JP4249966B2 - Printed wiring board inspection method and inspection apparatus - Google Patents

Description

（ii）Ｓ（ｐ_０）＝１、ここで、Ｓ（ｐ_０）はｐ_０の近傍画素の値Ｂ（ｐ_１），Ｂ（ｐ_２），Ｂ（ｐ_３），・・・・，Ｂ（ｐ_８），Ｂ（ｐ_１）をこの順序で調べたときの、画素値が０から１に変化している回数である。たとえば、図４の場合、Ｎ（ｐ_０）＝３，Ｓ（ｐ_０）＝２である。
（iii）Ｂ（ｐ_１）・Ｂ（ｐ_３）・Ｂ（ｐ_７）＝０
（iv） Ｂ（ｐ_１）・Ｂ（ｐ_５）・Ｂ（ｐ_７）＝０
「ステップ２」;ステップ１の条件（iii）並びに（iv）が次のように変わる。他の条件、処理の手順はステップ１と同じである。
（iii）’Ｂ（ｐ_１）・Ｂ（ｐ_３）・Ｂ（ｐ_５）＝０
（iv）’ Ｂ（ｐ_３）・Ｂ（ｐ_５）・Ｂ（ｐ_７）＝０
この細線処理は、一定の線幅が得られるまで回数を指定して行うことができ、又は上に述べたように、終局的には線幅１の図形が得られるまで繰り返すことができるが、一定の線幅が得られるまで回数を指定して行うことが好ましい。
【００４２】
例えば、図５の文字Ｅの画像について細線化を行なうと、図（ａ）は原画像、図（ｂ）はステップ１の処理を１回行なった画像、図（ｃ）はステップ２の処理を１回行なった画像、図（ｄ）は細線化された画像である。そして、本発明における画像細線化操作は、さらに、形成された細線の始端部（p1点）及び終端部（p2点）をラベリングし、これら両端部（p1点）（p2点）と連結部分の該特徴データとを含む標準骨格線データ(MSLD)を形成するものであることが好ましい。
【００４３】
また、この細線化処理に付随して、形成された各細線の始端部（p1点）及び終端部（p2点）をラベリングし、これら両端部（p1点）（p2点）と連結部分の該特徴データを得て、これを細線化データに加え、回路画像の細線化された標準骨格線データ(MSLD)を形成する。
【００４４】
〈３．膨張化〉
通常、取り込まれた基板画像と細線化したフィルム画像の標準骨格線データを比較させた場合、画像（スキャナ等）の読み取り位置精度などの問題から、基板画像と標準骨格線データが正確に重ならない。そのため、画像（スキャナ等）の読み取り精度による位置のズレを数ドット単位で無効化するために、フィルム画像の幾何学的特徴を損なわないように、元のパターン線幅を膨張させた画像を用意し、比較時には膨張画像の中にある基板画像に対しては位置のズレを無視するようにしている。また、画像の膨張化の膨張回数をドット単位の指定数により処理するようにする。また膨張時に隣接する図形要素とつながらないように線間が狭い場合、この部分は最低幅１ドットのギャップを保証するように設定する。この最低幅は、無論場合によってはドットより大きいギャップとすることができる。
【００４５】
本発明においては、オリジナルフィルムの画像の読取りにより取り込まれ細線化したフィルム画像の標準骨格線データと基板上の配線パターンとを比較させた場合、画像やパターンの（スキャナ等による）読み取り位置精度などの問題から、基板パターンと標準骨格線データが正確に重ならない。そのため、画像（スキャナ等）の読み取り精度による位置のズレを数ドット単位で無効化するためにフィルム画像の幾何学的特徴を損なわないように、元のパターン線幅を膨張させた画像を用意し、比較時には膨張画像の中にある基板画像に対しては位置のズレを無視するようにしている。また、画像の膨張化の膨張回数をドット単位の指定数により処理するようにする。
【００４６】
また、膨張時に隣接する図形要素とつながらないように線間が狭い場合、この部分は最低幅１ドットのギャップを保証するように設定して標準膨張化画像データ(MEID)を得る。この標準膨張化画像データ(MEID)を得るための膨張化操作におけるフィルム画像の幾何学的特徴には、例えばＬ型曲点、Ｔ型交点、Ｙ型交点、端部太点、湾曲部位、ショルダー部、Ｘ型交点のような画像の幾何学的特徴が含まれる。
【００４７】
〈４．画像合成〉
細線化した標準骨格線データ(MSLD)と、標準膨張化画像データ(MEID)を合成させる。このとき、細線化した標準骨格線データ(MSLD)を指定色（例えば赤色）に変換してから標準膨張化画像データ（例えば白色）と重ね合わせる。又は細線化した画像と、膨張化画像（背景黒色、膨張部白色）を重ね合わせる。このとき、細線化したデータは別の色（背景黒色、細線部赤）に変換してから重ね合わせる。こうすることで最低保証すべき導通データと太り、細りと誤差を考慮した膨張画像を同一画像上で扱うことができる。合成された画像は、１画素＝１バイト（８ビット）の色合いに変換する。この合成画像のデータが複合標準画像データ(CMID)である。
【００４８】
〈５．基準点検出〉
また、この複合標準画像データ(CMID)に基いて、オリジナル画像を、その四隅から画像要素を含む各小領域に分割し、分割された各小領域毎に、それらの中に含まれるそれぞれの画像要素の中心点を、基準点データとして形成する。
基準点は、フィルムデータの四隅より指定範囲内の図形要素を分離し、そこに含まれる一定範囲内の円形度（指定可能で通常０．８５〜１．２程度）又は、中心検出誤差の少ない形状を有する画像要素の中心点を検出する。本発明において、基準点は画像と配線パターンを重ね合わせるための回転方向及び水平垂直方向、伸縮を合わせるのに用いる。各基準点を結ぶ範囲内を検査対象範囲とするものである。中心点と関連する図形中の特徴点はつぎのようなものがあり、求める方法には色々のものがある。
【００４９】
即ち、図６に示されるような２値画像の特徴点として、
（ａ）対象物の面積、
（ｂ）対象物の周囲長、
（ｃ）重心Ｇからエッジまでの距離の最大値Ｒ_ｍａｘ並びに最小値Ｒ_ｍｉｎ、
（ｄ）重心Ｇからエッジまでの距離の全周についての平均値Ｒ_ｍｅａｎ
（ｅ）最大半径Ｒ_ｍａｘの軸並びに最小半径Ｒ_ｍｉｎの軸と、Ｘ軸（基準軸）とのなす角ψ_Ｒｍａｘ並びにψ_Ｒｍｉｎ，、
（ｆ）慣性等価楕円の長径ｄ_ｅｍａｘ並びに短径ｄ_ｅｍｉｎ、
（ｇ）慣性主軸Ｘと軸のなす角度ψ_ｍｍ、
（ｈ）慣性主軸に並行な外接長方形の長辺Ｌ_Ｒｍ並びに短辺Ｌ_Ｒｓの長さ、
（ｉ）対象物の重心の位置（Ｘ_Ｇ，Ｙ_Ｇ）、
（ｊ）対象物のＸ軸並びにＹ軸への射影の最大値Ｌ_ＰＸｍ並びにＬ_ＰＹｍ、最小値Ｌ_ＰＸｓ並びにＬ_ＰＹｓ、
（ｋ）対象物内の孔の数と面積の和）、が挙げられる。
【００５０】
円形度は、対象物領域の形状を知るのに用いられ、面積Ａ、周囲長Ｌを用いて次のように定義されている。
【００５１】
【数２】

領域が円形であればＲは１（最大）になり、円形からずれるに従ってこの値は１より小さくなる（たとえば、正方形では０．７８５）。
本発明においては、それぞれの画像要素の中心点を、各画像要素の円形度に基いて検出することにより該各小領域中の画像要素の基準点を、基準点データ(SPOD)として形成する。
そして、前記複合標準画像データ(CMID)に、基準点データ(SPOD)をマージして合成基準画像データ(CMSID)を形成する。
これら［検査標準データの作成］操作は、図１７に示されるように要約される。
【００５２】
［被検データの作成］
〈６．被検基板配線パターン読取り〉
本発明における（Ｂ）の被検基板配線パターンデータ(SVCD)は、被検プリント基板４隅に、配線パターン情報を囲繞するに必要充分な広さの長方形を形成する４つの位置を４基準点として、被検基板の配線パターン情報を読み取って２値データ(BCPD)化することにより形成される。この配線パターン情報を読み取って２値データ(BCPD)する際には、どの個所に生じるか不確定で大きさも不定のノイズ除去のため、読取られた配線パターンの（ｆ１）メディアンフイルタ処理、（ｆ２）平滑化フイルタ処理、（ｆ３）孤立雑音除去フイルタ処理を含むことが好ましい。
【００５３】
〈７．平滑化フイルタ処理〉
１枚の雑音のある画像からの雑音消去の場合、ランダムに雑音により隠されてしまった画素の真の値は、決して知ることはできないが、ただ、雑音が邪魔にならないようにするという目的のためには、視覚的に目立たなくすればよい。
雑音の乗った画像を拡大図として、図７を見てわかることは、雑音の濃度と、その周りの濃度に急激な濃度差があることである。また、急激な濃度差があるから目障りなわけであり、この雑音の性質を利用して雑音除去を行う手法を、一般に平滑化（smoothing）と呼んでいる。ただし、目的画像のエッジの部分も急激な濃度差があるので、このエッジ部分と雑音部分をいかにして分離して雑音だけを除去するかが、平滑化の腕の見せどころとなる。
移動平均法は、最も簡単な雑音除去法であり、これは、図８のように、ある画素の周辺の３×３の画素の平均値をその画素の値と置き替える手法である。要するに画像をぼかしてしまえば、細かい雑音は見えなくなるという理屈を用いたものである。この方法では、雑音だろうとエッジだろうとおかまいなしにぼかしてしまうので、雑音を除去できても、目的の画像はぼけてしまうことがある。
【００５４】
〈８．メディアンフイルタ処理〉
