JPH1048152A - Method and apparatus for detecting fine linear defect - Google Patents
Method and apparatus for detecting fine linear defectInfo
- Publication number
- JPH1048152A JPH1048152A JP20202296A JP20202296A JPH1048152A JP H1048152 A JPH1048152 A JP H1048152A JP 20202296 A JP20202296 A JP 20202296A JP 20202296 A JP20202296 A JP 20202296A JP H1048152 A JPH1048152 A JP H1048152A
- Authority
- JP
- Japan
- Prior art keywords
- image
- defect
- linear defect
- linear
- segments
- 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.)
- Granted
Links
Landscapes
- Length Measuring Devices By Optical Means (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
Description
【０００１】[0001]
【発明の属する技術分野】本発明は、検出対象物を撮像
した画像より微細線状欠陥を検出する方法及び装置に関
するものである。BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for detecting a fine linear defect from an image of an object to be detected.
【０００２】[0002]
【従来の技術】溶接部や金属のヘアクラック或いはコン
クリート面のひび割れなどの微細線状欠陥を画像処理技
術で自動的に検出することが試みられている。2. Description of the Related Art Attempts have been made to automatically detect minute linear defects such as welds, metal hair cracks or cracks in concrete surfaces by image processing techniques.
【０００３】従来、検出対象物を撮像した画像より欠陥
を検出するには、その欠陥部が比較的大きくかつ濃淡差
の大きい場合には、２段階しきい値法が用いられてい
る。Conventionally, in order to detect a defect from an image of an object to be detected, a two-step threshold method has been used when the defect portion is relatively large and has a large difference in shading.
【０００４】２段階しきい値法では、先ず、低いしきい
値で２値化することによって、図１０（ａ）に示すよう
に一部分ではあるが、確実に欠陥である非常に暗い部分
１１ａのみを抽出し、ノイズを含まない画像１０ａを作
成する。次に、高いしきい値で２値化することによっ
て、図１０（ｂ）に示すように、ノイズ１２ｂを含むも
のの欠陥領域１１ｂを全て抽出した画像１０ｂを作成す
る。最後に、図１０（ｃ）に示すように、両画像１０
ａ，１０ｂを重ね合わせた画像１０ｃにおいて、高いし
きい値により２値化画像の中で、低いしきい値による２
値化画像の領域１１ｃを含む領域１３を残し、低いしき
い値による２値化画像の領域を含まない領域１２ｃをノ
イズとして除去する。結果的に図１０（ｄ）の画像１０
ｄに示すように高いしきい値で２値化された領域の中
で、低い濃度値を持つ領域を欠陥領域１１ｄとして抽出
する。In the two-step threshold value method, first, binarization is performed at a low threshold value, so that only a very dark portion 11a, which is a defect as shown in FIG. Is extracted, and an image 10a containing no noise is created. Next, as shown in FIG. 10B, an image 10b is created by extracting all the defective regions 11b, including the noise 12b, by binarizing the image with a high threshold value. Finally, as shown in FIG.
a and 10b are superimposed in the binarized image 10c with the high threshold value.
The region 13 including the region 11c of the binarized image is left, and the region 12c not including the region of the binarized image due to the low threshold is removed as noise. As a result, the image 10 shown in FIG.
As shown by d, an area having a low density value is extracted as a defective area 11d among the areas binarized with a high threshold value.
【０００５】[0005]
【発明が解決しようとする課題】しかしながら、ヘアク
ラックやひび割れなど抽出すべき領域が細く、かつ周囲
との濃淡差が小さい場合、２値化しきい値を低くする
と、領域となる線がバラバラに途切れ、高くすると多く
のノイズも抽出するため、２段階しきい値法を用いても
有効に動作しない。However, if the region to be extracted, such as a hair crack or a crack, is small and the difference in shading from the surroundings is small, if the binarization threshold is lowered, the lines forming the region are discontinuously broken. However, if it is set high, a lot of noise is also extracted, so that the two-step threshold method does not operate effectively.
【０００６】これを図９により説明すると、図９（ａ）
は、ノイズを含まない低いしきい値による２値化画像１
５ａ、図９（ｂ）は、全線が抽出できるしきい値で２値
化した画像１５ｂ、図９（ｃ）は、図９（ａ）のしきい
値よりも高いしきい値で２値化した画像１５ｃを示して
いる。This will be described with reference to FIG.
Is a binarized image 1 with a low threshold without noise
5A and FIG. 9B are binarized images 15b with a threshold from which all lines can be extracted, and FIG. 9C is binarized with a threshold higher than the threshold of FIG. 9A. An image 15c is shown.
【０００７】図９（ａ）の画像１５ａと図９（ｂ）の画
像１５ｂを重ね合わせて図９（ｄ）に示す画像１５ｄと
し、この画像１５ｄより、画像１５ａに含まれる線状欠
陥領域１６ａを含む画像１５ｂの領域１６ｂを抽出して
も図９（ｆ）の画像１５ｆに示すように、ノイズ１７ｂ
は除去できるものの、画像１５ｂの線状欠陥領域１６ｂ
のように、抽出すべき領域が、バラバラに途切れている
ため抽出結果の１５ｆで、一部の線状欠陥を抽出できな
い。An image 15a shown in FIG. 9A and an image 15b shown in FIG. 9B are superimposed to form an image 15d shown in FIG. 9D. From this image 15d, a linear defect area 16a included in the image 15a is obtained. Even if the region 16b of the image 15b containing the noise 17b is extracted, as shown in the image 15f of FIG.
Can be removed, but the linear defect area 16b of the image 15b
As described above, since the region to be extracted is discontinuous, some linear defects cannot be extracted with the extraction result 15f.
