JP3598196B2

JP3598196B2 - Periodic pattern defect inspection equipment

Info

Publication number: JP3598196B2
Application number: JP5734597A
Authority: JP
Inventors: 憲一川路; 真幸川島; 晃正森田
Original assignee: Nireco Corp; Olympus Corp
Current assignee: Nireco Corp; Olympus Corp
Priority date: 1997-03-12
Filing date: 1997-03-12
Publication date: 2004-12-08
Anticipated expiration: 2017-03-12
Also published as: JPH10253332A

Description

【０００１】
【発明の属する技術分野】
本発明は、液晶パネルやそのカラーフィルタ、シャドウマスク等の周期的パターンを有する検査対象における傷や、ピンホール、黒点、ゴミ、パターンの欠け等の欠陥を検査する周期パターン検査装置に関する。
【０００２】
【従来の技術】
従来、周期的パターンの欠陥を検出する方法として特開平４−３１６３４６号、特開平６−１２９９９２号、特開平７−１５９３４３号公報等があり、これらは周期的パターンをラインセンサで走査し、元の画像とこれをずらした画像とを比較し、相違した場合、その相違部分を欠陥として検出する技術を開示している。
【０００３】
【発明が解決しようとする課題】
欠陥には傷、ピンホール、汚れ、ゴミ、パターンの欠け等があり、これらは画像上、黒くまたは白く表れる。このため、欠陥が黒か白かがわかれば、欠陥の原因もかなり推定できる。また欠陥は１周期長以内の場合が多いが、１周期長を超えるものもある。しかし、従来の技術では白と黒の欠陥を同時刻で区別して検出することができなかった。また１周期長を超える欠陥と、その欠陥が白か黒かの区別を検出することはできなかった。
【０００４】
本発明は上述の問題点に鑑みてなされたもので、周期的パターンに生じる欠陥を白欠陥、黒欠陥に区別して検出する検査装置を提供することを目的とする。さらに１周期長を超える大欠陥についても、白欠陥、黒欠陥を区別して検出する装置を提供することを目的とする。
【０００７】
【課題を解決するための手段】
請求項１の発明では、一定の周期でパターンが反復する検査対象物の欠陥を検査する装置であって、前記パターンを線型走査してイメージセンサで読み取り多値画像を得る画像入力手段と、前記多値画像をＡとし、このＡより１周期遅延した多値画像をＢとし、前記Ａより２周期遅延した多値画像をＣとし、各多値画像の同時刻における差分であって、ＡとＢとの差分Ａ−Ｂ，ＢとＡとの差分Ｂ−Ａ，ＢとＣとの差分Ｂ−Ｃ，ＣとＢとの差分Ｃ−Ｂを演算する差分演算手段と、前記各差分Ａ−Ｂ，Ｂ−Ａ，Ｂ−Ｃ，Ｃ−Ｂが所定のしきい値より大きいときは１，未満のときは負の場合も含めて０とする２値化を行う２値化手段と、この２値化手段による各差分の２値化値を対応するアルファベットの小文字によりａ−ｂ，ｂ−ａ，ｂ−ｃ，ｃ−ｂと表わした場合、ｂ−ａとｂ−ｃの論理積を演算する第１論理積演算手段と、ａ−ｂとｃ−ｂとの論理積を演算する第２論理積演算手段と、を備え、前記多値画像はＭからＮ（Ｍ，Ｎ共正数でＭ＜Ｎ）の多値で表されている。
【０００８】
多値画像について、最暗レベルは０、最明レベルはＮとした場合、白欠陥は周期長だけ離れた両隣の濃度値と比べてあるしきい値（Ｌ１）よりも高い濃度レベルを持っており、黒欠陥は他のしきい値（Ｌ２，Ｌ２＜ＬＩ）よりも低い濃度レベルを持っている。また欠陥がない場合は両隣とほぼ同様の濃度レベルを持っている。いま、１周期長内に欠陥があって、それが白欠陥であるとする。欠陥の発生した周期の画像Ｂを中心として同時刻におけるＡとの差分Ｂ−Ａ，Ｃとの差分Ｂ−Ｃをとると、差分Ｂ−Ａ，差分Ｂ−Ｃ，は正の値になる。この差分Ｂ−Ａ，差分Ｂ−Ｃを所定のしきい値Ｌ１と比較し、これより大きいとき１，それ以外のとき０として２値化する。しきい値として適切に欠陥を表す値とすることにより、確実に欠陥を取り出すことができる。そして２値化差分ｂ−ａ，ｂ−ｃが共に１のときは白欠陥と判定することができる。つまり、ある周期とこの１つ前の周期の比較、ある周期とこの１つ後の周期の比較をすることにより、ある周期内に欠陥が有るか否か、その欠陥は白欠陥か黒欠陥か、その周期内の欠陥かが分かる。
【０００９】
また１周期長以下の黒い欠陥が存在するとき、欠陥の発生した周期の画素Ｂを中心として同時刻におけるＡとの差分Ａ−Ｂ，Ｃとの差分Ｃ−Ｂをとる。差分Ａ−Ｂ，差分Ｃ−Ｂは正の値となる。この差分Ａ−Ｂ，差分Ｃ−Ｂを所定のしきい値Ｌ２と比較し、これより大きい時は１、それ以外は０として２値化する。しきい値として適切に欠陥を表す値とすることにより、確実に欠陥を取り出すことができる。この２値化差分ａ−ｂ，ｃ−ｂが共に１のときは黒い欠陥と判定することができる。これにより欠陥の検出のみならず、白い欠陥か黒い欠陥かも判定することができる。
【００１０】
請求項４の発明では、前記多値画像で、最暗レベルを０、最明レベルをＮ（正数）で多値化表示しているときは、前記第１論理積演算手段の出力が１ときは１周期長以下の白欠陥、前記第２論理積演算手段の出力が１のときは１周期長以下の黒欠陥と判定する。
【００１１】
多値画像で０を最暗レベル、Ｎを最明レベルとして０よりＮまでで多値を表示している場合、１周期長両隣の濃度と比較して、しきい値Ｌ１より高いレベルにあるときは白い欠陥である。故に請求項２の場合で説明した２値化差分ｂ−ａ，ｂ−ｃが共に１の場合に相当し、これは第１論理積演算手段の出力が１の場合である。また、１周期長両隣の濃度と比較して、しきい値Ｌ２（Ｌ１＞Ｌ２）より低いレベルにあるときは黒い欠陥である。故に請求項２の場合で説明した２値化差分ａ−ｂ，ｃ−ｂが共に１の場合に相当し、これは第２論理積演算手段の出力が１の場合である。
【００１２】
請求項５の発明では、前記多値画像で最明レベルを０、最暗レベルをＮ（正数）で多値化表示している時は、前記第１論理積演算手段の出力が１のときは１周期長以下の黒欠陥、前記第２論理積演算手段の出力が１のときは１周期長以下の白欠陥と判定する。
【００１３】
