the reference window in the test image pattern. If the point giving the maximum value of the correlation value r (i, j) is set to (i * , j * ), it is considered that the position registration accuracy is the highest at this scanning point, and the difference image W D (k, l )
【００２６】[0026]
【数３】 (Equation 3)
【００２７】を求める。また，次式に示すように規準ウ
ィンドウの各画素毎に１次微分強度Δ（ｋ，ｌ）を求め
る（Ｐｒｅｗｉｔｔのオペレータ）。Is obtained. Also, as shown in the following equation, the primary differential intensity Δ (k, l) is obtained for each pixel of the reference window (Prewitt operator).
【００２８】[0028]
【数４】 (Equation 4)
【００２９】この微分強度Δ（ｋ，ｌ）をしきい値処理
すれば，規準ウィンドウのエッジ画像が得られる。抽出
されたエッジに基づいて，それらエッジやウィンドウ矩
形の線分によって囲まれた閉領域を一つの領域とみな
し，順次ラベリング付けする．こうして分割された各領
域Ａi 毎に，残差の最大値ｒｅｓi を求める。ただし，
規準画像のエッジ付近はマスキングし，その部分は差分
画像の画素値（残差）を０とした。If the differential intensity Δ (k, l) is subjected to threshold processing, an edge image of the reference window can be obtained. Based on the extracted edges, the closed area surrounded by the edges and the line segment of the window rectangle is regarded as one area, and labeling is sequentially performed. The maximum value res i of the residual is obtained for each of the divided areas A i . However,
The vicinity of the edge of the reference image is masked, and the pixel value (residual) of the difference image is set to 0 in that portion.
【００３０】そして，各領域Ａi 毎にしきい値をｔｈで
残差ＷD （ｋ，ｌ）〔（ｋ，ｌ）∈Ａi 〕に関して２値
化し，画素レベルの連結性解析を行うと，（残差がｔｈ
以上の）領域が抽出される。その中で面積最大の領域
（最大連結領域）を求め，その領域の分散ｖａｒ（ｔ
ｈ）を計算する（図３参照）。Then, the threshold value is binarized with respect to the residual W D (k, l) [(k, l) ∈A i ] for each region A i by th, and a pixel-level connectivity analysis is performed. (The residual is th
The above region is extracted. The area with the largest area (maximum connected area) is determined, and the variance var (t
h) is calculated (see FIG. 3).
【００３１】図４は，しきい値を変化させたときの差分
画像の残差分散の推移例を示し，図５は，しきい値を変
化させたときの差分画像の残差分散の，前しきい値のそ
れとの減少量の推移例を示す。なお，図５において負
（の数量）は増加（量）を意味している。FIG. 4 shows an example of transition of the residual variance of the difference image when the threshold value is changed, and FIG. 5 shows the change in the residual variance of the difference image when the threshold value is changed. 6 shows an example of a transition of the amount of decrease from that of the threshold. In FIG. 5, a negative value (a quantity) means an increase (amount).
【００３２】しきい値ｔｈを初期値ｔｈ0 からｒｅｓi
まで変化させた時，残差分散は例えば図４のように，し
きい値が小さいと良品のバラツキやノイズの影響で変動
し，さらに上げていくと次第に減少するが，不良品の場
合には欠陥がまとまった領域として抽出される前後で大
きく変化する。The threshold th is changed from the initial value th 0 to res i
When the threshold is small, the residual variance fluctuates due to the variation of non-defective products and the effect of noise as shown in FIG. 4 and gradually decreases as the threshold value is further increased. It changes greatly before and after a defect is extracted as a grouped area.