【０００８】また全ての線状欠陥を抽出しようとして、
図９（ｂ）の画像１５ｂと図９（ｃ）の画像１５ｃを重
ね合わせて、図９（ｅ）に示す画像１５ｅとし、この画
像１５ｅより画像１５ｂの中で画像１５ｃの線状領域１
６ｃを含む領域を抽出すると、図９（ｇ）の画像１６ｇ
に示すように欠陥領域１６ｇにはノイズ１７ｇも含んで
しまう問題が起こる。In order to extract all linear defects,
The image 15b shown in FIG. 9B and the image 15c shown in FIG. 9C are superimposed to form an image 15e shown in FIG. 9E.
When an area including the image 6g is extracted, an image 16g shown in FIG.
As shown in (1), a problem occurs in that the defective area 16g also includes noise 17g.
【０００９】このように、２段階しきい値により抽出す
るには、しきい値の設定によって、抽出すべき領域が抽
出できなかったり、抽出すべき領域を抽出できても、ノ
イズを多く抽出してしまう場合があり、最適なパラメー
タの設定は困難である。As described above, in order to perform the extraction using the two-step threshold value, depending on the setting of the threshold value, a region to be extracted cannot be extracted, or even if the region to be extracted can be extracted, a large amount of noise is extracted. In some cases, it is difficult to set optimal parameters.
【００１０】そこで、本発明の目的は、上記課題を解決
し、対象物の抽出ごとにパラメータの変更を行うことな
く線状の欠陥を精度良く抽出できる微細線状欠陥の検出
方法及びその装置を提供することにある。An object of the present invention is to solve the above-mentioned problems and to provide a method and an apparatus for detecting a fine linear defect capable of accurately extracting a linear defect without changing parameters every time an object is extracted. To provide.
【００１１】[0011]
【課題を解決するための手段】上記目的を達成するため
に請求項１の発明は、検出対象物を撮像した画像より微
細線状欠陥を検出する方法において、画像に正規化処理
を施した後、ノイズを除去できる低いしきい値で２値化
した領域を初期の核とし、多段階の濃度範囲で２値化し
た領域が、核領域と近接していれば核領域に反復的に統
合する多重多段階スライス法を用いて線状欠陥候補を検
出する微細線状欠陥の検出方法である。請求項２の発明
は、線状欠陥候補を抽出した後、この各線状欠陥候補を
直線セグメントに記述し、これら直線セグメントの方向
の連続性が局所的に保たれているセグメントを統合する
と共に、２つ以上のセグメントが大局的に直線或いは２
次曲線を構成しているかを判定し、２次曲線を構成する
場合には、構成要素となる各セグメントを結合し、局所
的欠陥領域の結合と大局的欠陥領域の結合より、微細線
状欠陥を抽出する請求項１記載の微細線状欠陥の検出方
法である。According to a first aspect of the present invention, there is provided a method for detecting a fine linear defect from an image obtained by capturing an object to be detected, after performing normalization processing on the image. A region binarized with a low threshold value capable of removing noise is used as an initial nucleus, and if the binarized region in a multi-step density range is close to the nucleus region, it is repeatedly integrated into the nucleus region. This is a fine linear defect detection method for detecting a linear defect candidate using a multiple multi-step slicing method. According to the second aspect of the present invention, after extracting the linear defect candidates, each of the linear defect candidates is described in a linear segment, and the segments in which the continuity of the directions of the linear segments is locally maintained are integrated, If two or more segments are globally straight or 2
It is determined whether or not a quadratic curve is formed, and when a quadratic curve is formed, the segments serving as constituent elements are combined, and a fine linear defect is formed by combining the local defect area and the global defect area. 2. The method for detecting a fine linear defect according to claim 1, wherein
【００１２】請求項３の発明は、検出対象物を撮像した
画像より微細線状欠陥を検出する装置において、検査対
象物を撮像する画像入力装置と、画像に正規化処理を施
した後、ノイズを除去できる低いしきい値で２値化した
領域を初期の核とし、多段階の濃度範囲で２値化した領
域が、核領域と近接していれば核領域に反復的に統合す
る多重多段階スライス法を用いて線状欠陥候補を検出す
る画像処理装置と、この各線状欠陥候補を直線セグメン
トに記述し、これら直線セグメントの方向の連続性が局
所的に保たれているセグメントを統合すると共に、２つ
以上のセグメントが大局的に直線或いは２次曲線を構成
しているかを判定し、２次曲線を構成する場合には、構
成要素となる各セグメントを結合し、局所的欠陥領域の
結合と大局的欠陥領域の結合より、微細線状欠陥を抽出
する判定装置とを備えた微細線状欠陥検出装置である。According to a third aspect of the present invention, there is provided an apparatus for detecting a fine linear defect from an image obtained by capturing an object to be detected. The area binarized with a low threshold value that can remove the nuclei is used as the initial nucleus, and if the binarized area in the multi-level concentration range is close to the nucleus area, the multiplexing is repeatedly integrated into the nucleus area An image processing apparatus that detects linear defect candidates using the step slice method, and describes each linear defect candidate in a straight line segment, and integrates segments in which the direction continuity of these straight line segments is locally maintained. At the same time, it is determined whether two or more segments form a straight line or a quadratic curve globally. If a quadratic curve is formed, the segments serving as constituent elements are combined to form a local defect area. Coupling and global defects Than the binding of the band, a fine line-shaped defect detection apparatus and a judging device for extracting a fine linear defects.
【００１３】上記構成によれば、画像の濃淡情報のみか
ら、良好な線状欠陥候補をノイズと区別して検出すると
共に、得られた線状欠陥候補を局所的／大局的探索で欠
陥部を結合することでさらに良好に線状欠陥を検出でき
る。According to the above arrangement, good linear defect candidates are detected and distinguished from noise only from the grayscale information of the image, and the obtained linear defect candidates are combined with defective portions by local / global search. By doing so, a linear defect can be detected more favorably.