そこで、なるべく目的画像のエッジをぼかさずに雑音を除去するように考えられたのが、メディアン・フィルタという有名な手法であり、本発明においてもこれを用いると有利である。
図９のような濃度を持った画像があるとする。今、丸で囲んだ画素の値を求めるために、３×３の領域内（実線で囲んだ領域）の九つの画素の濃度を調べ、それを小さい順に並べると次のようになる。
２ ２ ３ ３ ▲４▼ ４ ４ ５ １０
このときの中央の値（これをメディアンという）、この場合は全部で九つなので左から５番目の濃度４が求める画素の濃度となる。この１０という画素は雑音として入ったものとすると、確かに雑音が除かれたことになる。これは、周りと比べて極端に濃度の違うものは、大きさの順に並べたとき、左はじか右はじに集まってしまい、中央値として選択されないからである。このように、メディアン・フィルタとは、ある画素の周辺の領域内の画素の濃度の中央値を求め、それを目的の画素の濃度とする処理である。では次に右隣の画素はどうなるかについて、点線で囲った領域内の画素を調べる。
２ ３ ３ ４ ▲４▼ ４ ４ ５ １０
中央値は４となり、本当は３なのに４になってしまいまった。これが処理によって被った被害であるが、しかし、視覚的にはあまり分らないことが多い。問題はエッジ部分が保存されるかどうかであるが、図１０（ａ）はエッジのある画像で、丸で囲んだ画素を順に求めていくと、同図（ｂ）のようになり、この場合は完全にエッジが保存されることが分る。移動平均では、雑音成分も平均の計算の中に入るので、出力は雑音の影響を受けるが、メディアン・フィルタでは、雑音成分は選択されにくいので出力にあまり影響を及ぼさない。したがって、同じ３×３で比較すると、メディアン・フィルタのほうが雑音除去能力においても勝っている。実際の画像でその効果を見ると、図１１は、雑音のある画像をメディアン・フィルタと移動平均法でそれぞれ処理した結果を示しており、メディアン・フィルタが、雑音の除去においても、エッジの保存においても非常に優れた手法であることが判る。計算時間は、メディアン・フィルタのほうが、移動平均の５倍ほど多くかかる。
【００５５】
〈９．孤立雑音除去フイルタ処理〉
領域抽出も、エッジ抽出も、出力画像は２値画像である。これらの処理を行う前に、平滑化によってあらかじめ画像の雑音を除去しておく前処理が大切であり、以後の処理をやりやすくする。しかしそれでも、出力２値画像に邪魔な雑音があるときは、後処理でこれを除いてやらなければならない。２値画像の雑音は、図１２のようなもので、ごま塩雑音と呼ばれているものである。この雑音も、もちろんメディアン・フィルタで除去できるが、２値であることを利用した、膨脹・収縮と呼ばれる処理によって本発明において除去することができる。この場合の膨脹とは、ある画素の近傍に一つでも１があればその画素を１に、そのほかは０にする処理であり、収縮とは、ある画素の近傍に一つでも０があればその画素を０に、そのほかは１にする処理である。この処理を膨脹→収縮と作用させると、結果の画像は膨脹で太って、収縮でやせて、結果的にはほとんど変わらないが、黒い孤立した雑音が膨脹のときに取り除かれ、逆に、収縮→膨脹と作用させると、白い孤立した雑音を収縮のときに取り除くことができる。
〈9a．反射光ノイズ除去処理〉
この雑音除去処理には、反射光に基くノイズ除去処理をも含む。この反射光ノイズ除去操作は、使用する光源の性質、被検基板表面の材料の分光的性質、３色分解する際に用いられる色フイルタの分光特性、及びカラーモニタの色表示特性により、被検基板表面の配線パターンと背景領域間の色相の偏りを超えて反射光ノイズとして認識される部分を消去する操作である。本発明において具体的には、被検基板パターンの各ＲＧＢの３色の成分について、（ｉ）赤色のＲ色成分については、検出されたいずれの濃度の赤色画素も、後に詳細に説明するようなヒストグラムを用い、このヒストグラムにおける濃度階調、例えば256の段階の濃度階調のうちの最大濃度値に属する画素集合部分の高濃度画素に変換し、(ii) 緑色のＧ色成分については、検出されたいずれの濃度の緑色画素も、該ヒストグラムにおける256の段階の濃度階調のうちの中間濃度以上の域で最大の画素集団が属する濃度に緑色画素を変換する一方、(iii)青色のＢ色成分については、検出されたいずれの濃度の青色画素も、該ヒストグラムにおける256の段階の濃度階調のうちの最低濃度値に属する画素集合の濃度値の低濃度画素に変換する。この画素濃度の変換処理をすることにより、主に配線パターンを構成する要素としてのＲ色画素成分情報をカットすることなく、背景域の非配線情報の主因たるＢ色成分情報を集中的にカットし、Ｒ＋Ｇ＋Ｂ＝白色、で表わされる白色バランスからＢ色画素成分を除去して色バランスを崩すことで白色をなくすることによって、白色光として認識される反射ノイズを除去することができる。
本発明においては、濃度階調は２５６の段階に限られることなく、コンピュータによる画像処理に適した他の数の段階、例えば６４段階であっても無論よい。しかし、理解を容易にするため、以下の説明では２５６段階の場合について記載する。
【００５６】
〈１０．閾値によるＲＧＢ成分分別のための濃度算出〉
本発明においては、被検基板配線パターンのデータ(SVCD)は、基板パターンのＲＧＢの３色の成分についてそれぞれ、算出した配線パターン濃度分布と、各成分基板の非パターン部分である背景部分について算出した濃度分布とを対比させることにより、ＲＧＢの各成分の基板の配線パターンを、背景部分から円滑に分別するため、閾値により算出された濃度であることが好ましい。すなわち、現在、基板画像はＲＧＢ各色８ビット（０〜２５６）の濃度値を持って取り込まれている。ここでは、基板上の対象物（回路パターン）とその他背景は比較的濃度差が大きいので、これを的確に分離できるように判別閾値選定法（例えば大津展之：判定及び最小２条基準に基づく自動閾値選定法、電子通信学会論文誌参照）を用いてＲＧＢ各色に対して濃度分散が最大になる濃度値を求め、得られた各色の濃度値に対してパターンを示す色が強調されるように重み付けした値を二値化の閾値として算出することができる。
【００５７】
本発明における、被検基板パターンの読み取られたＲＧＢの３色の成分についてそれぞれ、背景部分から分別するための濃度閾値を算出する分別用濃度閾値検出操作は、具体的には、それぞれの色を構成する全画素を仕分けして、濃度階調、例えば256の段階の濃度階調うちのいずれの濃度に属するかを示すヒストグラムを自動的に作成し、このヒストグラムにおいて或る濃度域に属する画素集合体と他濃度域に属する他の画素集合体との間の存在画素が少ない谷間の濃度を見い出し、この谷間濃度に基いて濃度閾値を検出する操作である。
【００５８】
図３０に示されるように、赤色、緑色、青色についてそれぞれ例えば２５６階調の濃度を横軸にとり、この各階調濃度毎に画素数を縦軸にとってヒストグラフ化し、例えばグラフに谷間が明瞭に出現してないときには、各種グラフィック手法を用いて明瞭化すると、赤色では２００番目の濃度が、画素が集合している部分の最大濃度値であるから、他の赤色画素をも、この２００の番目の濃度に変換する。また、緑色では２５６階調の濃度のうち中間濃度以上の域で最大の画素集団が属する濃度は１８１番目の濃度であるから、他の緑色画素をも、この２００の番目の濃度に変換する。さらに、青色の画素は常時、０値の濃度のものに変換する。
【００５９】
ヒストグラムについて説明を続けると、例えば図１３には、「絵」という文字が他の被写体と共に写っているが、この中から文字の部分だけを抜き出しため、自動選定された閾値を用いることができる。閾値処理は、入力画面の各画素について、明るさがある一定の閾値以上の場合には、対応する出力画像の画素の値を１とし、それ以外の場合には、０にするものであり、式で示すと、
【００６０】
【数３】

のようになる。ここで、ｆ（ｘ，ｙ），ｇ（ｘ，ｙ）は、それぞれ処理前、処理後の画像の（ｘ，ｙ）の場所にある画素の濃度値を、ｔは閾値を示す。図１３の画像の場合は、文字のほうが暗くて、背景が明るいので、上式を適用すると、背景が抜き出されてしまうから、閾値より小さいものを抜き出す次式を使用する。このような、濃度閾値の算出は、本発明においては判別閾値選定法により自動的に行なうことができる。
【００６１】
【数４】

閾値処理を行なうプログラム例を、つぎのリストに示す。
【００６２】
【表１】

【００６３】
このリスト中のｍｏｄｅは、ｍｏｄｅ＝１のとき［式３］が選択され、ｍｏｄｅ＝２のとき［式４］が選択される。また、閾値の処理後の値も、１と０ではなくＨＩＧＨとＬＯＷになっている。原画像と同じ表示プログラムを使用する場合は、ＨＩＧＨ＝２５５、ＬＯＷ＝０とする。
このプログラムで、閾値処理した例を図１４に示す。閾値の与え方により、ずいぶん抜け具合が変化する。この例からも明らかなように、大きすぎる閾値は、余分なものまで取り出し、一方、小さすぎる閾値は、必要なものも切り捨ててしまう。
【００６４】