本発明は多値化の設定を請求項４の場合と反対にしたもので、０を最明レベル、Ｎを最暗レベルとして多値化を表した場合である。１周期長離れた両隣よりもしきい値Ｌ２より小さな値のときは白い欠陥であり、１周期長離れた両隣よりもしきい値Ｌ１（Ｌ１＞Ｌ２）より大きな値のときは黒い欠陥である。黒の欠陥の場合、請求項２の場合で説明した２値化差分ｂ−ａ，ｂ−ｃが共に１の場合に相当し、これは第１論理積演算手段の出力が１の場合である。また、１周期長両隣の濃度と比較して、しきい値Ｌ２より小さな値のときは白い欠陥である。白い欠陥の場合、請求項１の場合で説明した２値化差分ａ−ｂ，ｃ−ｂが共に１の場合に相当し、これは第２論理積演算手段の出力が１の場合である。
【００１４】
請求項２の発明では、一定の周期でパターンが反復する検査対象物の欠陥を検査する装置であって、前記パターンを線形走査してイメージセンサで読み取り多値画像を得る画像入力手段と、前記多値画像をＡとし、このＡより１周期長遅延した多値画像をＢとし、前記Ａより２周期長遅延した多値画像をＣとし、各多値画像の同時刻における差分であって、ＡとＢとの差分Ａ−Ｂ，ＢとＡとの差分Ｂ−Ａ，ＢとＣとの差分Ｂ−Ｃ，ＣとＢとの差分Ｃ−Ｂを演算する差分演算手段と、前記各差分Ａ−Ｂ，Ｂ−Ａ，Ｂ−Ｃ，Ｃ−Ｂが所定のしきい値より大きいときは１，未満のときは負の場合も含めて０とする２値化を行う２値化手段と、この２値化手段による各差分の２値化値を対応するアルファベットの小文字によりａ−ｂ，ｂ−ａ，ｂ−ｃ，ｃ−ｂと表わした場合、ｂ−ａの反転値と、ｂ−ｃと、ｃ−ｂの１周期遅延値の反転値との論理積を得る第３論理積演算手段と、ａ−ｂの反転値と、ｃ−ｂと、ｂ−ｃの１周期遅延値の反転値との論理積を得る第４論理積演算手段とを、備える。
【００１５】
白い欠陥で１周期長を超える長欠陥の特徴を図８を用いて説明する。まず長欠陥の始まる１周期目について、Ｐの期間では、ＢとＡが同じレベルで、ＣはＡとＢより低いレベルである。さらに、Ｐ−１の期間ではＢとＣは同じレベルである。これは２値化差分ｂ−ａが０，ｂ−ｃが１，ｃ−ｂの１周期遅延値が０であることを意味する。故にｂ−ａの反転値と、ｂ−ｃの値と、ｃ−ｂの１周期遅延値の反転値とが全て１になるので、第３論理積演算手段の出力が１となる。
【００１６】
また長欠陥の終わる周期について、Ｑの期間ではＡとＢは同じレベルで、ＣはＡ，Ｂより高いレベルである。さらにＱ−１期間ではＣとＢは同じレベルである。これは請求項５の発明で２値化差分ａ−ｂが０、ｃ−ｂが１、ｂ−ｃの１周期遅延値が０であることを意味する。故にａ−ｂの反転値と、ｂ−ｃの値と、ｂ−ｃの１周期遅延値の反転値とが全て１になるので第４論理積演算手段の出力は１となる。
【００１７】
黒い欠陥で１周期長を超える長欠陥の特徴を図９を用いて説明する。長欠陥の始まる１周期目について、Ｐの期間では、ＢとＡが同じレベルで、ＣはＢ，Ａより高いレベルである。さらにＰ−１の期間ではＢとＣは同じレベルである。これは請求項５の発明で２値化差分ａ−ｂが０、ｃ−ｂが１、ｂ−ｃの１周期遅延値が０であることを意味する。故にａ−ｂの反転値と、ｂ−ｃの値と、ｂ−ｃの１周期遅延値の反転値とが全て１となるので、第４論理積演算手段の出力は１となる。
【００１８】
また長欠陥の終わる周期について、Ｑの期間では、ＢとＡが同じレベルであり、ＣはＢ，Ａよりも低レベルである。さらにＱ−１の期間ではＢとＣは同じレベルである。これは請求項５の発明で２値化差分ｂ−ａが０、ｂ−ｃが１、ｃ−ｂの１周期遅延値が０であることを意味する。故にｂ−ａの反転値と、ｂ−ｃの値と、ｃ−ｂの１周期遅延値の反転値とが全て１になるので、第３論理積演算手段の出力が１となる。
【００１９】
以上のように白い欠陥で１周期長を超える長欠陥が存在するとまず第３論理積演算手段の出力が１となり、続いて第４論理積演算手段の出力が１となる。また黒い欠陥で１周期長を超える長欠陥が存在すると、まず第４論理積演算手段の出力が１となり、続いて第３論理積演算手段の出力が１となる。これにより１周期長を超える長欠陥をそれが白い欠陥か、黒い欠陥かを区別して検出することができる。
【００２０】
請求項６の発明では、前記多値画像で最暗レベルを０、最明レベルをＮ（正数）で多値化表示しているときは、前記第３論理積演算手段の出力が１となり、これに続いて前記第４論理積演算手段の出力が１となったとき、１周期長を超える白欠陥と判定し、前記第４論理積演算手段の出力が１となり、これに続いて前記第３論理積演算手段の出力が１となったとき１周期長を超える黒欠陥と判定する。
【００２１】
多値化表示を０が最暗レベル、Ｎが最明レベルとした場合、請求項２の発明によれば、白欠陥で１周期長を超える長欠陥が存在する場合、第３論理積演算手段の出力が１となり、続いて第４論理積演算手段の出力が１となる。また黒欠陥で１周期長を超える長欠陥が存在する場合、第４論理積演算手段の出力が１となり、続いて第３論理積演算手段の出力が１となる。
【００２２】
請求項７の発明では、前記多値画像で最明レベルを０、最暗レベルをＮ（正数）で多値化表示している時は、前記第３論理積演算手段の出力が１となり、これに続いて前記第４論理積演算手段の出力が１となったとき、１周期長を超える黒欠陥と判定し、前記第４論理積演算手段の出力が１となり、これに続いて前記第３論理積演算手段の出力が１となったとき、１周期長を超える白欠陥と判定する。
【００２３】
多値化表示を０が最明レベル、Ｎが最暗レベルとした場合、請求項２の発明によれば、黒欠陥で１周期長を超える長欠陥が存在すると、まず第３論理積演算手段の出力が１となり、続いて第４論理積演算手段の出力が１となる。また白欠陥で１周期長を超える長欠陥が存在すると、まず第４論理積演算手段の出力が１となり、続いて第３論理積演算手段の出力が１となる。
【００２４】
請求項３の発明では、一定の周期でパターンが反復する検査対象物の欠陥を検査する装置であって、前記パターンを線形走査してイメージセンサで読み取り多値画像を得る画像入力手段と、前記多値画像をＡとし、このＡより１周期遅延した多値画像をＢとし、前記Ａより２周期遅延した多値画像をＣとし、各多値画像の同時刻における差分であって、ＡとＢとの差分Ａ−Ｂ，ＢとＡとの差分Ｂ−Ａ，ＢとＣとの差分Ｂ−Ｃ，ＣとＢとの差分Ｃ−Ｂを演算する差分演算手段と、前記各差分Ａ−Ｂ，Ｂ−Ａ，Ｂ−Ｃ，Ｃ−Ｂが所定のしきい値より大きいときは１，未満のときは負の場合も含めて０とする２値化を行う２値化手段と、この２値化手段による各差分の２値化値を対応するアルファベットの小文字によりａ−ｂ，ｂ−ａ，ｂ−ｃ，ｃ−ｂと表わした場合、ｂ−ａとｂ−ｃの論理積を演算する第１論理積演算手段と、ａ−ｂとｃ−ｂとの論理積を演算する第２論理積演算手段と、ｂ−ａの反転値と、ｂ−ｃと、ｃ−ｂの１周期遅延値の反転値との論理積を得る第３論理積演算手段と、ａ−ｂの反転値と、ｃ−ｂと、ｂ−ｃの１周期遅延値の反転値との論理積を得る第４論理積演算手段とを、備える。