【００３３】つまり，前しきい値ｔｈ−１に対する減少
量δ（ｔｈ） δ（ｔｈ）＝ｖａｒ（ｔｈ−１）−ｖａｒ（ｔｈ） は，例えば図５のように極大値をもつことが予想され
る。そこで，残差分散の減少量を最大にするｔｈを求
め，その値を残差に関する２値化しきい値として採用
し，欠陥検出に用いる。すなわち，判定規則としては， δ（ｔｈ）＜Ｃｏｎｓｔ．ならば，良品； δ（ｔｈ）≧Ｃｏｎｓｔ．ならば，不良品； となる。Ｃｏｎｓｔ．は良否判定のしきい値であり，検
査対象によって異なる。なお，残差しきい値の初期値ｔ
ｈ0 は，現在着目している領域Ａi の，規準画像の濃度
の標準偏差，あるいは検査画像の各領域における残差の
期待値の最小値のいずれか小さい方の数値を用いる。That is, the decrease amount δ (th) δ (th) = var (th−1) −var (th) with respect to the previous threshold value th−1 is expected to have a maximum value as shown in FIG. Is done. Therefore, th which maximizes the reduction amount of the residual variance is obtained, and its value is adopted as a threshold for binarizing the residual and used for defect detection. That is, as a determination rule, δ (th) <Const. If δ (th) ≧ Const. If so, it is defective. Const. Is a threshold value for pass / fail judgment, which differs depending on the inspection target. The initial value of the residual threshold value t
As h 0 , the smaller value of the standard deviation of the density of the reference image of the area A i currently focused on or the minimum value of the expected value of the residual in each area of the inspection image is used.
【００３４】以上のように，本実施の形態では，規準画
像と検査画像を用いて，規準画像のエッジ抽出結果に基
づき領域分割を行い，各領域毎に残差しきい値を変化さ
せたときの両者の差分画像の分散を計算する。そして，
その分散の変化に着目して最適なしきい値を決定する処
理により，検査サンプル表面の文字や固有の模様，良品
間のバラツキに起因する画像の濃度変動が存在する環境
においても，良品／不良品判別の誤り率が小さく信頼度
が高い表面欠陥検出方法が実現できる。As described above, in the present embodiment, the region is divided based on the edge extraction result of the reference image using the reference image and the inspection image, and the residual threshold value is changed for each region. The variance of the difference image between the two is calculated. And
The process of determining the optimal threshold value by focusing on the variation of the variance enables the non-defective / defective products even in the environment where the density of the image due to the characters and unique patterns on the surface of the test sample and the variation between non-defective products exists. It is possible to realize a surface defect detection method with a low error rate of the determination and high reliability.
【００３５】[0035]
【実施例】本発明の方法を適用して得られた処理例を以
下に示す。図６は１０円硬貨の画像を示す図，図７はキ
ズのついた１０円硬貨の画像を示す図であり，図６中に
設定した１２８×１２８画素の規準ウィンドウ（処理ウ
ィンドウ）に対して正規化相関マッチングにより位置合
わせを行った結果を図７中に矩形（位置合わせ後のウィ
ンドウ）で表している。The following is an example of processing obtained by applying the method of the present invention. FIG. 6 is a view showing an image of a 10-yen coin, and FIG. 7 is a view showing an image of a 10-yen coin with a flaw. The reference window (processing window) of 128 × 128 pixels set in FIG. The result of the alignment performed by the normalized correlation matching is represented by a rectangle (window after the alignment) in FIG.
【００３６】この２枚の画像から前記ウィンドウに対し
て処理した結果，残差しきい値が２３と求まり（図８参
照），そのしきい値で２値化すれば図９に示すような欠
陥検出の結果が得られる。このように，ウィンドウ中に
含まれる文字部分や模様に左右されることなく，キズだ
けが検出されている。As a result of processing the window from the two images, a residual threshold value of 23 is obtained (see FIG. 8). If the threshold value is binarized, a defect is detected as shown in FIG. Is obtained. As described above, only the flaw is detected without being influenced by the character portion or the pattern included in the window.
【００３７】[0037]
【発明の効果】以上説明したように，本発明によれば，
被検査物の表面に文字や固有の模様などのテクスチャが
存在しても，良品／不良品判別の誤り率が小さく信頼度
が高い表面欠陥検出方法を実現することが可能となる。As described above, according to the present invention,
Even if a texture such as a character or a unique pattern is present on the surface of the inspection object, it is possible to realize a highly reliable surface defect detection method with a small error rate for non-defective / defective product discrimination.
【図１】本発明における処理手順の一例を示すフローチ
ャートである。FIG. 1 is a flowchart illustrating an example of a processing procedure according to the present invention.
【図２】実施の形態で用いた画像入力系の概略図であ
る。FIG. 2 is a schematic diagram of an image input system used in the embodiment.
【図３】領域分割から２値化，最大連結領域算出，分散
計算までの処理の概念図である。FIG. 3 is a conceptual diagram of processing from area division to binarization, maximum connected area calculation, and variance calculation.