どのようにして、最適な閾値を決めるべきかについて記述すると、抽出したい物体と背景とは明るさが異なっているはずである。でなければ、人間の目でも識別できなくなる。そこで、抽出したい部分と切り捨てたい部分の明るさを、何点か位置を指定してチェックして見る。図１３の例の文字の部分は１４０ぐらいで、背景の部分は１６０ぐらいである。したがって、閾値を１５０ぐらいに選べば、文字と背景が分離できそうである。しかし、このように１点ごとに画素の明るさをチェックしないで、もう少しスマートな方法として、ヒストグラム（頻度分布）を使う方法がある。ヒストグラムとは、図１５に示すように濃度値ｉの画素何個あるのかの頻度を示すものであり、上記リストに示すプログラムにより求めることができる。上記リストだけでは、ヒストグラムを見ることができないので、ヒストグラムを印字するプログラムによりヒストグラムを打ち出し、あるいはヒストグラムを画像化するプログラムにより印刷する。図１６に示されるような一目瞭然のヒストグラムを得ることができる。
このヒストグラムによれば、濃度値１４０前後の山が文字部分の画素に相当し、濃度値１６０前後の山が背景の画素に相当する。この二つの山の境目、すなわち谷のところを閾値とすればうまく分離できそうにも思えるが、しかし、このヒストグラムでは、でこぼこが激しく、谷がよく分らない。そこで、谷を見つけやすくする方法として、ヒストグラム上で近傍同士を平均化してでこぼこを減らしてヒストグラム表示することがよく行われており、本発明においてもこの手法を好適に用いることができる。このようにすると、谷が見つけられやすくなり、濃度値１４８のあたりであると判断できる。照明が均一でない、文字がぼけているなどの理由から完全には抜き出せていないが、他の閾値の場合に比べかなりうまく切り出すことができており、このようにヒストグラムの谷を使って、閾値を決定する方法はモード法と呼ばれている。
閾値の決め方には、モード法のほかにも、Ｐタイル法、判別分析法、可変閾値法などがあり、Ｐタイル法は、全画像に占める物体の割合がわかっているとき（例えばｐ％とする）、ヒストグラム上で全度数のうち、暗いほうから（あるいは明るいほうから）ｐ％のところを閾値とする方法であり、判別分析法は、ヒストグラムを物体と背景の二つの集団に分けたときに二つの集団の統計量が異なるように閾値を決める方法である。可変閾値法は、背景が一様な明るさでない場合に有効な方法で、場所ごとに閾値を変化させる方法である。
【００６５】
〈１１．基準点検出〉
基準点の検索は、フィルムの場合と同様に四隅から指定範囲内の画像領域を切り出した後、二値化処理を行ない、そこに含まれる一定範囲内の円形度、中心誤差の少ない形状を有する画像要素の中心点を基準点として検出する。この基準点は、フィルム複製物とし、そのためパターン形成時にフィルムと同位置に形成される。この、基板配線パターン読取〜基準点検出までの処理で合成被検データを得る。
これら［被検データの作成］操作は、図１８に概要が示される。
【００６６】
［検査処理］
検査のための画像処理は二段階で行なう。第一段階は合成被検データ（被検基板配線画像データ(SVCD)）と合成基準データ（標準仮想導通データ(MVCD)）とを重ね合わせる処理、第二段階は、重ね合わせた画像から有効な相違点を抽出し、断線、ショート、太り、細りを判別する処理である。
【００６７】
（１）二値化と重ね合わせ
〈１２．二値化処理〉
基板画像を二値化してパターン図形のみを有効図形として切り出す。その際、基板画像に含まれる微小ノイズ、スキャン時の光の当たり具合による局所的な色合いの違い、画像のエッジ部分のぼやけ（境界画素の不明瞭な個所）等の影響を低減するために二値化前処理として、メディアンフィルタ、平均化フィルタ、孤立雑音除去フィルタ等をユーザーの選択によって適用する。実際ノイズの多い画像では、二値化後の画像の欠落や微小なノイズが多いため、正確な断線、ショート、太り、細りの判別が難しくなる。
二値化する前の画像に対して上記のような前処理を適用することによって、パターンのエッジを極力保存しながら、ノイズの影響を低減することが必要である。二値化した画像は、１画素＝１バイト（８ビット）の画像に変換し、パターンの存在する個所は、青色で表示する。
【００６８】
〈１３．画像重ね合わせ〉
前記作成した被合成検査データと合成基準データを重ね合わせた画像を作成する。この際、先に検出した基準点の座標を用いて、被合成検査データと合成基準データの回転角度、伸縮率を算出し、これを考慮しながら画像を重ね合わせる。具体的には、基板上の注目画素の位置に存在するフィルム画像の画素の排他論理輪をとり、その値を重ね合わせた画像として出力する。こうすることで、二値化した基板画像の余分な部分に関しては、二値化したときの青色がそのまま出力され、不足する部分に関しては合成基準データ作成時に色づけした赤色がそのまま出力された画像が得られる。それ以外の部分である黒色部分は検査対象外領域となり、画像同士が重なり合った部分は、もとの赤／青以外の色で表示される。さらに上記の処理後、基準点の座標を用いて各画像全体を小領域に分割し（例えば、６４ドット×６４ドットに分割）、その各小領域を回転、伸縮させ画像の重ね合わせを行なう。具体的には、分割した小領域に縦横数画素分のオーバーラップ画素分（例えば、３ドット×３ドット）をもたせ、その画素分の範囲で回転、伸縮（例えば、３×３ドットの範囲を３６回幾何学的補正）させ、各小領域の重なり率を算出し、一番高率の個所を選択し固定する。これを各小領域毎に行なう。このような、２段階の重ね合わせを処理することで、全体画像での大きな重ね合わせズレを各小領域に分散し、さらに回転、伸縮補正することで、各小領域でのズレ幅が分散し小さくなり、重ね合わせ処理の高速化もできるようになり、精度、速度が各段に向上する。
【００６９】
（２）相違点の検出（検査処理の第２段階である）
本発明において重ね合せ、比較操作する際には、回路オリジナルの標準仮想導通データと、プリント基板の被検基板配線画像データのうちの少なくとも１方を、部分的回転又は傾け補正、伸縮補正及び／又は座標確認調整して重ね合わせることが好ましく、特に回路オリジナルの標準仮想導通データを部分的回転又は傾け補正、伸縮補正及び／又は座標確認調整して重ね合わせることが好ましい。
【００７０】
〈１４．断線、細りの検出〉
重ね合わせた画像は、前述のとおり不足部分は赤、余分な領域は青で表示されている。ここで、不足部分は断線、細りとして検出するのであるが、赤色で表示されているものはフィルム画像に線化を施したものであるため、部分的には線幅１画素になっている。このため画像が正確に重ね合わさった場合、断線は１画素でも赤色が存在すれば、断線として判定できる。
【００７１】
〈１５．短絡、太り検出〉
次に、ショート、太りは、基板画像のズレ等を考慮するためにフィルム画像に膨張処理を加えて重ね合わせている。フィルム画像は膨張時に隣接する図形要素とつながってしまわないようにしているので、線間が狭い場合、この部分は最低幅１ドットの開きしか無くなっている。そこで、ショート、太り検出には２通りの判断基準を設定する。
【００７２】
本発明において、例えば（ｉ）前記（Ｄ）の被検基板上の配線における断線は、前記（Ａ）中の前記（ａ）による２値データ(BCID)と、前記（Ｂ）の被検基板配線パターンデータとを重ね合わせのための比較をしたとき、該（ａ）による２値データ(BCID)の余剰個所として検出され、
（ii) 前記（Ｅ）の被検基板上の配線における短絡は、前記（Ａ）中の前記（ａ）による２値データ(BCID)と、前記（Ｂ）の被検基板配線パターンデータとを重ね合わせのための比較をしたとき、該（ａ）による２値データ(BCID)の欠損個所として検出され、
（iii) 前記（Ｆ）の被検基板上の配線における潜在的断線は、前記（Ａ）中の前記（ｂ）による標準骨格線データ(MSLD)と、前記（Ｂ）の被検基板配線パターンデータとを重ね合わせのための比較をしたとき、該（ｂ）による標準骨格線データ(MSLD)の余剰個所として検出され、
（iv) 前記（Ｇ）の被検基板上の配線における潜在的短絡は、前記（Ａ）中の前記（ｃ）による標準膨張画像データ(MEID)、前記（Ｂ）の被検基板配線パターンデータとを重ね合わせのための比較をしたとき、該（ｃ）による標準膨張画像データ(MEID)の欠損個所として検出される。
【００７３】
先に説明したように、本発明においては、２値データ(BCID)化された画素のうち背景領域との境界に位置する画像画素を、縦行３画素×横列３画素の二次元展開画素の行列の中心としたときに、周囲の８画素のうち背景領域画素はいずれであり画像画素はいずれであるかに基いて非画像画素への書き替え操作が行なわれるが、このような操作過程を、ＲＧＢの３色で表示できるモニタを用いた場合について具体的に説明すると、つぎの（Ｉ）のように、３×３ドットの画像領域で、中心画素を含む縦、横、斜めの連続線上に青色画素が３つ並んで存在する（図１９）か、或いは（II）のように、２×２ドットの画像領域で、その全てが青色画素であるか（図２０）の基準により、膨張処理による線幅１のギャップ内に存在するショート、太りに相当する部分と、広い空間上に存在する不良領域（フィルムに存在しない個所）の全てを的確に選別、検出できる。
図２１、図２２には、このような［検査処理］操作の概要が示される。
【００７４】