【００２５】
本発明は、請求項１の発明と請求項２の発明を連結したもので、１周期長以内の欠陥を白欠陥と黒欠陥を区別して検出できると共に、１周期長を超える長欠陥を白欠陥と黒欠陥とに区別して検出できる。
【００２６】
【発明の実施の形態】
以下、本発明の実施の形態について図面を参照して説明する。
図１は本実施の形態の構成を示すブロック図である。検査対象物１は矢印方向に移動し、この移動方向と直交する方向に周期的パターンを有している。ラインセンサカメラ２は内蔵するラインセンサの長さ方向（センサ素子の並び方向）を検査対象物１の移動方向と直交するように配置し、移動する周期パターンを持った画像を取り込む。Ａ／Ｄ変換器３はラインセンサカメラ２からのビデオ信号の濃淡画像をアナログ／デジタル変換して８ビット（２５６階調）の多値画像に変換し、この多値画像をＡとする。１周期長遅延回路４はＡを１周期長遅延させこの出力をＢとする。２周期長遅延回路５はＡを２周期長遅延させこの出力をＣとする。欠陥判定回路６はＡ，Ｂ，Ｃを入力し周期的なパターンに発生した欠陥を検出する。欠陥は傷、ピンホール、汚れ、ゴミ、パターンの欠け等で、これらは多値画像で０を最暗レベル、２５５を最明レベルとしその間を順次最暗レベルから最明レベルに変化させる階調とする場合、１周期長離れた両隣に比べてあるしきい値Ｌ１より高いレベルにあるとき白欠陥、他のしきい値Ｌ２（Ｌ１＞Ｌ２）より低いレベルにあるとき黒欠陥とする。以下の説明は０を最暗レベル、２５５を最明レベルとして説明する。欠陥判定回路６は欠陥をこの白欠陥または黒欠陥に分類して２値化画像として出力する。
【００２７】
図２は検査対象物１としての液晶パネルの一例を示す。これはリード線１１とゲート線１２が格子状に配線され、交点にＴＦＴ（ＴｈｉｎＦｉｌｍＴｒａｎｓｉｓｔｅｒ）が配置されている。格子間隔は横方向、縦方向共規則的に配置され、ラインセンサカメラ２はゲート線方向（横方向）に走査してゆき、リード線１１の間隔を１周期長とする画像Ａが得られる。欠陥は１周期長以内の白欠陥１４、黒欠陥１５があり、また１周期長を超える白長欠陥１６、黒長欠陥１７がある。
【００２８】
ラインセンサカメラ２のラインセンサにより撮像される画像は、走査方向（図２で横方向、ゲート線１２方向）に１列に並んだ画素列より構成される。ラインセンサは検出精度を高めるように構成されている。つまりパターン検査に必要な分解能が得られるように１周期中の画素数を定めている。ラインセンサのセンサ素子の大きさが撮像される画像の１つの画素の大きさとなっている。この画素の大きさは例えば１０μｍ程度で、１周期長が１００μｍの場合、ほぼ１０個ぐらいの画素で１周期長を表している。しかし、周期長と、センサ素子の大きさはそれぞれ独立のものであるので、１周期長を表す画素の数が整数個とならない場合が普通である。例えば１周期長が９．４画素などで表される。
【００２９】
このように１周期長が整数の画素数で表されない場合、１周期遅延画像Ｂ、２周期遅延画像Ｃの作成方法を説明する。例えば、１周期長が９．４画素の場合、１周期を構成する各画素を１周期長（つまり９．４画素）遅延させるには、次に示す（１）式により線形補間して、まず０．４画素遅延した場所のデータを算出する。さらにそのデータを９画素（整数部）遅らせることで、９．４画素遅延させることができる。
【００３０】
線形補間式：Ｑｉ＝（ｍＰ_ｉ＋ｎＰ_ｉ＋１）／１０ ……（１）
ここでｍ＋ｎ＝１０である。Ｐ_ｉは遅延させようとする画素の濃淡階調、Ｐ_ｉ＋１は隣の画素の濃淡階調、０．４画素遅延する場合はｍ＝６，ｎ＝４であり、Ｑｉは０．４画素遅延した位置の濃淡階調である。これにより０．１画素刻みの遅延を行うことができる。２周期遅延の場合は、９．４×２＝１８．８画素になるので、先と同様に（１）式を用いてｍ＝２，ｎ＝８とし、０．８画素遅延した場所のデータを線形補間で算出し、さらにそのデータを１８画素（整数部）遅らせることにより、２周期遅延画像を得ることができる。
【００３１】
図３は白欠陥、黒欠陥の検出方法を示す。白欠陥、黒欠陥の検出信号の立ち上がりエッジで座標値を読み取ることにより、欠陥の所在位置が判明する。座標値はラインセンサの１画素毎にインクリメントするＸ座標と、ラインセンサの画像取り込み開始信号毎にインクリメントするＹ座標とがある。本装置では、欠陥の立ち上がりと、立ち下がりの座標値をランレングスデータとして記憶している。このランレングスデータを連結処理することにより、欠陥の大きさを知ることができる。
【００３２】
図４はラインセンサより得られる１ライン画像Ａと、このＡの１周期遅延画像Ｂ、およびＡの２周期遅延画像Ｃの関係を表す図である。▲１▼，▲２▼，▲３▼は連続した３周期長の画像を表す。以下に述べるＡ，Ｂ，Ｃ間の差分は同一時刻における各周期長の差分を演算する。例えば、差分Ｂ−ＡはＢの▲２▼とＡの▲３▼の差分，差分Ｂ−ＣはＢの▲２▼とＣの▲１▼の差分である。
【００３３】
図５は図１の欠陥判定回路６の内容を表す。Ａ，Ｂ，Ｃは図４に示す値である。差分２値化器２１ａは画像ＡとＢとの差分Ａ−Ｂを演算し、この値を所定のしきい値で２値化し、これを２値化差分ａ−ｂで表す。差分は値が負となった場合は０とする。所定のしきい値は、差分が欠陥を表す限界を示すもので、このしきい値を超える差分は欠陥となり、未満の差分は正常なものとする。このしきい値は検査対象物１の種類などに応じて定められるものである。同様に差分２値化器２１ｂは差分Ｂ−Ａを演算し、２値化差分ｂ−ａを出力し、差分２値化器２１ｃは差分Ｂ−Ｃを演算し、２値化差分ｂ−ｃを出力し、差分２値化器２１ｄは差分Ｃ−Ｂを演算し、２値化差分ｃ−ｂを出力する。
【００３４】
１周期遅延器２２は２値化差分値を１周期遅延する。これにより２値化差分ｂ−ｃは１周期遅延され、これをＤ（ｂ−ｃ）で表す。同様に２値化差分ｃ−ｂの１周期遅延値もＤ（ｃ−ｂ）で表す。第１論理積演算器２３は２値化差分ｂ−ａとｂ−ｃの論理積を演算する。第２論理積演算器２４は２値化差分ａ−ｂとｃ−ｂの論理積を演算する。第３論理積演算器２５は２値化差分ｂ−ａの反転値と、ｂ−ｃと、ｃ−ｂの１周期遅延値Ｄ（ｃ−ｂ）の反転値の論理積を演算する。第４論理積演算器２６は２値化差分ａ−ｂの反転値と、ｃ−ｂと、ｂ−ｃの１周期遅延値Ｄ（ｂ−ｃ）の反転値の論理積を演算する。判定回路２７は各論理積演算器２３〜２６の出力から欠陥の判定を行う。