【００１４】[0014]
【発明の実施の形態】以下、本発明の好適一実施の形態
を添付図面に基づいて詳述する。DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of the present invention will be described below in detail with reference to the accompanying drawings.
【００１５】先ず、微細線状欠陥の検出装置の装置構成
を図２により説明する。First, the configuration of a fine linear defect detecting device will be described with reference to FIG.
【００１６】被検査対象である対象物２０を画像入力装
置２１で撮像し、その取り込んだ画像を画像処理装置２
２で画像処理して判定装置２３にて抽出した微細線状欠
陥を判定する。An object 20 to be inspected is imaged by the image input device 21 and the captured image is processed by the image processing device 2.
In step 2, the image processing is performed, and the fine linear defect extracted by the determination device 23 is determined.
【００１７】この画像処理と判定の概略を図１により説
明すると、処理が開始２５され、画像入力２６がなされ
た後、画像正規化処理２７を行って、濃淡ヒストグラム
の平均値及び分散値を一定値に変換した後、多重多段階
スライス法による線状候補領域の抽出２８を行い、その
抽出した線状候補について連続している線かどうかの判
定を大局的探索による領域結合２９と局所的探索による
領域結合３０の双方で評価して、微細線状欠陥領域の抽
出３１を行った後、処理を終了３２する。The outline of the image processing and determination will be described with reference to FIG. 1. After the processing is started 25 and an image input 26 is performed, an image normalization processing 27 is performed to keep the average value and the variance value of the density histogram constant. After conversion to a value, a line candidate region is extracted 28 by the multiple multi-step slice method, and it is determined whether or not the extracted line candidate is a continuous line by a global search 29 and a local search by a global search. After performing the extraction 31 of the fine linear defect region by performing the evaluation in both of the region couplings 30, the process ends 32.
【００１８】以下、これらの処理を順に説明する。Hereinafter, these processes will be described in order.
【００１９】（１）画像の正規化処理 入力した画像は、その撮像状況によって濃淡にバラツキ
があるため、濃度の正規化処理を行う。(1) Image Normalization Processing The density of an input image varies depending on the imaging condition, and therefore, the density normalization processing is performed.
【００２０】正規化処理とは、濃度平均値ｍ、標準偏差
σの入力画像を、式(1) に従って、濃度平均ｍN 、標準
偏差σN の画像に変換する処理をいう。一般に、正規化
処理は、感度が異なるセンサで撮像された同種の画像間
の標準化を行う際に有効である。The normalization process is a process of converting an input image having a density average value m and a standard deviation σ into an image having a density average m N and a standard deviation σ N according to equation (1). Generally, the normalization processing is effective when standardizing the same type of images captured by sensors having different sensitivities.
【００２１】 ＩN （ｘ，ｙ）＝（σN ／σ）（Ｉ（ｘ，ｙ）−ｍ）＋ｍN …(1) ただし、 Ｉ（ｘ，ｙ） ： 入力原画像の点（ｘ，ｙ）における
濃淡値 ＩN （ｘ，ｙ）： 正規化画像の点（ｘ，ｙ）における
濃淡値 である。I N (x, y) = (σ N / σ) (I (x, y) −m) + m N (1) where I (x, y) is a point (x, y) of the input original image. Gray value IN (x, y) in y): Gray value at point (x, y) in the normalized image.
【００２２】図３（ａ）は、画像全体が暗く、コントラ
ストが低い画像の濃淡ヒストグラムを表している。この
ような画像を正規化することによって、図３（ｂ）に示
すように全体に明るさのバランスがとれ、コントラスト
の良い画像に変換される。すなわち、照明条件や入力装
置の設定条件によって、暗すぎたり、明るすぎたり、或
いはコントラストが低いような画像が入力されたとして
も、正規化することによって、欠陥領域の濃度範囲が略
同じ範囲に変換される。その結果、以後の処理における
パラメータを略固定することが可能となり、自動化を図
ることができる。FIG. 3A shows a density histogram of an image in which the entire image is dark and the contrast is low. By normalizing such an image, the brightness is balanced as a whole as shown in FIG. 3B, and the image is converted into an image with good contrast. That is, even if an image that is too dark, too bright, or has a low contrast is input depending on the lighting conditions and the setting conditions of the input device, the density range of the defect area is reduced to approximately the same range by normalization. Is converted. As a result, parameters in subsequent processing can be substantially fixed, and automation can be achieved.
【００２３】（２）多重多段階スライス法による線状候
補領域の抽出 多重多段階スライス法とは、図４（ａ）に示すような局
所的な濃淡変化を繰り返すヘアクラックを抽出し、か
つ、ノイズの包含を避ける方法である。(2) Extraction of Linear Candidate Regions by Multi-Stage Multi-Slice Method The multi-stage multi-slice method extracts hair cracks which repeat local density changes as shown in FIG. This is a method to avoid inclusion of noise.
【００２４】図４（ａ）に示すよう、正規化した画像中
の線状分の濃度分布４０に、線状領域４１とノイズ４２
があるとする。この場合、抽出される線状領域４１は、
濃度範囲Ｔ0 〜Ｔ4 に濃度が分布し、線状領域４１は、
Ｔ1 〜Ｔ4 に濃度が分布し、Ｔ4 以上は、対象物の表面
である。As shown in FIG. 4A, a linear region 41 and a noise 42 are included in a density distribution 40 of a linear portion in a normalized image.
There is In this case, the extracted linear region 41 is
The density is distributed in the density range T 0 to T 4 , and the linear area 41 is
The concentration is distributed from T 1 to T 4 , and T 4 or more is the surface of the object.