【発明の実施の形態】
以下、本発明のより具体的な実施の形態例について説明する。
図２３、図２４は、本発明の１具体例中の(i)オリジナル読取処理、(ii)フイルム画像の細線化処理、(iii)フイルム画像の膨張化処理、(iv)フイルム画像の細線化データと膨張化データの画像合成処理、(v)合成基準データの基準点算出処理からなる（Ａ）の標準仮想導通データ(MVCD)作成を説明し、図２５、図２６は、(vi)基板画像読取処理、(vii)基板画像の二値化前処理としてのＲＧＢの３色フイルタを適用し孤立ノイズ除去、平滑化フイルタ、メディアンフイルタによる処理、(viii)基板画像の二値化処理、(ix)被検合成配線データの基準点算出からなる（Ｂ）の被検基板配線画像データ(SVCD)作成を説明し、図２７、図２８、図２９は、(Ｘ）合成基準データと被検配線データの画像重ね合せ処理、(xi)相違点の検出処理からなる（Ｃ）〜（Ｇ）の比較判定操作を説明するものである。
【００７５】
〈合成基準データの作成〉
（ｉ）この例におけるフィルム画像読み取りにおいては、白黒の二値化データとして取り込みを行なう。この例では、スキャナとＰＣ間の送受はＴＷＡＩＮ規格に沿って行なっている。最上段の図は、端部を拡大した配線のフィルム原画像を示す。
(ii) フィルム画像の細線化処理においては、線化の回数をドット単位で指定して、指定回数の線化処理を行なう。線化されたドットの連結要素は、断線、細りの最低保証幅領域の判定基準になる。この図の例は、細線化回数１後の状態を示す。
(iii) においては、膨張化の回数をドット単位で指定して、指定回数の膨張化処理を行なう。膨張化されたドットの連結要素は、画像比較時の位置ズレの無効化領域と太りの最大保証幅領域とショートの判定基準になる。この図の例は、膨張化回数１後の状態を示す。
(iV) フィルム画像の細線化データと膨張化データの画像合成処理により画像合成されたデータは、ドット単位の導通保証、細り、太りの判定基準となる合成基準データである。
(ｖ) この例においては、画像合成処理における合成基準データの基準点として、オリジナル画像の各隅に４箇所算出する。この図には色が示されてないが、この例では(ii)のフィルム画像の細線化処理データの結果は、赤色で、(iii)の膨張化の結果は白色で、それぞれ表示されているので、識別が容易となる。
図には確かに表示されていないが、本発明においては、細線化の処理速度を向上させるため、輪郭全体から１画素になるまで線化を行なわず、最低保証幅まで削ることで画素を削る回数を少なくすることができる。また、最低保証幅まで線化された線化データは、導通ルートの保証と細りの最低保証幅を同時に検査することができる。さらに、膨張化処理は、画像重ね合わせのとき、ズレによる検査誤認をなくすための無効範囲と最大保証幅を得ることができる。
【００７６】
〈合成被検パターンデータの作成〉
(Vi) 基板画像読取りにおいては、２４ビットカラーで、フィルム画像読み取り時同様の解像度で行なう。現在、データの送受はＴＷＡＩＮ規格による。
(Vii) 基板画像の二値化前処理として、取り込んだ基板画像に、孤立雑音除去フィルタ、平滑化フィルタ、メディアンフィルタの３種類のフィルタを選択により適用する。
(Viii) 基板画像の二値化処理は、判別閾値選定方を用いて、回路パターン部が強調されるような値で、二値化の閾値として算出する。この図には色が示されてないが、この例では回路パターン部は青色で、背景領域は黒色で、それぞれ表示されているので、識別が容易となる。得られたものを被合成検査データとする。孤立雑音除去は、二値化処理後、表示の直前に連結していない青色の画素を雑音と考え、１ドットの青色画素を除去する方法で行なった。また、３種類のフィルタを用いる場合の使用順位は、１．メディアンフィルタ、２．平滑化フィルタ３．孤立雑音除去フィルタの順であった。本発明においては、このように、各フィルタは、必要に応じて選択できるようにしているが、必要なフィルタのみを使用することで二値化の処理速度を上げることができる。
(iｘ) 被合成検査データの基準点算出において、合成被検データは、不良個所検出の精度（断線、ショート、太り、細り個所の形状、不要な付着物識別、パターン形成部材と基材の密着不良個所の色合いによる識別）を向上させるため、検査基板の生画像にできるだけ近い二値化画像を得ることが重要である。そのため、前記の基板画像の二値化画像処理以外（例：細線化、膨張化等）は、施さないことでより正確な情報を得ることができる。
〈両データの重ね合せ〉
(ｘ) さらに、合成基準データと被合成検査データの画像重ね合わせ処理がおこなわれ、
(ｘi) 相違点の検出が行なわれた。良好な検査結果が得られた。
【００７７】
この例における検査前処理と検査時間（基板検査）はつぎのとおりであった。
（ｉ）検査範囲
Ｂ５サイズ：約１８０×２５０ｍｍ
取り込みデータ：約５０ＭＢ
（Ａ）検査前処理（スキャナによる取り込み−フィルター適用−二値化処理）＝約５５ｓｅｃ
（Ｂ）検査実行・表示＝約２５ｓｅｃ
（Ｃ）検査総合時間：５５＋２５＝約７５ｓｅｃ
（ii)検査範囲
Ａ４サイズ：約２１０×３００ｍｍ
取り込みデータ：約１００ＭＢ
（Ａ）検査前処理（スキャナによる取り込み−フィルター適用−二値化処理）＝約１０５ｓｅｃ
（Ｂ）検査実行・表示＝約２５ｓｅｃ
（Ｃ）検査総合時間：１０５＋２５＝約１３０ｓｅｃ
（iii)検査範囲
Ａ３サイズ：約４２０×３００ｍｍ
取り込みデータ：約１８０ＭＢ
（Ａ）検査前処理（スキャナによる取り込み−フィルター適用−二値化処理）＝約１８０ｓｅｃ
（Ｂ）検査実行・表示＝約３０ｓｅｃ
（Ｃ）検査総合時間：１８０＋３０＝約２１０ｓｅｃ
基準データ作成時間（フィルム）
（ｉ)フィルム基準範囲
Ｂ５サイズ、取り込みデータ約２．５ＭＢ
（細りの回数 ２、太りの回数 ３の場合）
基準データ作成時間（スキャナによる取り込み−線化・膨張化処理−表示）＝約８０ｓｅｃ
(ii)フィルム基準範囲
Ａ４サイズ、取り込みデータ約４．５ＭＢ
（細りの回数 ２、太りの回数 ３の場合）
基準データ作成時間（スキャナによる取り込み−線化・膨張化処理−表示）＝約１２０ｓｅｃ
(iii)フィルム基準範囲
Ａ３サイズ、取り込みデータ約６．５ＭＢ
（細りの回数 ２、太りの回数 ３の場合）
基準データ作成時間（スキャナによる取り込み−線化・膨張化処理−表示）＝約２００ｓｅｃ
【００７８】
図３１は、本発明によるプリント配線基板の検査方法及び検査装置の概要を説明するためのものである。同図において、シャーテン台（１０ａ）上のオリジナル画像フイルム（１０）の回路画像（１１）は、画像の光学読取部（１３ａ）、画像情報処理部（１３１）を有する画像情報読取手段としてのスキャナ（１３）により読み取られ、コンピュータ（１５）のＣＰＵ（１６）を介して２値データ化されＲＡＭ（１）及びＲＡＭ（２）からなるメモリ（１８）中に格納される。画像情報読取手段としてのスキャナ（１３）は、画像情報処理部（１３１）を有し、画像情報処理部（１３１）は、読み取られた画像情報をＲＧＢの３色情報に色分解するフイルタ（１３１ａ）、ＲＧＢの３色情報のためのシフトレジスタ（１３１ｂ）、データラッチ部（１３１ｃ）、デコーダ（１３１ｄ）、これらを制御し、光学読取部（１３ａ）の駆動を制御する制御装置（１３１ｅ）、そのための制御プログラムを呼出可能に格納するメモリ（１３１ｆ）、デコーダ（１３１ｄ）から払い出された画像情報を一時保存するメモリ（１３１ｇ）を有する。
図中ではメモリ（１８）がＲＡＭ（１）及びＲＡＭ（２）に分けて示されているが、無論、２つである必要はなく、２５６階調を有する３色の多数のカラー画素をデータとし取り扱い、かつこれら多数のデータを次々に変換するための大規模な計算、及びその繰り返し計算回数の多さに鑑みて、それに適した容量の格納呼出し自在なものであればよい。メモリ（１９）として記載された主に画像処理プログラムを読出可能に収納せるＲＯＭ（１）及びＲＯＭ（２）の場合も同様である。
【００７９】
したがって、この検査装置は、（Ａ）プリント回路基板作成のためのフォトレジスト面への像露光に用いるオリジナル画像フイルムの回路画像の標準仮想導通データ(MVCD)と、（Ｂ）検査されるプリント基板の被検基板配線パターンデータ(SVCD)との（Ｃ）比較により、該被検基板上の配線における（Ｄ）断線及び（Ｅ）短絡を判定し、かつ該配線における（Ｆ）潜在的断線及び（Ｇ）潜在的短絡を判定するプリント配線基板の検査装置であって、
（Ａ）のオリジナル画像フイルムの回路画像の標準仮想導通データ(MVCD)のための、（ａ）プリント回路基板作成のためのフォトレジスト面への像露光に用いるオリジナル画像フイルムの回路画像の画像情報を読み取って２値データ(BCID)化する画像情報読取手段（１３）の他、（ｂ）ＣＰＵ（１６）と、（ｃ）該２値データ(BCID)、前記オリジナルフイルムの回路画像の標準仮想導通データ(MVCD)を呼出自在に格納するメモリ手段（１８）と、（ｄ）前記オリジナルフイルムの回路画像の標準仮想導通データ(MVCD)、該メモリ手段から呼出された２値データ(BCID)に基いて、読み取られた画像の基準点を検出するための基準点検出プログラム、該メモリ手段から呼出された２値データ(BCID)を細線化するための細線化プログラム、膨張化するための膨張化プログラム、細線化された２値データと膨張化された２値データをマージするためのマージプログラムを格納するメモリ手段（１９）と、（ｅ）モニタ手段とを有するものと還元することができる。