【００３５】
判定回路２７は次の欠陥判定を行う。第１論理積演算器２３の出力が１のとき、白欠陥と判定し、かつ、第３および第４論理積演算器２５，２６の出力が１でないとき、１周期長以内の白欠陥と判定する。第２論理積演算器２４の出力が１のとき、黒欠陥と判定し、かつ、第３および第４論理積演算器２５，２６の出力が１でないとき１周期長以内の黒欠陥と判定する。第３論理積演算器２５の出力が１で、これに続いて第４論理積演算器２６の出力が１となるとき、１周期長以上の白欠陥と判定し、第４論理積演算器２６の出力が１で、これに続いて第３論理積演算器２５の出力が１となる場合、１周期長以上の黒欠陥と判定する。
【００３６】
図６〜図９は図５の欠陥判定回路６のタイミングチャートである。図６は１周期長以内の白欠陥を検出するタイミングチャート、図７は１周期長以内の黒欠陥を検出するタイミングチャート、図８は１周期長を超える白欠陥を検出するタイミングチャート、図９は１周期長を超える黒欠陥を検出するタイミングチャートである。
【００３７】
図６において、Ａはラインセンサより出力される原画像で、検査対象物１としては図２に示した液晶パネルである。周期的に表れる下方に向いたパルス３０はリード線１１を表し、このパルス間の長さが１周期長となる。パルス間を結ぶ水平線はリード線１１間のフィルタの濃淡レベル３１を表し、このレベル３１より黒は下方、白は上方である。レベル３１より上方に盛り上がった形状が１周期長以内の白欠陥３２を表している。ＢはＡの１周期遅延画像、ＣはＡの２周期遅延画像である。差分Ｂ−Ａ，Ｂ−Ｃ，Ａ−Ｂ，Ｃ−Ｂでは負の値は０とする。これら差分を所定のしきい値で２値化した２値化差分ｂ−ａ，ｂ−ｃ，ａ−ｂ，ｃ−ｂは、差分Ｂ−Ａ，Ｂ−Ｃ，Ａ−Ｂ，Ｃ−Ｂと同じ形状となっている。第１論理積演算器２３はｂ−ａとｂ−ｃの値が１のとき、出力１となる。白欠陥の場合、図６に示すようにｂ−ａ，ｂ−ｃは１となっているので、第１論理積演算器２３の出力は１となる。これにより１周期長以内の白欠陥を検出することができる。
【００３８】
次に１周期長以下の黒欠陥の検出を説明する。図７において、黒欠陥があるとパルス３０間の濃淡レベル３１に黒欠陥３３が表れる。第２論理積演算器２４は２値化差分ａ−ｂ，ｃ−ｂが共に１のとき１を出力する。黒欠陥の場合、ａ−ｂ，ｃ−ｂは共に１となるので第２論理積演算器２４の出力が１のとき１周期長以内の黒欠陥と判定する。
【００３９】
次に１周期長を超える白欠陥の検出を説明する。図８において、１周期長を超える白欠陥３４が表れている。第３論理積演算器２５は２値化差分ｂ−ａの反転値と、ｂ−ｃと、ｃ−ｂの１周期遅延値Ｄ（ｃ−ｂ）の反転値が１のとき１を出力する。また、第４論理積演算器２６は２値化差分ａ−ｂの反転値と、ｃ−ｂと、ｂ−ｃの１周期遅延値Ｄ（ｂ−ｃ）の反転値が１のとき、１を出力する。１周期長を超える白欠陥の場合、周期Ｐ（最初に白欠陥が表れる周期Ｐ−１の次の周期）において、第３論理積演算器２５の出力が１となる。また周期Ｑ（Ｃ画像で白欠陥が表れる最後の周期）で第４論理積演算器２６の出力が１となる。このように第３論理積演算器２５の出力が１となり、これに続いて第４論理積演算器２６の出力が１となったとき、１周期長を超える白欠陥と判定する。なお、１周期長以内の白欠陥の場合は第３および第４論理積演算器２５，２６は１とならない。しかし第１論理積演算器２３は１周期長を超える白欠陥の場合、１となる場合がある。故に１周期長を超える白欠陥か否かは第３および第４論理積演算器２５，２６の出力によって判定する。
【００４０】
次に１周期長を超える黒欠陥の検出を説明する。図９において、１周期長を超える黒欠陥３５が表れている。１周期長を超える黒欠陥の場合、周期Ｐ（最初に黒欠陥が表れる周期Ｐ−１の次の周期）において、第４論理積演算器２６の出力が１となる。また周期Ｑ（Ｃ画像で黒欠陥が表れる最後の周期）で第３論理積演算器２５の出力が１となる。このように第４論理積演算器２６の出力が１となり、続いて第３論理積演算器２５の出力が１となったとき、１周期長を超える黒欠陥と判定する。なお、１周期長以内の黒欠陥の場合は第３および第４論理積演算器２５，２６の出力は１とならない。しかし、第２論理積演算器２４は１周期長を超える黒欠陥の場合１となる場合がある。故に１周期長を超える黒欠陥か否かは第３および第４論理積演算器２５，２６の出力によって判定する。
【００４１】
以上の説明は濃淡階調設定において、最暗レベルを０、最明レベルをＮ（８ビットの濃淡表示の場合Ｎ＝２５５）としたが、これを逆にして最暗レベルをＮ，最明レベルを０としても本発明は成立する。また、実施形態では、リニアセンサを用いた場合で説明したが、面撮像素子の１走査線について行ってもよい。
【００４２】
【発明の効果】
以上の説明より明らかなように、本発明は元の画像Ａ，この１周期遅延画像Ｂ，Ａの２周期遅延画像Ｃについて、Ｂに対するＡまたはＣの差分を求め、この２値化値の論理積を演算することにより、１周期長以内またはこれを超えるか否かを区別し、さらに白欠陥または黒欠陥を判定することができる。
【図面の簡単な説明】
【図１】実施形態の構成を示すブロック図である。
【図２】検査対象物の一例を示す図である。
【図３】欠陥を検出する方法を説明する図である。
【図４】ラインセンサより得られる１ライン画像Ａ，１周期長遅延画像Ｂ，２周期長遅延画像Ｃを示す図である。
【図５】欠陥判定回路の構成を示す図である。
【図６】１周期長以内の白欠陥を検出するタイミングチャートである。
【図７】１周期長以内の黒欠陥を検出するタイミングチャートである。
【図８】１周期長を超える白欠陥を検出するタイミングチャートである。
【図９】１周期長を超える黒欠陥を検出するタイミングチャートである。
【符号の説明】
１検査対象物
２ラインセンサカメラ
３Ａ／Ｄ変換器
４１周期長遅延回路
５２周期長遅延回路
６欠陥判定回路[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a periodic pattern inspection apparatus for inspecting defects such as scratches, pinholes, black spots, dust, and missing patterns in an inspection object having a periodic pattern such as a liquid crystal panel, a color filter thereof, and a shadow mask.