【００２５】さて先ず、正規化画像を濃淡モフォロジィ
処理によって、画像撮影時の濃淡ムラを除去する。すな
わち、正規化画像に対して所定のフィルタサイズで濃淡
モフォロジィ処理をし、その画像と正規化画像の濃淡値
の差分を求めることで、濃淡ムラの除去された画像が得
られる。この画像を、非常に低い濃度しきい値Ｔ1 で２
値化することによって、図４（ｂ）のに示すように、
断片的ではあるが確実に線状欠陥の一部４１ａであると
して判定される部分を抽出する。この抽出画像５０は、
図５（ａ）に示すように、ノイズを含まない、線状欠陥
５１ａの領域を示している。First, the normalized image is subjected to shading morphology processing to remove shading unevenness during image shooting. That is, a grayscale morphology process is performed on the normalized image with a predetermined filter size, and a difference between grayscale values of the image and the normalized image is obtained, whereby an image from which grayscale unevenness has been removed can be obtained. The image, at very low concentrations threshold T 1 2
By quantifying, as shown in FIG.
A portion that is fragmentary but is definitely determined to be the portion 41a of the linear defect is extracted. This extracted image 50
As shown in FIG. 5A, an area of the linear defect 51a that does not include noise is shown.
【００２６】次いで、図４（ｂ）のに示すように、そ
のしきい値Ｔ1 よりも少し高い濃度しきい値範囲Ｔ1 〜
Ｔ2 の部分４１ｂ，４２ｂを２値化により取り出し、こ
れがで抽出した部分４１ａと近接しているかどうかを
調べ、近接していれば、図４（ｂ）のに示すように結
合して連結部分４１ｃとする。この場合、ノイズの部分
４２ｂは、線状欠陥の部分４１ａに近接していないため
連結されず、同様にこの段階では、図４（ｂ）のの中
央の部分４１ｂ0 は連結されない。Next, as shown in FIG. 4B, the density threshold range T 1 to T 1 which is slightly higher than the threshold T 1 .
T 2 of the portion 41b, 42b were removed by the binarization, to determine whether it is a neighbor extracted portion 41a in, if the proximity, linking moiety attached as shown in FIG. 4 (b) 41c. In this case, the noise portion 42b is not connected because it is not close to the linear defect portion 41a. Similarly, at this stage, the central portion 41b0 in FIG. 4B is not connected.
【００２７】以下同様にして｛Ｔ1 ＋（ｉ−１）α＜Ｔ
＜Ｔ1 ＋ｉα ｉ＝１,,, Ｎ｝の部分を取りだし、それ
がこれまでに抽出した部分に近接するか否かを調べ、近
接している場合には線状欠陥として延長登録して、の
連結部分４１ｃに、のＴ2 〜Ｔ3 の部分４１ｄを連結
しての連結部分４１ｅとし、さらにこの連結部分４１
ｅにのＴ3 〜Ｔ4 の部分４１ｆを連結しての連結部
分４１ｇとする。Similarly, ΔT 1 + (i−1) α <T
<T 1 + iα i = 1 ,,, N} is taken out, and it is checked whether or not it is close to the part extracted so far, and if it is close, it is extended and registered as a linear defect. linked to a portion 41c, and the connecting portion 41e of connecting portion 41d of the T 2 through T 3 of, further the connecting portion 41 of the
the connecting portion 41g of the linked portions 41f of e Nino T 3 through T 4.
【００２８】この操作をＮ回（図４の例では３回）繰り
返して、濃度範囲Ｔ1 〜Ｔ4 までの連結を終えた後、さ
らに以上のことを、低いしきい値Ｔ1 からＮ回繰り返
し、これらを集計した部分を線状領域とする。このよう
に同じ処理ループをＮ回（図では３回）繰り返すことに
より、の部分４１ｄ0 は、の連結部分４１ｇと連続
しているため、の連結部分４１ｈとなり、さらに、
の中央の部分４１ｂ0 は、の連結部分４１ｈと連続し
ているため、の連結部分４１ｉとされて全ての線状領
域が抽出され、またノイズ４２ｂ，４２ｄ，４２ｆ分
は、の部分４１ａと連続していないため、抽出されな
い。This operation is repeated N times (three times in the example of FIG. 4), and after the connection in the concentration range T 1 to T 4 is completed, the above operation is further repeated N times from the low threshold value T 1. Repeatedly, a portion obtained by summing them is defined as a linear region. By repeating the same processing loop N times (three times in the figure), the portion 41d0 is connected to the connected portion 41g, and thus becomes the connected portion 41h.
Since the central portion 41b0 is continuous with the connecting portion 41h, it is regarded as the connecting portion 41i and all the linear regions are extracted, and the noises 42b, 42d, and 42f are continuous with the portion 41a. Not extracted because it is not.
【００２９】このように図５（ａ）に示した線状欠陥５
１ａの初期領域に含まれない連結領域が、１ループ目の
成長で、図５（ｂ）に示すように線状欠陥５１ｇが、次
いで２，３ループで、図５（ｃ），図５（ｄ）に示すよ
うに、順次成長した線状欠陥５１ｈ，５１ｉを抽出する
ことが可能となる。As described above, the linear defect 5 shown in FIG.
The connected region not included in the initial region 1a is the growth of the first loop, and as shown in FIG. 5B, a linear defect 51g is formed, followed by a few loops, as shown in FIG. 5C and FIG. As shown in d), it is possible to extract the linear defects 51h and 51i that have grown sequentially.
【００３０】このように多重多段階スライス法にて線状
候補を抽出することで、２段階しきい値法よりノイズを
抑え、精度良く線状欠陥を抽出することが可能となる。As described above, by extracting the linear candidates by the multiplex multi-step slicing method, it is possible to suppress noise and extract a linear defect with high accuracy compared to the two-step threshold method.
【００３１】次に、本発明をさらに説明する。Next, the present invention will be further described.