前記（ａ）の画像情報読取手段（１３）は、前記プリント基板の被検基板配線画像データ(SVCD)のための、（ｆ）配線パターン情報を読み取って２値データ(BCPD)化する画像情報読取手段である。
前記（ｂ）のＣＰＵ（１６）は、前記オリジナルフイルムの回路画像の標準仮想導通データ(MVCD)と、前記被検基板配線パターンデータ(SVCD)とを比較して被検基板配線の断線及び短絡を判定し、かつ該配線における潜在的断線及び潜在的短絡を判定する比較手段でもある。
前記（ｃ）のメモリ手段（１８）は、読み取られた被検基板パターンの２値データ(BCPD)、前記被検基板配線パターンデータ(SVCD)を前記ＣＰＵに呼出可能に格納するメモリ手段でもある。
前記（ｄ）のメモリ手段（１９）は、読み取られた被検基板パターンのＲＧＢの３色の成分についてそれぞれ、背景部分から分別するための濃度閾値を算出する分別用濃度閾値検出用プログラムを前記ＣＰＵに呼出可能に格納するメモリ手段でもある。
【００８０】
【発明の効果】
以上、詳細かつ具体的な説明から明らかなように、本発明により、プリント基板やＬＳＩ（大規模集積回路）、そのためのフォトマスク複製物における配線パターンの配線欠陥を配線の外観検査で見つけ出して、良、不良判断を容易確実かつ迅速に行なえるだけでなく、目視による外観検査では見い出せない配線の太い細い等に起因する導通欠陥を一定の定量的判断基準を保持しつつ見い出すためのバーチャルコンダクティング（仮想導通）テストをプリント基板の製造過程において行なえる改良されたパターンマッチング検査方法及び検査装置が提供されるという極めて優れた効果が奏される。
【図面の簡単な説明】
【図１】本発明のプリント配線基板のパターン情報の２値化法の概要を示す図である。
【図２】本発明の細線化を経て標準骨格データ形成方法を模式的に示す示す図である。
【図３】本発明における３×３論理マスクを用いた細線化操作の説明図である。
【図４】本発明における３×３論理マスクを用いた細線化操作の説明図である。
【図５】本発明における３×３論理マスクを用いた細線化操作の詳細説明図である。
【図６】本発明における特徴点、基準点検出を説明する図である。
【図７】典型的なノイズを含む画像例を示す図である。
【図８】移動平均法によるノイズ除去処理を示す図である。
【図９】本発明におけるメディアンフイルタ使用によるノイズ除去処理を示す図である。
【図１０】本発明におけるメディアンフイルタ使用によるノイズ除去処理を示す図である。
【図１１】本発明におけるメディアンフイルタと移動平均法の使用例を説明する図である。
【図１２】典型的なごま塩ノイズを含む画像例を示す図である。
【図１３】本発明における処理対象画像の例を概念的に説明する図である。
【図１４】本発明における処理対象画像の閾値処理を概念的に説明する図である。
【図１５】本発明における処理対象画像の閾値のヒストグラムを示す図である。
【図１６】本発明における処理対象画像の閾値の調整済みのヒストグラムを示す図である。
【図１７】本発明における合成基準データの作成プロセスを示す図である。
【図１８】本発明における合成被検データの作成プロセスを示す図である。
【図１９】本発明における被検基板の配線の短絡、太り判定を説明する図である。
【図２０】本発明における被検基板の配線の別の短絡、太り判定を説明する図である。
【図２１】本発明における検査第１段階プロセスを示す図である。
【図２２】本発明における検査第２段階プロセスを示す図である。
【図２３】本発明における合成基準データの作成を説明する図である。
【図２４】本発明における合成基準データの基準点算出を示す図である。
【図２５】本発明における合成被検データの作成を説明する図である。
【図２６】本発明における合成被検データの基準点算出を示す図である。
【図２７】本発明の合成基準データと、合成被検データの重ね合わせによる相違点検出を説明する図である。
【図２８】本発明の合成基準データと、合成被検データの重ね合わせ画像の部分拡大図である。
【図２９】本発明における被検基板の配線の別の短絡、太り判定を説明する図である。
【図３０】本発明における被検基板の反射光ノイズ除去のためのＲＧＢ３色それぞれの濃度基準の目安閾値のためのヒストグラム例である。
【図３１】本発明のプリント配線基板の検査装置の１例の概要を示す図である。
【符号の説明】
１０ オリジナルパターンフィルム
１０ａ シャーテン台
１１ 回路パターン領域（画像領域）
１２ 背景領域
１３ 画像情報読取装置
１３ａ 光学読取部
１５ コンピュータ
１６ ＣＰＵ
１７ モニタ
１８ 記憶装置（メモリ）
１９ 記憶装置（メモリ）
２０ 入力手段（キーボード）
１３ａ 光学読取部
１３１ 画像情報処理部
１３１ａ 色分解フイルタ
１３１ｂ シフトレジスタ
１３１ｃ データラッチ部
１３１ｄ デコーダ
１３１ｅ 制御装置
１３１ｆ メモリ
１３１ｇ メモリ[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an improved inspection method and inspection apparatus for determining whether or not a wiring defect of a printed circuit board is found by visual inspection of the wiring and determining whether or not it is defective. The technique of the present invention is also applicable to an inspection method and an inspection apparatus for determining whether or not a wiring defect of a wiring pattern in an LSI (Large Scale Integrated Circuit) and a photomask replica therefor can be found by visual inspection of the wiring. Can do.
[0002]
[Prior art]
Generally, there are disconnection defects, short circuit defects, thin defects, and fat defects as the main defects in printed circuit boards and LSI wiring patterns. Conventionally, printed circuit boards and LSI circuit patterns and photomask replicas ( Image recognition technology used for defect detection methods in the mechanization and automation of inspections (such as copies of originals) is mainly a design rule that statistically checks specific special parameters that faithfully and prominently show the features of the design image Method (statistical identification method) and pattern matching method (comparison method) that compares whether the pattern is the same pattern as the standard image set in advance or a pattern that is very close. The suitability is determined by the degree of matching with the standard image compared with the image. The pattern to be inspected is a printed wiring board product or a wiring pattern of a printed wiring board being manufactured, and the standard image is selectively image-exposed in the form of a wiring pattern on the photoresist material surface provided on the original printed wiring board. Therefore, it is a wiring image of the original photomask or the second original drawing photomask obtained therefrom.