[0002]
[Prior art]
Conventionally, methods for detecting a defect in a periodic pattern include JP-A-4-316346, JP-A-6-129998, and JP-A-7-159343, which scan a periodic pattern with a line sensor and scan the original pattern. A technique is disclosed in which an image is compared with an image shifted from the image, and if the image is different, the difference is detected as a defect.
[0003]
[Problems to be solved by the invention]
Defects include scratches, pinholes, dirt, dust, missing patterns, etc., which appear black or white on the image. For this reason, if it is known whether the defect is black or white, the cause of the defect can be considerably estimated. In many cases, the defects have a length of one cycle or less, but some defects exceed one cycle. However, according to the conventional technique, it is not possible to detect the white and black defects separately at the same time. Further, it was not possible to detect a defect exceeding one cycle length and whether the defect was white or black.
[0004]
SUMMARY OF THE INVENTION The present invention has been made in view of the above-described problems, and has as its object to provide an inspection apparatus that detects a defect occurring in a periodic pattern by distinguishing the defect into a white defect and a black defect. It is still another object of the present invention to provide a device for distinguishing and detecting a white defect and a black defect even for a large defect exceeding one cycle length.
[0007]
[Means for Solving the Problems]
The invention according to claim 1 is an apparatus for inspecting a defect of an inspection object in which a pattern is repeated at a constant cycle, wherein the pattern is linearly scanned, and an image sensor is read by an image sensor to obtain a multi-valued image. A multi-valued image is denoted by A, a multi-valued image delayed by one cycle from A is denoted by B, a multi-valued image delayed by two cycles from A is denoted by C, and the difference of each multi-valued image at the same time is A and Difference calculating means for calculating a difference AB between B, a difference BA between B and A, a difference BC between B and C, and a difference CB between C and B; A binarizing means for performing binarization to set B, BA, BC, and CB to 0 when the value is larger than a predetermined threshold value and to 0 when the value is smaller than a predetermined threshold value, including negative values; The binarized value of each difference by the binarizing means is represented by ab, ba, bc, c by the corresponding lowercase alphabet. When expressed as b, first AND operation means for calculating the logical AND of ba and bc, and second AND operation means for calculating the logical AND of ab and cb, The multi-valued image is represented by multi-values from M to N (M and N are both positive and M <N).
[0008]
Assuming that the darkest level is 0 and the brightest level is N for the multi-valued image, the white defect has a density level higher than a certain threshold value (L1) as compared with the density values on both sides separated by the cycle length. Therefore, the black defect has a density level lower than other threshold values (L2, L2 <LI). When there is no defect, the density level is almost the same as that on both sides. Now, it is assumed that there is a defect within one cycle length, which is a white defect. Taking the difference B-A and the difference B-C from A at the same time centering on the image B of the cycle in which the defect occurs, the difference B-A and the difference B-C have positive values. The difference BA and the difference BC are compared with a predetermined threshold value L1. By appropriately setting the threshold value to a value representing the defect, the defect can be reliably taken out. When both the binarized differences ba and bc are 1, it can be determined as a white defect. That is, by comparing a certain period with the immediately preceding period, and comparing a certain period with the immediately following period, it is determined whether or not there is a defect within a certain period, and whether the defect is a white defect or a black defect. , A defect within the period.
[0009]
Further, when there is a black defect of one cycle length or less, the difference AB between A and C at the same time and the difference CB from C at the same time are calculated with the pixel B of the cycle in which the defect occurs as the center. The difference AB and the difference CB are positive values. The difference AB and the difference CB are compared with a predetermined threshold value L2. By appropriately setting the threshold value to a value representing the defect, the defect can be reliably taken out. When both the binarized differences ab and cb are 1, it can be determined as a black defect. As a result, it is possible to determine not only a defect but also a white defect or a black defect.
[0010]
According to the fourth aspect of the present invention, when the darkest level is 0 and the brightest level is N (positive number) in the multivalued image, the output of the first AND operation means is 1 In this case, it is determined that the defect is a white defect of one cycle length or less, and when the output of the second AND operation means is 1, the defect is a black defect of one cycle length or less.
[0011]
When a multi-value image is displayed with multi-values from 0 to N, where 0 is the darkest level and N is the brightest level, the density is higher than the threshold value L1 as compared with the density on both sides of one cycle length. Sometimes a white defect. Therefore, this corresponds to the case where both the binarized differences ba and bc described in claim 2 are 1, which is the case where the output of the first AND operation means is 1. Further, when the density is lower than the threshold value L2 (L1> L2) as compared with the density on both sides of one cycle length, it is a black defect. Therefore, this corresponds to the case where both the binarization differences ab and cb described in claim 2 are 1, which is the case where the output of the second AND operation means is 1.
[0012]
According to the fifth aspect of the present invention, when the brightest level is displayed as 0 and the darkest level is displayed as N (positive number) in the multivalued image, the output of the first AND operation means is 1 In this case, it is determined that the defect is a black defect having one cycle length or less, and when the output of the second AND operation means is 1, the defect is a white defect having one cycle length or less.
[0013]
In the present invention, the setting of the multi-value conversion is reversed from the case of the fourth aspect , and the multi-value conversion is represented with 0 being the brightest level and N being the darkest level. A defect is a white defect when the value is smaller than the threshold L2 than on both sides one cycle away, and a black defect when the value is larger than the threshold L1 (L1> L2) than on both sides one cycle apart. In the case of a black defect, this corresponds to a case where both the binarized differences ba and bc described in claim 2 are 1, and this is a case where the output of the first AND operation means is 1. . When the density is smaller than the threshold value L2 as compared with the density on both sides of one cycle length, it is a white defect. In the case of a white defect, this corresponds to the case where both the binarized differences ab and cb described in claim 1 are 1, which is the case where the output of the second AND operation means is 1.
[0014]
In a second aspect of the present invention, there is provided an apparatus for inspecting a defect of an inspection object in which a pattern is repeated at a constant period, wherein the pattern input is linearly scanned and an image sensor is read by an image sensor to obtain a multi-valued image. A multi-valued image is denoted by A, a multi-valued image delayed by one cycle length from A is denoted by B, a multi-valued image delayed by two cycles from A is denoted by C, and the difference of each multi-valued image at the same time, Difference calculating means for calculating a difference AB between A and B, a difference BA between B and A, a difference BC between B and C, and a difference CB between C and B; A binarizing means for performing binarization to set AB to 0 when the values of AB, BA, BC, and CB are larger than a predetermined threshold value, including 1 when the value is smaller than 1; The binarized value of each difference by the binarizing means is represented by ab, ba, bc by the corresponding lowercase alphabet. When expressed as cb, a third AND operation means for obtaining a logical product of the inverted value of ba, bc, and the inverted value of the one-cycle delay value of cb, A fourth AND operation means for obtaining a logical product of the inverted value, cb, and the inverted value of the one-cycle delay value of bc.