【図４】しきい値を変化させたときの差分画像の残差分
散の推移例を示す図である。FIG. 4 is a diagram illustrating an example of transition of residual variance of a difference image when a threshold value is changed.
【図５】しきい値を変化させたときの差分画像の残差分
散の，前しきい値のそれとの減少量の推移例（負は増
加）を示す図である。FIG. 5 is a diagram illustrating a transition example (negative increase) of a decrease amount of the residual variance of the difference image when the threshold value is changed from that of the previous threshold value.
【図６】実施例で用いた１０円硬貨の画像ならびに処理
対象のウィンドウを示す図である。FIG. 6 is a view showing an image of a 10-yen coin used in the embodiment and a window to be processed.
【図７】実施例で用いたキズのついた１０円硬貨の画像
ならびに位置合わせ後のウィンドウを示す図である。FIG. 7 is a diagram showing an image of a scratched 10-yen coin used in the embodiment and a window after alignment.
【図８】実施例で用いたキズのついた１０円硬貨の画像
に関して，しきい値を変化させたときの差分画像の残差
分散の，前しきい値のそれとの減少量の推移を示す図で
ある。FIG. 8 shows a change in the amount of decrease in the residual variance of the difference image with respect to that of the previous threshold value when the threshold value is changed for the image of the 10-yen coin with scratches used in the embodiment. FIG.
【図９】実施例で用いた図６および図７に示す２枚の画
像から欠陥検出を行った結果を示す図である。FIG. 9 is a diagram showing a result of performing defect detection from the two images shown in FIGS. 6 and 7 used in the example.
【図１０】従来手法の一つである２値化処理方式を示す
図である。FIG. 10 is a diagram showing a binarization processing method which is one of the conventional methods.
【図１１】従来手法の一つである濃度正規化相関方式を
示す図である。FIG. 11 is a diagram showing a density normalized correlation method, which is one of the conventional methods.
【図１２】良品／不良品の検査パターンの画像例を示す
図である。FIG. 12 is a diagram showing an example of an image of an inspection pattern of a good / defective product.
１０１ 第１の処理 １０２ 第２の処理 １０３ 第３の処理 １０４ 第４の処理 １０５ 第５の処理 １０６ 第６の処理 101 first processing 102 second processing 103 third processing 104 fourth processing 105 fifth processing 106 sixth processing
───────────────────────────────────────────────────── フロントページの続き (72)発明者 奥平 雅士 東京都新宿区西新宿三丁目19番２号 日本 電信電話株式会社内 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Masashi Okuhira 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Nippon Telegraph and Telephone Corporation
Claims (2)
査サンプルの検査画像とを用いて，両者の差分画像の分
散を計算し，残差の２値化しきい値を決定する処理によ
り，被検査物の表面に文字や模様が存在しても安定に良
品と不良品を判別する検査方法であって，被検査物のう
ち規準となる代表的な良品サンプルを選び，その画像を
撮影する第１の処理と，前記第１の処理と同じ撮像条件
下で，検査するサンプルの画像を取得する第２の処理
と，前記規準画像と前記検査画像との濃度差分画像を求
める第３の処理と，前記規準画像のエッジ抽出により領
域分割を行う第４の処理と，前記第４の処理で得られた
各領域毎に残差の各しきい値に対する残差分散の変化量
を計算し，その最大値を求める第５の処理と，前記第５
の処理で求めた変化量の最大値をもとに，当該検査サン
プルの良否判定を行う第６の処理とを有することを特徴
とする画像を用いた表面欠陥検出方法。1. A method for calculating a variance of a difference image between a reference image of a non-defective sample and an inspection image of an inspection sample, which is a reference, and determining a binarization threshold value of a residual. This is an inspection method for stably discriminating a non-defective product from a non-defective product even if a character or a pattern is present on the surface of the product. A second process of obtaining an image of a sample to be inspected under the same imaging conditions as the first process, a third process of obtaining a density difference image between the reference image and the inspection image, A fourth process of dividing the area by edge extraction of the reference image, and calculating a change amount of a residual variance with respect to each threshold value of the residual for each region obtained in the fourth process; A fifth process for obtaining a value;
And a sixth process of judging the quality of the inspection sample based on the maximum value of the amount of change obtained in the process of (i).