[0003]
Examples of techniques using the pattern matching method (comparison method) include those described in Japanese Patent Application Laid-Open No. 4-73759, Japanese Patent No. 2502854, Japanese Patent Publication No. 8-7155, and Japanese Patent No. 2502853. The one described in the publication and the one described in the Japanese Patent Publication No. 7-43252 can be cited. As an improvement of such a technique, we first described "(a) a photoresist surface for producing a printed circuit board". An operation of reading the image information of the circuit pattern of the original pattern film used for image exposure to form binary data (BCD), and (b) the circuit pattern based on the binary data (BCD) of the circuit pattern image The standard skeleton line data (MSLD) of the circuit pattern is converted into a delineated line segment by recognizing the start end (p1 point) and end (p2 point) of the formed thin line while thinning the image. And (c) connecting the circuit pattern of the board to be inspected by comparing the standard skeleton line data (MSLD) with the inspected skeleton line data (SSLD) of the printed circuit board to be inspected. (D) Separately, (d) Separate operation for extracting non-standard contour data (MCCD) of circuit pattern from binary data (BCD) of circuit pattern image information formed by the operation of (a), e) Based on the size of the separation width between the outer contour position of the circuit pattern based on the non-standard contour data (MCCD) and the skeleton line position of the circuit pattern based on the binary data (BCD) of the standard skeleton line. Operation to form standard virtual continuity data (MBCD), (f) Printed by comparing the standard virtual continuity data (MVCD) of the circuit pattern and the virtual continuity data (SVCD) to be inspected of the printed circuit board Circuit board virtual continuity test The printed circuit board inspection method "and" inspection apparatus "characterized in that it includes an operation for performing the above-mentioned operations, and a patent application has been filed as Japanese Patent Application No. 2000-168509.
[0004]
We continue further examination, and the above-mentioned “circuit connection inspection of the substrate to be inspected” is an inspection standard data obtained from an image pattern as a standard pattern, that is, an image pattern of a photoresist mask (an image pattern of an exposure film), This is done by comparing the data to be inspected obtained from the circuit pattern created on the substrate to be inspected. However, depending on the inspection conditions, inspection results with a wide range of density and line thickness can be obtained. Has been found to be suitable for classifying.
[0005]
For example, the “virtual continuity inspection of a circuit on a substrate to be inspected” includes standard skeleton data (pilot data) obtained from a photoresist mask pattern (image of an exposure film) and binary wiring data obtained from a substrate to be inspected. This is performed by comparing the wiring pattern data of the board to be inspected set based on the threshold value. The selected image processing operation, such as the selection of the gap width, provides a sufficiently wide inspection result from insufficient continuity to a circuit short, and therefore it is preferable to classify the degree of defects, and It is practically advantageous to use a color monitor and display the read substrate image separately for each of the three RGB components according to the nature and type of the image. There, For this purpose, each of the read-out PCB pattern three color components of RGB, and found that is desirable to calculate the density threshold for the strongest separated from the background portion.