【００３２】多重多段階スライス法による連結処理にお
いて、線状領域に対し、大きく離れたノイズ分は、基本
的には検出されないが、線状欠陥５２に近接しているノ
イズ５３，５４については、図５（ｃ），図５（ｄ）に
示すようにが検出されてしまう。In the connection processing by the multiplex multi-step slicing method, noise components far apart from the linear region are basically not detected, but noises 53 and 54 close to the linear defect 52 are not detected. 5 (c) and 5 (d) are detected.
【００３３】そこで、以下の処理で、線状欠陥に近接す
るノイズ分を除去する。Therefore, a noise component close to the linear defect is removed by the following processing.
【００３４】（３）局所的及び大局的探索による欠陥領
域の結合 多重多段階スライス法を用いても、線状欠陥領域に近接
するノイズだけは除去することはできない。このような
ノイズは、抽出したい線状欠陥領域の濃度範囲、領域形
状、領域面積等が同じであるため、単なる画像処理手法
を組み合わせても除去することができない。(3) Coupling of Defect Regions by Local and Global Search Even by using the multiple multi-step slicing method, it is not possible to remove only the noise close to the linear defect region. Since such noise has the same density range, region shape, region area, and the like of the linear defect region to be extracted, it cannot be removed even by a combination of simple image processing methods.
【００３５】そこで、飛び飛びに得られた線状領域を直
線セグメントに記述し、幾何学的な連続性、及びフィッ
ト性を基に、同一欠陥部分として直線セグメントを結合
していく。これにより、１本の連続した線状欠陥を抽出
すると共に、線状欠陥に近接したノイズを区別すること
ができる。直線セグメントを結合するのに、局所的探索
及び大局的探索の２つのアプローチをとる。Accordingly, the linear regions obtained at intervals are described as straight line segments, and the straight line segments are combined as the same defective portion based on the geometric continuity and the fit. This makes it possible to extract one continuous linear defect and to discriminate noise close to the linear defect. Two approaches are used to combine the straight line segments: local search and global search.
【００３６】ａ．局所的探索による欠陥領域の結合 本処理は、図６に示すように、直線に記述されたセグメ
ント群Ｓの中から、局所的にみて近距離にあり、かつ方
向の連続性が保たれているものを統合していくものであ
る。すなわち、先ず比較的長い直線セグメントＳ0 をベ
ースセグメントと名付け、このベースセグメントＳ0 を
幅Ｗだけ太らせたリボン状領域８０を生成する。次に、
ベースセグメントＳ0 の端点から、距離ｒ以内に端点を
持ち、セグメント全体がリボン領域８０内に含まれる直
線セグメントの中で方向が同じセグメントＳ1 を統合
し、この操作を成長が止まるまで繰り返す。その際、リ
ボン領域８０内に含まれない点線で示したセグメントｓ
は連結候補に入れない。このようにリボン状領域８０に
ある連続性のあるセグメントＳを逐次的に探索すること
で、セグメントＳと方向が異なるノイズ分を除去できる
こととなる。A. Combining Defect Regions by Local Search In this process, as shown in FIG. 6, a segment group S described in a straight line is located at a short distance locally and continuity of direction is maintained. It integrates things. That is, first, a relatively long straight line segment S0 is named a base segment, and a ribbon-shaped region 80 is created by increasing the width of the base segment S0 by the width W. next,
The segment S1 having the end point within the distance r from the end point of the base segment S0 and having the same direction among the straight segments included in the ribbon region 80 as the whole segment is integrated, and this operation is repeated until the growth stops. At this time, the segment s indicated by a dotted line not included in the ribbon area 80
Is not included in the concatenation candidate. By successively searching for segments S having continuity in the ribbon-like region 80 in this manner, it is possible to remove a noise component having a direction different from that of the segment S.
【００３７】ｂ．大局的探索による欠陥領域の結合 図６に示した局所的探索では、周囲を見渡して隣接する
欠陥候補を結合してしまうため、大きく離れた欠陥候補
同士を結び付けることはできない。そこで、２つ以上の
直線セグメントが直線或いは２次曲線を構成しているか
否かを仮説検証によって判定し、大きく離れていても、
統合する処理を行う。これにより、マクロ的にみて、直
線或いは２次曲線で構成される直線群を検出することが
できる。B. Coupling of Defect Regions by Global Search In the local search shown in FIG. 6, adjacent defect candidates are combined while looking around, so that defect candidates that are far apart cannot be connected. Therefore, it is determined by hypothesis verification whether two or more straight line segments form a straight line or a quadratic curve.
Perform the integration process. This makes it possible to detect a group of straight lines constituted by straight lines or quadratic curves from a macro perspective.
【００３８】図７（ａ）は、各ベースセグメントＳ0 ご
とに、他の候補セグメントＳ'0とのペアで直線ｌを形成
できるか否かを判定し、可能である場合には最小二乗法
によって直線を生成する。FIG. 7A shows, for each base segment S0, whether it is possible to form a straight line l with a pair with another candidate segment S'0, and if possible, the least square method is used. Generate a straight line.
【００３９】次に、図７（ｂ）に示すように、この直線
ｌ上に存在する候補セグメントＳを全て検出する。そし
て、セグメントＳ群が構成する全直線の長さがあるしき
い値以上で、かつその全体の長さに占めるセグメントＳ
の比率が高ければ、図７（ｃ）に示すように線状欠陥７
１と判定する。その際、点線で示したセグメントｓのよ
うに、あまりにも大きく離れている場合は、分断して、
それぞれの直線群に対して仮説検証を行う。Next, as shown in FIG. 7 (b), all candidate segments S present on the straight line 1 are detected. Then, the length of all the straight lines that constitute the segment S group is equal to or greater than a certain threshold value and the segment S occupying the entire length thereof
Is high, the linear defect 7 as shown in FIG.
Judge as 1. At this time, if the distance is too large, such as the segment s indicated by the dotted line, the segment is
Hypothesis testing is performed for each straight line group.
【００４０】局所的及び大局的探索で線状欠陥を抽出す
る模式図を図８に示す。FIG. 8 is a schematic diagram for extracting a linear defect by local and global search.