[0015]
The feature of a white defect longer than one cycle length will be described with reference to FIG. First, in the first period in which a long defect starts, in the period P, B and A are at the same level, and C is at a level lower than A and B. Further, during the period of P-1, B and C are at the same level. This means that the one-cycle delay value of the binary difference ba is 0, bc is 1 and cb is 0. Therefore, the inverted value of ba, the value of bc, and the inverted value of the one-cycle delay value of cb all become 1, and the output of the third AND operation means becomes 1.
[0016]
In the period in which the long defect ends, in the period Q, A and B are at the same level, and C is at a higher level than A and B. Further, during the Q-1 period, C and B are at the same level. This means that the binary difference ab is 0, cb is 1, and the one-cycle delay value of bc is 0 in the invention of claim 5. Therefore, since the inverted value of ab, the value of bc, and the inverted value of the one-cycle delay value of bc are all 1, the output of the fourth AND operation means is 1.
[0017]
The feature of a black defect longer than one cycle length will be described with reference to FIG. In the first period in which a long defect starts, in the period P, B and A are at the same level, and C is at a higher level than B and A. Further, during the period P-1, B and C are at the same level. This means that the binary difference ab is 0, cb is 1, and the one-cycle delay value of bc is 0 in the invention of claim 5. Therefore, since the inverted value of ab, the value of bc, and the inverted value of the one-cycle delay value of bc are all 1, the output of the fourth AND operation means is 1.
[0018]
Further, in the cycle in which the long defect ends, in the period Q, B and A are at the same level, and C is at a lower level than B and A. Further, during the period of Q-1, B and C are at the same level. This means that the binary difference ba is 0, bc is 1 and the one-cycle delay value of bb is 0 in the invention of claim 5. Therefore, the inverted value of ba, the value of bc, and the inverted value of the one-cycle delay value of cb all become 1, and the output of the third AND operation means becomes 1.
[0019]
As described above, when there is a white defect having a length exceeding one cycle length, the output of the third AND operation means becomes 1 first, and then the output of the fourth AND operation means becomes 1. If a black defect having a length exceeding one cycle length exists, the output of the fourth AND operation means becomes 1 and then the output of the third AND operation means becomes 1. As a result, a long defect exceeding one cycle length can be detected by distinguishing whether it is a white defect or a black defect.
[0020]
In the invention according to claim 6, when the multi-valued image is displayed by multi-valued display with the darkest level being 0 and the brightest level being N (positive number), the output of the third AND operation means becomes 1. Then, when the output of the fourth AND operation means becomes 1, it is determined that the white defect exceeds one cycle length, and the output of the fourth AND operation means becomes 1, and When the output of the third AND operation unit becomes 1, it is determined that the black defect exceeds one cycle length.
[0021]
According to the second aspect of the present invention, when the multivalued display is such that 0 is the darkest level and N is the brightest level, and there is a long defect exceeding one cycle length as a white defect, the third AND operation means Becomes 1, and then the output of the fourth AND operation means becomes 1. When a black defect having a length exceeding one cycle length exists, the output of the fourth AND operation means becomes 1, and subsequently, the output of the third AND operation means becomes 1.
[0022]
According to the seventh aspect of the present invention, when the brightest level is represented by 0 and the darkest level is represented by N (positive number) in the multivalued image, the output of the third AND operation means becomes 1. When the output of the fourth AND operation unit subsequently becomes 1, it is determined that the black defect exceeds one cycle length, the output of the fourth AND operation unit becomes 1, and When the output of the third AND operation means becomes 1, it is determined that the white defect exceeds one cycle length.
[0023]
According to the second aspect of the present invention, when the multilevel display is 0 for the brightest level and N for the darkest level, if there is a black defect longer than one cycle length, the third AND operation means Becomes 1, and then the output of the fourth AND operation means becomes 1. If a white defect having a length exceeding one cycle length exists, the output of the fourth AND operation means becomes 1 and then the output of the third AND operation means becomes 1.
[0024]
In a third aspect of the present invention, there is provided an apparatus for inspecting a defect of an inspection object in which a pattern is repeated at a constant cycle, wherein the image input means for linearly scanning the pattern and reading the image with an image sensor to obtain a multi-valued image. A multi-valued image is denoted by A, a multi-valued image delayed by one cycle from A is denoted by B, a multi-valued image delayed by two periods from A is denoted by C, and the difference of each multi-valued image at the same time is A and Difference calculating means for calculating a difference AB between B, a difference BA between B and A, a difference BC between B and C, and a difference CB between C and B; A binarizing means for performing binarization to set B, BA, BC, and CB to 0 when the value is larger than a predetermined threshold value and to 0 when the value is smaller than a predetermined threshold value, including negative values; The binarized value of each difference by the binarizing means is represented by ab, ba, bc, c by the corresponding lowercase alphabet. When expressed as b, first AND operation means for calculating the AND of ba and bc, second AND operation means for calculating the AND of ab and cb, b A third AND operation means for obtaining a logical product of an inverted value of -a, bc, and an inverted value of the one-cycle delay value of cb, an inverted value of ab, cb, and a fourth AND operation means for obtaining an AND with an inverted value of the one-cycle delay value of bc.
[0025]
The present invention is a combination of the invention of claim 1 and the invention of claim 2 , wherein defects within one cycle length can be detected by distinguishing between white defects and black defects, and long defects exceeding one cycle length can be detected as white defects. And black defects can be detected separately.
[0026]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of the present embodiment. The inspection object 1 moves in the direction of the arrow, and has a periodic pattern in a direction orthogonal to the moving direction. The line sensor camera 2 arranges the length direction of the built-in line sensor (the arrangement direction of the sensor elements) so as to be orthogonal to the moving direction of the inspection object 1, and captures an image having a moving periodic pattern. The A / D converter 3 converts the grayscale image of the video signal from the line sensor camera 2 from analog to digital and converts it into an 8-bit (256 gradation) multivalued image. The one cycle length delay circuit 4 delays A by one cycle length and sets the output to B. The two-cycle delay circuit 5 delays A by two cycles, and sets the output to C. The defect determination circuit 6 receives A, B, and C, and detects a defect occurring in a periodic pattern. Defects are flaws, pinholes, dirt, dust, missing patterns, etc. These are gradations in which 0 is the darkest level, 255 is the brightest level in the multi-valued image, and the interval between them is sequentially changed from the darkest level to the brightest level. In this case, a white defect is determined when the level is higher than a certain threshold value L1 as compared with the adjacent pixels one cycle long apart, and a black defect is determined when the level is lower than another threshold value L2 (L1> L2). In the following description, 0 is the darkest level and 255 is the brightest level. The defect judging circuit 6 classifies the defect as a white defect or a black defect and outputs it as a binary image.
[0027]
FIG. 2 shows an example of a liquid crystal panel as the inspection object 1. In this configuration, a lead line 11 and a gate line 12 are arranged in a grid pattern, and a TFT (Thin Film Transistor) is arranged at an intersection. The grid intervals are arranged regularly in both the horizontal and vertical directions, and the line sensor camera 2 scans in the gate line direction (horizontal direction) to obtain an image A in which the interval between the lead wires 11 is one cycle length. The defects include a white defect 14 and a black defect 15 within one cycle length, and a white length defect 16 and a black length defect 17 exceeding one cycle length.