[0006]
In addition, when obtaining both the original image and the wiring pattern of the substrate to be inspected, when reading the image of the original film for exposure and the circuit pattern created on the substrate to be inspected by the scanner, the image capture angle of the scanner, etc. Due to the difference in the conditions at the time of capture, the obtained pattern or image may be distorted, for example, and may not be completely the same shape, and the captured pattern or image may be misaligned on the coordinate axis There may be cases where they are not at the exact same position, or between the two patterns and images, during the manufacturing process or during image reading for inspection, some inevitable misalignment and pattern rotation may occur, and minute fine images and patterns are converted into data When handling, make it possible to perform image processing to compensate for errors caused by these positional coordinate deviations, and incorporate some errors into the tolerance. It was found that preferably be provided.
[0007]
Furthermore, the wiring pattern information created on the board to be inspected contains a lot of noise when read by the scanner, and these noises are different for each board to be inspected, and therefore the original wiring pattern on the board to be inspected. It has been found that smooth and efficient removal of these noises without losing information is more effective than noise removal of image information on an exposure film.
[0008]
In addition, the image of the exposure film and the wiring pattern created on the substrate to be inspected do not have the same size, but have a similar shape, due to various reasons such as differences in optical properties. It has been found that it is preferable to perform an operation correction for a slight allowance that incorporates this error.
[0009]
In addition, for image of exposure film, standard skeleton line data (MSLD) and standard virtual continuity data (MBCD) are smoothly synthesized, and the skeleton line data (SSLD ) And the virtual continuity data (SVCD) to be inspected can be smoothly synthesized to compare the standard data on the exposure film image and the inspected data on the circuit pattern created on the substrate to be inspected. Created a rational method for
[0010]
In addition, when comparing the image of the exposure film with the circuit pattern created on the substrate to be inspected, a color image display technology was used to create a system that could recognize the difference between the two patterns more clearly. .
[0011]
Further, when obtaining both the virtual continuity data (MBCD) and (SVCD), the minimum number of times is obtained on the outer contour of the image of the exposure film and on the outer contour of the circuit pattern created on the substrate to be inspected. A rational method for extracting singular points as pixels from the most effective position was created.
[0012]
[Problems to be solved by the invention]
Therefore, in view of the above circumstances, the object of the present invention is to find wiring defects of wiring patterns in printed circuit boards, LSIs (large-scale integrated circuits), and photomask reproductions therefor by visual inspection of wirings, and to easily judge defects. In addition to being able to perform surely and quickly, a virtual conducting test (virtual continuity) test for finding continuity defects caused by thick and thin wiring that cannot be found by visual inspection while maintaining a certain quantitative judgment standard It is an object of the present invention to provide an improved pattern matching inspection method and inspection apparatus capable of performing the above in the manufacturing process of printed circuit boards.
[0013]
[Means for Solving the Problems]
  The above objects are (1) “(A) Standard virtual continuity data (MVCD) of the circuit image of the original image film used for image exposure on the photoresist surface for producing the printed circuit board, and (B) inspection of the present invention. The (D) disconnection and (E) short circuit in the wiring on the test board are determined by the (C) comparison operation with the test board wiring image data (SVCD) of the printed circuit board to be processed, and (F) A method of inspecting a printed wiring board to determine potential disconnection and (G) potential short circuit,
  (A) The standard virtual continuity data (MVCD) of the circuit image of the original image film (A) includes (a) image information of the circuit image of the original image film used for image exposure on the photoresist surface for creating a printed circuit board. Read binary data (BCID) as pilot line dataCircuit image and(B) the image information reading operationBinary data (BCID) Circuit imageAn image thinning operation for forming a standard skeleton line data (MSLD) of a thinned circuit image by thinning the image, and (c)The binary data (BCID) Circuit imageThe image expansion operation for forming the standard expansion image data (MEID) of the circuit image by performing expansion processing, and (d) the standard expansion image data (MEID) in the standard skeleton line data (MSLD)Overlay synthesisTo form composite standard image data (CMID) of the circuit image, and based on the composite standard image data (CMID),Circuit imageIs divided into small areas including image elements from the four corners, and for each divided small area, each of the image elements included therein is divided.Based on circularitydetectiondidCenter pointAs a reference point,An image reference point detection operation for forming a reference point of an image element in each small region as reference point data (SPOD), and (e) the reference point data (SPOD) in the composite standard image data (CMID) TheOverlay synthesisAnd a standard composite reference image data forming operation for forming composite reference image data (CMSID).
  (B) Test substrate wiringimageThe data (SVCD) is (f) a wiring pattern reading operation for reading the wiring pattern information of the printed circuit board to be inspected and converting it into binary data (BCPD);Binary dataA printed wiring board inspection method characterized by being obtained by a process including a density threshold value detection operation for calculating a density threshold value for classifying each of the three RGB components of a substrate pattern from a background portion. Is achieved.
[0014]
Further, (2) of the present invention, wherein the image reading operation for converting to binary data (BCID) in (a) includes an operation of reading as binary data by a TWAIN compatible scanner. This is achieved by the “printed wiring board inspection method according to the item”.
[0015]
In addition, (3) “the image thinning operation for forming the standard skeleton line data (MSLD) of (b) above” is further performed with the start end (p1 point) and the end (p2 point) of the formed thin line. Item (1) is characterized in that it is labeled to form standard skeletal line data (MSLD) including both end portions (p1 point) (p2 point) and the feature data of the connected portion. This is achieved by “a method for inspecting a printed wiring board”.
[0016]
  According to the present invention, (4) “the image thinning operation for forming the standard skeleton line data (MSLD) of the above (b) is performed by the binary data (BCID)Is a thin lineThis is achieved by the printed wiring board inspection method according to any one of (1) to (3).
[0017]
  In addition, according to the fifth aspect of the present invention, the image thinning operation for forming the standard skeleton line data (MSLD) of (b) is an originalImage filmRead from circuit imageObtainedBased on binary data (BCID), the pixels that form the image area areOne pixel at a timeIt is achieved by the printed wiring board inspection method according to any one of (1) to (4) above, which includes an operation of cutting into a background region.
[0018]
According to the sixth aspect of the present invention, “the operation of trimming the pixels forming the image area into the background area bit by bit from the outside is the background area among the pixels read from the original circuit image and converted into binary data (BCID)”. Is the center of the matrix of two-dimensionally developed pixels of 3 pixels in the vertical row and 3 pixels in the horizontal row, which is the background region pixel and which is the image pixel By repeating the operation of rewriting to a non-image pixel based on whether there is no parallel part of the image pixel-image pixel in contact with each other in the alternate direction along the outer periphery of the pattern, until there is no part where the line width exceeds one pixel This is achieved by the “printed wiring board inspection method according to item (5)”, which includes an operation of repeating in an alternating direction along the outer periphery of the image.
[0019]
In addition, according to the present invention (7), the image expansion operation for forming the standard expansion image data (MEID) of the circuit image of (c) recognizes the geometric feature portion of the image in the original of the circuit. The printed wiring board inspection method according to any one of (1) to (6), wherein the inspection is performed without impairing the geometric characteristics of the image in the original. Is done.
[0020]
In addition, (8) “(d) image reference point detection operation of the present invention, for each divided small area, the center point of each image element included in each of the divided small areas, the circle of each image element. This is achieved by the method for inspecting a printed wiring board according to any one of (1) to (7), characterized in that the operation includes detection based on the degree.
[0021]
Further, (9) “(f) The wiring pattern reading operation for converting the wiring pattern of the substrate to be tested (BCPD) into binary data (BCPD)” is further performed. (F1) Median filter processing of the read wiring pattern, 3. The method for inspecting a printed wiring board according to any one of (1) to (8) above, comprising three processes of f2) smoothing filter process and (f3) isolated noise removal filter process Is achieved.