【００４１】先ず、図８（ａ）は、上述した多重多段階
スライス法で抽出した線状欠陥候補を直線セグメント８
１に記述した画像８０ａを示す。直線セグメント８１か
ら図８（ｂ）に示すように、比較的長さが長いベースセ
グメントＳ0 を抽出し、図８（ｃ）では、局所的探索法
を用いて、近接し、かつ方向性が保たれるセグメント群
を抽出して線状欠陥８１ｃとしている。ただし、８１ｄ
のように大きく離れたセグメント同士は抽出することが
できない。図８（ｄ）では、大局的探索法を用いて大き
く離れたセグメント群を直線として線状欠陥８１ｄを抽
出している。２つの画像８０ｃ，８０ｄを合成すること
により、図８（ｅ）のように線状欠陥８１ｅのみの画像
８０ｅを得ることができる。First, FIG. 8A shows a linear defect candidate extracted by the above-described multiple multi-step slicing method.
1 shows an image 80a described in FIG. As shown in FIG. 8B, a base segment S0 having a relatively long length is extracted from the straight line segment 81, and in FIG. A group of segments to be dropped is extracted and set as a linear defect 81c. However, 81d
Cannot be extracted. In FIG. 8D, a linear defect 81d is extracted by using a global search method and setting a group of segments far apart as a straight line. By combining the two images 80c and 80d, an image 80e including only the linear defect 81e can be obtained as shown in FIG.
【００４２】[0042]
【発明の効果】以上要するに本発明によれば、画像の濃
淡情報のみから、良好な線状欠陥候補をノイズと区別し
て検出すると共に、得られた線状欠陥候補を局所的／大
局的探索で欠陥部を結合することでさらに良好に線状欠
陥を検出できる。In summary, according to the present invention, good linear defect candidates are distinguished and detected from noise only from the grayscale information of an image, and the obtained linear defect candidates are detected by local / global search. By combining the defective portions, a linear defect can be detected more favorably.
【図面の簡単な説明】[Brief description of the drawings]
【図１】本発明の微細線状欠陥を検出する処理フローを
説明する図である。FIG. 1 is a diagram illustrating a processing flow for detecting a fine linear defect according to the present invention.
【図２】本発明の微細線状欠陥を検出する装置構成を示
す図である。FIG. 2 is a diagram showing a configuration of an apparatus for detecting a fine linear defect according to the present invention.
【図３】本発明において、正規化処理を説明する図であ
る。FIG. 3 is a diagram illustrating a normalization process in the present invention.
【図４】本発明において、多重多段階スライス法で、線
状欠陥を抽出するための説明図である。FIG. 4 is an explanatory diagram for extracting a linear defect by a multiple multi-step slicing method in the present invention.
【図５】図４の多重多段階スライス法で、線状領域を連
結成長させた過程の画像を示す図である。FIG. 5 is a diagram showing an image of a process in which linear regions are connected and grown by the multiplex multi-step slicing method of FIG. 4;
【図６】本発明において、局所的探索による領域の結合
を説明する図である。FIG. 6 is a diagram for explaining connection of regions by a local search in the present invention.
【図７】本発明において、大局的探索による領域の結合
を説明する図である。FIG. 7 is a diagram for explaining connection of regions by a global search in the present invention.
【図８】本発明において、多重多段階スライス法で、抽
出した線状欠陥候補から線状欠陥を抽出するまでの模式
図を示す図である。FIG. 8 is a diagram showing a schematic diagram until a linear defect is extracted from extracted linear defect candidates by the multiplex multi-step slicing method in the present invention.
【図９】従来の２段階しきい値法による線状欠陥の抽出
を説明する図である。FIG. 9 is a diagram illustrating extraction of a linear defect by a conventional two-step threshold method.
【図１０】同じく従来の２段階しきい値法による線状欠
陥の抽出を説明する図である。FIG. 10 is a diagram illustrating extraction of a linear defect by a conventional two-step threshold method.
４１ａ，ｂ，ｄ，ｆ 線状欠陥 ４２ ノイズ Ｔ1 〜Ｔ4 多段階の濃度範囲 ４１ｉ 線状欠陥候補41a, b, d, f linear defect 42 Noise T 1 through T 4 multi-step concentration range 41i linear defect candidate
───────────────────────────────────────────────────── フロントページの続き (72)発明者 出川 定男 東京都江東区豊洲三丁目１番15号 石川島 播磨重工業株式会社東二テクニカルセンタ ー内 ──────────────────────────────────────────────────の Continued on the front page (72) Inventor Sadao Degawa 3-1-1, Toyosu, Koto-ku, Tokyo Ishikawajima Harima Heavy Industries, Ltd.
Claims (3)
欠陥を検出する方法において、画像に正規化処理を施し
た後、ノイズを除去できる低いしきい値で２値化した領
域を初期の核とし、多段階の濃度範囲で２値化した領域
が、核領域と近接していれば核領域に反復的に統合する
多重多段階スライス法を用いて線状欠陥候補を検出する
ことを特徴とする微細線状欠陥の検出方法。1. A method for detecting a fine linear defect from an image of an object to be detected, comprising: performing a normalization process on an image; If the binarized region in the multi-level density range is close to the nucleus region as a nucleus, linear defect candidates are detected using the multiple multi-step slicing method that repeatedly integrates into the nucleus region. Method for detecting fine linear defects.