[0028]
The image picked up by the line sensor of the line sensor camera 2 is composed of a single pixel line in the scanning direction (horizontal direction in FIG. 2, direction of the gate line 12). The line sensor is configured to increase detection accuracy. That is, the number of pixels in one cycle is determined so that the resolution required for pattern inspection can be obtained. The size of the sensor element of the line sensor is the size of one pixel of the image to be captured. The size of this pixel is, for example, about 10 μm, and when one cycle length is 100 μm, about ten pixels represent one cycle length. However, since the cycle length and the size of the sensor element are independent of each other, it is common that the number of pixels representing one cycle length does not become an integer. For example, one cycle length is represented by 9.4 pixels.
[0029]
In the case where one cycle length is not represented by an integer number of pixels, a method of creating a one-cycle delayed image B and a two-cycle delayed image C will be described. For example, when one cycle length is 9.4 pixels, in order to delay each pixel constituting one cycle by one cycle length (that is, 9.4 pixels), first, linear interpolation is performed by the following equation (1). Calculate the data at the place delayed by 0.4 pixels. Further, by delaying the data by 9 pixels (an integer part), the data can be delayed by 9.4 pixels.
[0030]
Linear interpolation formula: Qi = (mP _i + nP _{i + 1} ) / 10 (1)
Here, m + n = 10. P _i is the gray scale of the pixel to be delayed, P _{i + 1} is the gray scale of the adjacent pixel, m = 6, n = 4 when delayed by 0.4 pixels, and Qi is 0.4 pixel delayed This is the gradation of the shaded position. Thus, a delay of 0.1 pixel can be performed. In the case of a two-cycle delay, 9.4 × 2 = 18.8 pixels, so that m = 2 and n = 8 using equation (1), and data at a location delayed by 0.8 pixels as in the above case. Is calculated by linear interpolation, and the data is further delayed by 18 pixels (an integer part), whereby a two-cycle delayed image can be obtained.
[0031]
FIG. 3 shows a method for detecting a white defect and a black defect. By reading the coordinate value at the rising edge of the detection signal of the white defect and the black defect, the location of the defect is determined. The coordinate value includes an X coordinate that is incremented for each pixel of the line sensor and a Y coordinate that is incremented for each image capture start signal of the line sensor. In this apparatus, the coordinate values of the rise and fall of the defect are stored as run-length data. By linking the run-length data, the size of the defect can be known.
[0032]
FIG. 4 is a diagram showing a relationship between a one-line image A obtained from the line sensor, a one-cycle delayed image B of A, and a two-cycle delayed image C of A. {Circle around (1)}, {circle around (2)}, and {circle around (3)} represent images having three continuous cycle lengths. The difference between A, B, and C described below calculates the difference of each cycle length at the same time. For example, the difference BA is the difference between (2) of B and (3) of A, and the difference BC is the difference between (2) of B and (1) of C.
[0033]
FIG. 5 shows the contents of the defect determination circuit 6 of FIG. A, B, and C are the values shown in FIG. The difference binarizer 21a calculates a difference AB between the images A and B, binarizes this value with a predetermined threshold value, and expresses this as a binarized difference ab. The difference is set to 0 when the value becomes negative. The predetermined threshold value indicates a limit at which the difference indicates a defect. A difference exceeding this threshold value is a defect, and a difference less than the threshold value is normal. This threshold is determined according to the type of the inspection object 1 and the like. Similarly, the difference binarizer 21b calculates the difference BA and outputs a binary difference ba, and the difference binarizer 21c calculates the difference BC and calculates the binary difference bc. , And the difference binarizer 21d calculates the difference CB, and outputs a binarized difference cb.
[0034]
The one-cycle delay unit 22 delays the binarized difference value by one cycle. As a result, the binary difference bc is delayed by one cycle, and is represented by D (bc). Similarly, the one-cycle delay value of the binary difference c-b is also represented by D (c-b). The first AND operator 23 calculates the AND of the binary differences ba and bc. The second AND operator 24 calculates the AND of the binarized differences ab and cb. The third AND operator 25 calculates the logical AND of the inverted value of the binary difference ba, the inverted value of bc and the inverted value of the one-cycle delay value D (cb) of cb. The fourth AND operator 26 calculates the AND of the inverted value of the binarization difference ab, the inverted value of cb and the one-cycle delay value D (bc) of bc. The determination circuit 27 determines a defect from the output of each of the logical product arithmetic units 23 to 26.
[0035]
The determination circuit 27 performs the following defect determination. When the output of the first AND operator 23 is 1, it is determined that the defect is a white defect, and when the outputs of the third and fourth AND operators 25 and 26 are not 1, it is determined that the defect is within one cycle length. I do. When the output of the second AND operator 24 is 1, it is determined as a black defect, and when the outputs of the third and fourth AND operators 25 and 26 are not 1, it is determined as a black defect within one cycle length. . When the output of the third AND operator 25 is 1 and the output of the fourth AND operator 26 subsequently becomes 1, it is determined that the white defect is longer than one cycle length, and the fourth AND operator 26 Is 1 and the output of the third AND operator 25 subsequently becomes 1, it is determined that the black defect has one cycle length or more.
[0036]
6 to 9 are timing charts of the defect determination circuit 6 of FIG. 6 is a timing chart for detecting a white defect within one cycle length, FIG. 7 is a timing chart for detecting a black defect within one cycle length, FIG. 8 is a timing chart for detecting a white defect exceeding one cycle length, and FIG. Is a timing chart for detecting a black defect exceeding one cycle length.
[0037]
In FIG. 6, A is an original image output from the line sensor, and the inspection object 1 is the liquid crystal panel shown in FIG. The downwardly directed pulses 30, which appear periodically, represent the leads 11, the length between the pulses being one period long. The horizontal line connecting the pulses represents the shading level 31 of the filter between the lead wires 11, from which black is below and white is above. The shape raised above the level 31 indicates a white defect 32 within one cycle length. B is a one-cycle delayed image of A, and C is a two-cycle delayed image of A. For the differences BA, BC, AB, and CB, the negative value is 0. The binarized differences ba, bc, ab, and bb obtained by binarizing these differences with a predetermined threshold value are defined as differences BA, BC, AB, and CB. It has the same shape as. The first AND operator 23 outputs 1 when the values of ba and bc are 1. In the case of a white defect, since ba and bc are 1 as shown in FIG. 6, the output of the first AND operator 23 is 1. As a result, a white defect within one cycle length can be detected.
[0038]
Next, detection of a black defect of one cycle length or less will be described. In FIG. 7, if there is a black defect, a black defect 33 appears at the gray level 31 between the pulses 30. The second AND operator 24 outputs 1 when the binary differences ab and cb are both 1. In the case of a black defect, both ab and cb become 1, so when the output of the second AND operator 24 is 1, it is determined that the black defect is within one cycle length.
[0039]
Next, detection of a white defect exceeding one cycle length will be described. In FIG. 8, a white defect 34 exceeding one cycle length appears. The third AND operator 25 outputs 1 when the inverted value of the binary difference ba and the inverted value of the one-cycle delay value D (c−b) of bc and c−b are 1. . The fourth AND operator 26 calculates 1 when the inverted value of the binary difference ab, the inverted value of cb, and the one-cycle delay value D (bc) of bc are 1, Is output. In the case of a white defect exceeding one cycle length, the output of the third AND operator 25 becomes 1 in the period P (the period next to the period P-1 in which the white defect first appears). The output of the fourth AND operator 26 becomes 1 in the cycle Q (the last cycle in which a white defect appears in the C image). As described above, when the output of the third AND operator 25 becomes 1, and subsequently the output of the fourth AND operator 26 becomes 1, it is determined that the white defect exceeds one cycle length. In the case of a white defect within one cycle length, the third and fourth AND operation units 25 and 26 do not become 1. However, the first AND operation unit 23 may become 1 in the case of a white defect exceeding one cycle length. Therefore, whether or not the white defect exceeds one cycle length is determined by the outputs of the third and fourth AND operators 25 and 26.