られた差分画像を用いて，前記第４の処理で求められた
各領域毎に，しきい値を変えた場合の当該差分画像の分
散の変化量を計算して行うことを特徴とする請求項１記
載の画像を用いた表面欠陥検出方法。2. The method according to claim 1, wherein the fifth process uses a difference image obtained in the third process and changes a threshold value for each region obtained in the fourth process. 2. The method according to claim 1, wherein a change amount of the variance of the difference image is calculated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP9016038A JPH10214327A (en) | 1997-01-30 | 1997-01-30 | Method of detecting surface defect using image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP9016038A JPH10214327A (en) | 1997-01-30 | 1997-01-30 | Method of detecting surface defect using image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| JPH10214327A true JPH10214327A (en) | 1998-08-11 |
Family
ID=11905422
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP9016038A Pending JPH10214327A (en) | 1997-01-30 | 1997-01-30 | Method of detecting surface defect using image |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JPH10214327A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1312469C (en) * | 2002-08-22 | 2007-04-25 | 丰田自动车株式会社 | Good-or-not decision device, decision programme, method and multi-variable statistical analyzing device |
| JP2007139621A (en) * | 2005-11-18 | 2007-06-07 | Omron Corp | Determination device, control program of determination device, and recording medium recording control program of determination device |
| WO2013111373A1 (en) * | 2012-01-27 | 2013-08-01 | オムロン株式会社 | Image examination method and image examination apparatus |
| JP2017013255A (en) * | 2015-06-29 | 2017-01-19 | 三星ダイヤモンド工業株式会社 | Break device |
-
1997
- 1997-01-30 JP JP9016038A patent/JPH10214327A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1312469C (en) * | 2002-08-22 | 2007-04-25 | 丰田自动车株式会社 | Good-or-not decision device, decision programme, method and multi-variable statistical analyzing device |
| JP2007139621A (en) * | 2005-11-18 | 2007-06-07 | Omron Corp | Determination device, control program of determination device, and recording medium recording control program of determination device |
| JP4645422B2 (en) * | 2005-11-18 | 2011-03-09 | オムロン株式会社 | Determination device, determination device control program, and recording medium recording determination device control program |
| WO2013111373A1 (en) * | 2012-01-27 | 2013-08-01 | オムロン株式会社 | Image examination method and image examination apparatus |
| JP2013156094A (en) * | 2012-01-27 | 2013-08-15 | Omron Corp | Image inspection method and image inspection device |
| JP2017013255A (en) * | 2015-06-29 | 2017-01-19 | 三星ダイヤモンド工業株式会社 | Break device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111179243A (en) | Small-size chip crack detection method and system based on computer vision | |
| US6879392B2 (en) | Method and apparatus for inspecting defects | |
| US6603877B1 (en) | Method of and apparatus for optical imaging inspection of multi-material objects and the like | |
| CN101464418B (en) | Flaw detection method and apparatus | |
| US20070053580A1 (en) | Image defect inspection apparatus, image defect inspection system, defect classifying apparatus, and image defect inspection method | |
| JP3279868B2 (en) | Defect inspection method and device for inspected pattern | |
| US7561751B2 (en) | Image processing method | |
| JP4230880B2 (en) | Defect inspection method | |
| JP2003083907A (en) | Defect inspecting method and its apparatus | |
| JPH08189904A (en) | Surface defect detector | |
| JPH06207909A (en) | Inspection apparatus for surface defect | |
| JP3788586B2 (en) | Pattern inspection apparatus and method | |
| JPH10214327A (en) | Method of detecting surface defect using image | |
| KR100940301B1 (en) | Inspecting method by using mark partitioning | |
| JP2001021332A (en) | Surface inspecting device and its method | |
| JP2003149160A (en) | Appearance inspection method and appearance inspection system | |
| US6888957B1 (en) | System and method for detecting blobs on an end surface of an optical fiber | |
| JP2001028059A (en) | Method and device for color unevenness inspection | |
| JP2000097869A (en) | Defect inspection method for pattern and its device | |
| JP2002168799A (en) | Pattern defect inspection method | |
| JP2000323541A (en) | Method and apparatus for inspecting appearance of object to be inspected | |
| JP2831273B2 (en) | Appearance inspection method | |
| CN117495846B (en) | Image detection method, device, electronic equipment and storage medium | |
| KR102546969B1 (en) | Particle and Plating Defect Inspection Method | |
| KR102655139B1 (en) | Camera module false feature classification method and system according to image segmentation inspection |