欠陥候補を直線セグメントに記述し、これら直線セグメ
ントの方向の連続性が局所的に保たれているセグメント
を統合すると共に、２つ以上のセグメントが大局的に直
線或いは２次曲線を構成しているかを判定し、２次曲線
を構成する場合には、構成要素となる各セグメントを結
合し、局所的欠陥領域の結合と大局的欠陥領域の結合よ
り、微細線状欠陥を抽出する請求項１記載の微細線状欠
陥の検出方法。2. After extracting linear defect candidates, each of the linear defect candidates is described in a straight line segment, and segments in which the continuity of directions of the straight line segments is locally maintained are integrated. It is determined whether the above segments form a straight line or a quadratic curve globally. If a quadratic curve is formed, the segments serving as constituent elements are connected to each other, and the local defect area connection and the global The method for detecting a fine linear defect according to claim 1, wherein the fine linear defect is extracted by combining the defect areas.
欠陥を検出する装置において、検査対象物を撮像する画
像入力装置と、画像に正規化処理を施した後、ノイズを
除去できる低いしきい値で２値化した領域を初期の核と
し、多段階の濃度範囲で２値化した領域が、核領域と近
接していれば核領域に反復的に統合する多重多段階スラ
イス法を用いて線状欠陥候補を検出する画像処理装置
と、この各線状欠陥候補を直線セグメントに記述し、こ
れら直線セグメントの方向の連続性が局所的に保たれて
いるセグメントを統合すると共に、２つ以上のセグメン
トが大局的に直線或いは２次曲線を構成しているかを判
定し、２次曲線を構成する場合には、構成要素となる各
セグメントを結合し、局所的欠陥領域の結合と大局的欠
陥領域の結合より、微細線状欠陥を抽出する判定装置と
を備えたことを特徴とする微細線状欠陥検出装置。3. An apparatus for detecting a fine linear defect from an image obtained by capturing an object to be detected, comprising: an image input device for capturing an image of an object to be inspected; Using the area binarized by the threshold value as the initial nucleus, and using the multiple multi-step slicing method that repeatedly integrates the binarized area into the nucleus area if the binarized area in the multi-level concentration range is close to the nucleus area An image processing apparatus for detecting linear defect candidates by using a linear segment, describing each linear defect candidate in a linear segment, integrating segments in which the continuity of the directions of the linear segments is locally maintained, and combining two or more linear defect candidates. It is determined whether or not a segment of the image globally forms a straight line or a quadratic curve. When a quadratic curve is formed, the segments serving as constituent elements are combined to combine the local defect area and the global defect. Finer than combining regions A fine linear defect detection device, comprising: a determination device for extracting a fine linear defect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP20202296A JP3635795B2 (en) | 1996-07-31 | 1996-07-31 | Method and apparatus for detecting fine line defects |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP20202296A JP3635795B2 (en) | 1996-07-31 | 1996-07-31 | Method and apparatus for detecting fine line defects |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| JPH1048152A true JPH1048152A (en) | 1998-02-20 |
| JP3635795B2 JP3635795B2 (en) | 2005-04-06 |
Family
ID=16450634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP20202296A Expired - Fee Related JP3635795B2 (en) | 1996-07-31 | 1996-07-31 | Method and apparatus for detecting fine line defects |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JP3635795B2 (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005227055A (en) * | 2004-02-12 | 2005-08-25 | Tokyo Metropolitan Sewerage Service Corp | Method for extracting crack in image in pipe culvert |
| JP2006024149A (en) * | 2004-07-09 | 2006-01-26 | Foundation For The Promotion Of Industrial Science | On-image moving object recognition method and device |
| JP2007078540A (en) * | 2005-09-14 | 2007-03-29 | Tdk Corp | Visual inspection method and visual inspection device |
| US7433508B2 (en) * | 2003-03-12 | 2008-10-07 | Hitachi High-Technologies Corporation | Pattern inspection method and its apparatus |
| JP2012098045A (en) * | 2010-10-29 | 2012-05-24 | Mitsubishi Electric Corp | Crack detector and crack detection program |
| US8391585B2 (en) | 2006-12-28 | 2013-03-05 | Sharp Kabushiki Kaisha | Defect detecting device, defect detecting method, image sensor device, image sensor module, defect detecting program, and computer-readable recording medium |
| JP2017072517A (en) * | 2015-10-08 | 2017-04-13 | 西日本高速道路エンジニアリング四国株式会社 | Detection method of surface crack of concrete |
| WO2019009214A1 (en) * | 2017-07-07 | 2019-01-10 | キヤノン株式会社 | Image processing device, image processing method and program for detecting deformation from image |
| WO2019138525A1 (en) * | 2018-01-12 | 2019-07-18 | 株式会社日立ハイテクノロジーズ | Defect inspection device and defect information display device |
| JP2020160616A (en) * | 2019-03-25 | 2020-10-01 | ブラザー工業株式会社 | Generation device and computer program and generation method |
| CN112598659A (en) * | 2020-12-29 | 2021-04-02 | 凌云光技术股份有限公司 | Method for detecting linear defects of printed product |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017130699A1 (en) * | 2016-01-26 | 2017-08-03 | 富士フイルム株式会社 | Crack information detection device, crack information detection method, and crack information detection program |
-
1996
- 1996-07-31 JP JP20202296A patent/JP3635795B2/en not_active Expired - Fee Related
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7433508B2 (en) * | 2003-03-12 | 2008-10-07 | Hitachi High-Technologies Corporation | Pattern inspection method and its apparatus |
| US7949178B2 (en) | 2003-03-12 | 2011-05-24 | Hitachi High-Technologies Corporation | Pattern inspection method and its apparatus |
| JP2005227055A (en) * | 2004-02-12 | 2005-08-25 | Tokyo Metropolitan Sewerage Service Corp | Method for extracting crack in image in pipe culvert |
| JP2006024149A (en) * | 2004-07-09 | 2006-01-26 | Foundation For The Promotion Of Industrial Science | On-image moving object recognition method and device |
| JP4628712B2 (en) * | 2004-07-09 | 2011-02-09 | 財団法人生産技術研究奨励会 | Method and apparatus for recognizing moving object on image |
| JP2007078540A (en) * | 2005-09-14 | 2007-03-29 | Tdk Corp | Visual inspection method and visual inspection device |