[0040]
Next, detection of a black defect exceeding one cycle length will be described. In FIG. 9, a black defect 35 exceeding one cycle length appears. In the case of a black defect exceeding one cycle length, the output of the fourth AND operator 26 becomes 1 in the period P (the period next to the period P-1 in which the black defect first appears). Further, the output of the third AND operator 25 becomes 1 in the cycle Q (the last cycle in which the black defect appears in the C image). As described above, when the output of the fourth AND operator 26 becomes 1, and subsequently the output of the third AND operator 25 becomes 1, it is determined that the black defect exceeds one cycle length. In the case of a black defect within one cycle length, the outputs of the third and fourth AND operators 25 and 26 do not become 1. However, the second AND operator 24 may become 1 when a black defect exceeds one cycle length. Therefore, whether or not the black defect exceeds one cycle length is determined based on the outputs of the third and fourth AND operators 25 and 26.
[0041]
In the above description, the darkest level is set to 0 and the brightest level is set to N (N = 255 in the case of 8-bit grayscale display) in the gray scale setting. The present invention holds even when the level is set to 0. Further, in the embodiment, the case where the linear sensor is used has been described, but it may be performed for one scanning line of the surface imaging device.
[0042]
【The invention's effect】
As is clear from the above description, according to the present invention, for the original image A, the one-cycle delayed image B, and the two-cycle delayed image C of A, the difference between A and C with respect to B is obtained, By calculating the product, it is possible to discriminate whether it is within one cycle length or more than one cycle length, and to determine a white defect or a black defect.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of an embodiment.
FIG. 2 is a diagram illustrating an example of an inspection object.
FIG. 3 is a diagram illustrating a method for detecting a defect.
FIG. 4 is a diagram showing a one-line image A, a one-cycle length delay image B, and a two-cycle length delay image C obtained from a line sensor.
FIG. 5 is a diagram illustrating a configuration of a defect determination circuit.
FIG. 6 is a timing chart for detecting a white defect within one cycle length.
FIG. 7 is a timing chart for detecting a black defect within one cycle length.
FIG. 8 is a timing chart for detecting a white defect exceeding one cycle length.
FIG. 9 is a timing chart for detecting a black defect exceeding one cycle length.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Inspection object 2 Line sensor camera 3 A / D converter 4 1 cycle length delay circuit 5 2 cycle length delay circuit 6 Defect judgment circuit

Claims

An apparatus for inspecting a defect of an inspection object in which a pattern is repeated at a constant cycle, wherein the pattern is linearly scanned, and an image input means for obtaining a multi-valued image read by an image sensor, and the multi-valued image is A, The multi-valued image delayed by one cycle from A is denoted by B, the multi-valued image delayed by two periods from A is denoted by C, and the difference of each multi-valued image at the same time, which is the difference A between B and A -B, a difference calculation means for calculating a difference BA between B and A, a difference BC between B and C, and a difference CB between C and B; and the difference AB, BA , BC and CB are larger than a predetermined threshold value, and are smaller than 1 when they are smaller than a predetermined threshold value. When the binarized value of each difference is expressed as ab, ba, bc, bb by the corresponding lowercase alphabetic character , A logical product of ba and bc, a second logical product operation device for calculating a logical product of ab and cb; A defect inspection apparatus for a periodic pattern, wherein an image is represented by a multivalue of M to N (M and N are both positive numbers and M <N).

An apparatus for inspecting a defect of an inspection object in which a pattern is repeated at a constant cycle, wherein the pattern is linearly scanned, an image input unit that reads a multivalued image by an image sensor, and the multivalued image is A, The multi-valued image delayed by one cycle from A is denoted by B, the multi-valued image delayed by two periods from A is denoted by C, and the difference of each multi-valued image at the same time, which is the difference A between B and A -B, a difference calculation means for calculating a difference BA between B and A, a difference BC between B and C, and a difference CB between C and B; and the difference AB, BA , BC and CB are larger than a predetermined threshold value, and are smaller than 1 when they are smaller than a predetermined threshold value. When the binarized value of each difference is expressed as ab, ba, bc, bb by the corresponding lowercase alphabetic character , Ba, an inverted value of bc, and an inverted value of the one-period delay value of cb, a third AND operation means; an inverted value of ab, cb And a fourth AND operation means for obtaining an AND of an inverted value of the one-cycle delay value of bc.

An apparatus for inspecting a defect of an object to be inspected in which a pattern is repeated at a constant cycle, wherein said pattern is read by a linear sensor to obtain a multi-valued image; A multi-valued image delayed by a cycle length is denoted by B, a multi-valued image delayed by two periods from A is denoted by C, and a difference between each of the multi-valued images at the same time, that is, a difference AB, B between A and B Calculating means for calculating a difference BA between A and A, a difference BC between B and C, and a difference CB between C and B; and each of the differences AB, BA, BC , CB is larger than a predetermined threshold value, and when it is less than 1, a negative value including a negative value is used, including a case where the difference is negative. When the quantified values are expressed as ab, ba, bc, and bb by the corresponding lowercase letters of the alphabet, ba and b first AND operation means for operating the AND of c, second AND operation means for operating the AND of ab and cb, an inverted value of ba, bc, third AND operation means for obtaining a logical product of the inverted value of the one-cycle delay value of cb and the inverted value of ab, cb, and the inverted value of the one-cycle delay value of bc; And a fourth logical product calculating means for obtaining a logical product of the following.

When the darkest level is 0 and the brightest level is N (positive number) in the multi-valued image, when the output of the first AND operation unit is 1, the output is shorter than one cycle length. 4. The periodic pattern defect inspection apparatus according to claim 1 , wherein when a white defect and an output of the second AND operation means are 1, the defect is determined to be a black defect having one cycle length or less.

When the brightest level is 0 and the darkest level is N (positive number) in the multi-valued image, when the output of the first AND operation means is 1, the length is less than one cycle length. black defect, the output is a defect inspection apparatus of the periodic pattern according to claim 1 or 3, wherein determining that one period length less white defect when one of the second aND operation means.

When the darkest level is 0 and the brightest level is N (positive number) in the multi-valued image, the output of the third AND operation means is 1, and subsequently the output of the third AND operation unit is 1. When the output of the four AND operation means becomes 1, it is determined that the white defect exceeds one cycle length, the output of the fourth AND operation means becomes one, and subsequently, the white defect of the third AND operation means becomes one. 4. The defect inspection apparatus for a periodic pattern according to claim 2 , wherein when the output becomes 1, a black defect exceeding one cycle length is determined.

When the brightest level is 0 and the darkest level is N (positive number) in the multi-valued image, the output of the third AND operation means becomes 1, and subsequently the output of the third AND operation means becomes 1. When the output of the four AND operation means becomes 1, it is determined that the black defect exceeds one cycle length, and the output of the fourth AND operation means becomes one. 4. The defect inspection apparatus for a periodic pattern according to claim 2 , wherein when the output becomes 1, a white defect exceeding one cycle length is determined.