| US8391585B2 (en) | 2006-12-28 | 2013-03-05 | Sharp Kabushiki Kaisha | Defect detecting device, defect detecting method, image sensor device, image sensor module, defect detecting program, and computer-readable recording medium |
| JP2012098045A (en) * | 2010-10-29 | 2012-05-24 | Mitsubishi Electric Corp | Crack detector and crack detection program |
| JP2017072517A (en) * | 2015-10-08 | 2017-04-13 | 西日本高速道路エンジニアリング四国株式会社 | Detection method of surface crack of concrete |
| WO2019009214A1 (en) * | 2017-07-07 | 2019-01-10 | キヤノン株式会社 | Image processing device, image processing method and program for detecting deformation from image |
| CN110892255A (en) * | 2017-07-07 | 2020-03-17 | 佳能株式会社 | Image processing apparatus, image processing method, and program for detecting defects from an image |
| JPWO2019009214A1 (en) * | 2017-07-07 | 2020-05-21 | キヤノン株式会社 | Image processing apparatus, image processing method, and program for detecting deformation from image |
| US11443500B2 (en) | 2017-07-07 | 2022-09-13 | Canon Kabushiki Kaisha | Image processing apparatus, image processing method, and program for detecting defect from image |
| CN110892255B (en) * | 2017-07-07 | 2024-03-01 | 佳能株式会社 | Image processing apparatus, method and storage medium for detecting defect from image |
| WO2019138525A1 (en) * | 2018-01-12 | 2019-07-18 | 株式会社日立ハイテクノロジーズ | Defect inspection device and defect information display device |
| JP2020160616A (en) * | 2019-03-25 | 2020-10-01 | ブラザー工業株式会社 | Generation device and computer program and generation method |
| CN112598659A (en) * | 2020-12-29 | 2021-04-02 | 凌云光技术股份有限公司 | Method for detecting linear defects of printed product |
| CN112598659B (en) * | 2020-12-29 | 2024-02-27 | 凌云光技术股份有限公司 | Printed linear defect detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| JP3635795B2 (en) | 2005-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP3635795B2 (en) | Method and apparatus for detecting fine line defects | |
| JP4189061B2 (en) | Crack detection method for concrete surface | |
| CN111638218A (en) | Method for detecting surface defects of coating | |
| JPH09119900A (en) | Method for automatically evaluating concrete defect | |
| CN108510513A (en) | A kind of PCB image circle detection method based on PCA and segmentation RHT | |
| JPH06207909A (en) | Inspection apparatus for surface defect | |
| JPH08305795A (en) | Character recognizing method | |
| JP2000003436A (en) | Device and method for recognizing isar picture | |
| JP2005031901A (en) | Coin identifying device and method | |
| EP0974830A3 (en) | Apparatus and method for detecting low-contrast blemishes | |
| JP2946095B2 (en) | Line segment extraction device | |
| JPH02242382A (en) | Defect checking method | |
| JP4189060B2 (en) | Extraction method of free lime on concrete surface | |
| JPH0217832B2 (en) | ||
| JP2711649B2 (en) | Surface scratch detection method for inspection object | |
| JP2831273B2 (en) | Appearance inspection method | |
| JPH041867A (en) | Picture quality evaluating method | |
| CN113344955B (en) | System and method for accurately detecting slope by fuzzy edge | |
| CN112767414B (en) | Image segmentation method and automatic detection method for micro-nano optical element | |
| JP2004125629A (en) | Defect detection apparatus | |
| JPH10214327A (en) | Method of detecting surface defect using image | |
| JP3339128B2 (en) | Defect inspection method and apparatus used for its implementation | |
| JPH0797410B2 (en) | Image processing method | |
| JP3409695B2 (en) | Surface defect inspection method | |
| JP3189604B2 (en) | Inspection method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A977 | Report on retrieval |
Free format text: JAPANESE INTERMEDIATE CODE: A971007 Effective date: 20040830 |
|
| A131 | Notification of reasons for refusal |
Effective date: 20040907 Free format text: JAPANESE INTERMEDIATE CODE: A131 |
|
| A521 | Written amendment |
Effective date: 20041105 Free format text: JAPANESE INTERMEDIATE CODE: A523 |
|
| TRDD | Decision of grant or rejection written | ||
| A01 | Written decision to grant a patent or to grant a registration (utility model) |
Effective date: 20041214 Free format text: JAPANESE INTERMEDIATE CODE: A01 |
|
| A61 | First payment of annual fees (during grant procedure) |
Free format text: JAPANESE INTERMEDIATE CODE: A61 Effective date: 20041227 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20080114 Year of fee payment: 3 |
|
| S531 | Written request for registration of change of domicile |
Free format text: JAPANESE INTERMEDIATE CODE: R313531 |
|
| S533 | Written request for registration of change of name |
Free format text: JAPANESE INTERMEDIATE CODE: R313533 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20080114 Year of fee payment: 3 |
|
| R350 | Written notification of registration of transfer |
Free format text: JAPANESE INTERMEDIATE CODE: R350 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20080114 Year of fee payment: 3 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Year of fee payment: 4 Free format text: PAYMENT UNTIL: 20090114 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Year of fee payment: 5 Free format text: PAYMENT UNTIL: 20100114 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20100114 Year of fee payment: 5 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Year of fee payment: 6 Free format text: PAYMENT UNTIL: 20110114 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20110114 Year of fee payment: 6 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Year of fee payment: 7 Free format text: PAYMENT UNTIL: 20120114 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20120114 Year of fee payment: 7 |
|
| FPAY | Renewal fee payment (prs date is renewal date of database) |
Year of fee payment: 8 Free format text: PAYMENT UNTIL: 20130114 |
|
| LAPS | Cancellation because of no payment of annual fees |