[0022]
In addition, according to the present invention, (10) “(g) the separation density threshold value detection operation for calculating the density threshold value for the separation of each of the three RGB components of the read substrate pattern from the background portion, By classifying all the pixels that make up a color, a histogram is automatically created to indicate which of the 256 levels of density gradations it belongs to. Items (1) to (1) are operations for finding a density of a valley where there are few pixels between other pixel aggregates belonging to the density range and detecting a density threshold based on the valley density. This is achieved by the “printed wiring board inspection method according to any one of items (9)”.
[0023]
In addition to (11) “noise processing, (f1) median filter processing of the read wiring pattern, (f2) smoothing filter processing, (f3) isolated noise removal filter processing”. Furthermore, an operation for removing reflected light noise of the substrate to be tested is included. The reflected light noise removing operation detects (RGB) a red R color component for each of the three RGB components of the substrate pattern. The red pixels of any density are converted into high density pixels in the pixel set portion belonging to the maximum density value among the 256 gradation levels in the histogram, and (ii) for the green G color component, The detected green pixel of any density converts the green pixel to the density to which the largest pixel group belongs in the region of the intermediate density or higher of the 256 gradation levels in the histogram, while (iii) the blue pixel B For the color component, any detected blue pixel of the density is converted into a low density pixel of the density value of the pixel set belonging to the lowest density value among the 256 levels of density gradation in the histogram,
The white balance represented by R + G + B = white, which mainly cuts the B color component information, which is the main cause of the non-wiring information in the background area, without cutting the R color pixel component information as an element constituting the wiring pattern. Item (9) is an operation for removing reflection noise recognized as white light by removing white color by removing the B color pixel component from the color balance and decoloring the color balance. This is achieved by the “printed wiring board inspection method”.
[0024]
  In addition, (12) “standard virtual continuity data (MVCD) of original circuit image (MV) and / or test board wiring image data (SVCD) of (B) of the above (A) is originalImage filmCircuit image4Corner and / or test substrateWiring imageThe first (1) is characterized in that four positions forming four rectangles forming a rectangle having a sufficient width to surround the wiring image and / or wiring pattern are included in the binary data as four reference points at the four corners. The printed wiring board inspection method according to any one of Items 1 to 11 is achieved.
[0025]
Further, (13) of the present invention is characterized in that the presence or absence of a defective point other than the wiring formed of the same kind of material as the wiring is further determined in a rectangular area having a width surrounding the wiring pattern. This is achieved by the “printed wiring board inspection method according to item (12)”.
[0026]
Further, according to the present invention, (14) “the comparison operation of (C) includes at least one of the circuit original standard virtual continuity data of (A) and the test board wiring image data of the printed circuit board of (B)”. The method according to any one of (1) to (13) above, further comprising a process of superimposing one side by partial rotation or tilt correction, expansion / contraction correction and / or coordinate confirmation adjustment. This is achieved by the “printed wiring board inspection method”.
[0027]
Further, according to the present invention, (15) “disconnection in the wiring on the test substrate in (D) is the binary data (BCID) according to (a) in (A) and the test in (B) The printed wiring as described in (14) above, wherein when the comparison is made with the substrate wiring pattern data for superposition, it is detected as a surplus portion of the binary data (BCID) according to (a). This is achieved by the “substrate inspection method”.
[0028]
Further, according to the present invention, (16) “the short circuit in the wiring on the test substrate in (E) is the binary data (BCID) according to (a) in (A) and the test in (B)” Printed wiring as set forth in (14), characterized in that when the comparison is made with the substrate wiring pattern data for overlaying, it is detected as a missing portion of the binary data (BCID) according to (a). This is achieved by the “substrate inspection method”.
[0029]
Further, (17) of the present invention, “the potential disconnection in the wiring on the test substrate of (F) is the standard skeleton line data (MSLD) according to (b) in (A), and (B) Item (14) is characterized in that, when the comparison is made with the test substrate wiring pattern data of (b), it is detected as an extra portion of the standard skeleton line data (MSLD) according to (b). This is achieved by the "printed wiring board inspection method" described.
[0030]
Further, (18) of the present invention, “a potential short circuit in the wiring on the test substrate of (G) is caused by the standard expansion image data (MEID) according to (c) in (A), (B) Item (14) is characterized in that when the test substrate wiring pattern data is compared for overlay, it is detected as a missing portion of the standard expansion image data (MEID) according to (c). This is achieved by “a method for inspecting a printed wiring board”.
[0031]
Further, (19) of the present invention “Comparison for superimposing the standard expansion image data (MEPD) according to (c) in (A) and the test substrate wiring pattern data in (B) described above, Geometric features (q1, q2,... Qn) of the image in the original of the circuit and the corresponding points (q1 ′, q2 ′,. qn ′), and the printed wiring board inspection method according to item (18) above.
[0032]
  Furthermore, the above object is20) "(A) Standard virtual continuity data (MVCD) of circuit image of original image film used for image exposure on photoresist surface for printed circuit board creation, and (B) Test board wiring of printed circuit board to be inspected (C) Comparison with pattern data (SVCD) determines (D) disconnection and (E) short circuit in the wiring on the test substrate, and (F) potential disconnection and (G) potential in the wiring A printed wiring board inspection device for determining a short circuit,
  (A) For the standard virtual continuity data (MVCD) of the circuit image of the original image film of (A), (a) The image of the circuit image of the original image film used for image exposure on the photoresist surface for the production of a printed circuit board Image information reading means (13) for reading information into binary data (BCID), (b) CPU, (c), binary data (BCID), standard virtual continuity data of circuit image of the original film Memory means (18) for storing (MVCD) in a callable manner, (d) Standard virtual continuity data (MVCD) of the circuit image of the original film, the memory means(18)Reference point detection program for detecting a reference point of a read image based on binary data (BCID) called from(18)Thinning program for thinning binary data (BCID) called fromBinary data called from the memory means (18) (BCID) TheExpansion program for expansion, thinned binary data and expanded binary dataOverlay synthesisA memory means (19) for storing a merge program for executing, and (e) a monitor means,
  The image information reading means (13) of (a) is for (f) the test board wiring image data (SVCD) of the printed circuit board.Board to be testedImage information reading means that reads wiring pattern information and converts it to binary data (BCPD)AlsoYes,
  The CPU in (b) compares the standard virtual continuity data (MVCD) of the circuit image of the original film with the test board wiring pattern data (SVCD) to determine disconnection and short circuit of the test board wiring. And a comparison means for determining a potential disconnection and a potential short circuit in the wiring,
  The memory means (18) of (c) is a memory means for storing the read binary data (BCPD) of the test substrate pattern and the test substrate wiring pattern data (SVCD) in a callable manner in the CPU. ,
  The memory means (19) of (d) stores a classification density threshold detection program for calculating a density threshold for classifying the three RGB components of the read test substrate pattern from the background portion. This is achieved by a printed circuit board inspection apparatus characterized in that it is a memory means that is stored in a callable manner in the CPU.
[0033]
Hereinafter, the present invention will be described in detail.
The printed wiring board inspection technique of the present invention is mainly configured to detect the difference in shape between the board and the film on the two-dimensional plane (both inspection that guarantees the line width and the conduction data route in pixel units). Has been. This inspection method of the present invention is particularly suitable for superimposing a film and a substrate image data captured by an optical image information reading device (eg, a scanner) and detecting a mismatched portion as a defect.
[0034]
Image area that exists on the test substrate but not on the original film is detected as a short (short) or thickening and corresponds to a defect in the superimposed image. Is presented as a disconnection or thinning.
The main image processing contents in the present invention include a geometric state changing process for superimposing a film image and a substrate image, and a process for classifying defect types when superposed. Typically, (1) the original film image is binarized and linearized to create data that recognizes the end point, (2) the circuit image of the board is binarized, and (3) standard The connection inspection is performed by comparing the skeleton line data and the wiring data of the test substrate. The data obtained by binarizing the circuit image of (2) includes standard skeleton line data obtained by thinning this and standard dilation line data obtained by dilatation processing. Since it is used for the virtual continuity test, it is distinguished from the original binarized data in this specification and may be referred to as standard virtual continuity data. The wiring data of the board to be tested (3) is also binarized (sometimes referred to as virtual continuity data to be tested). As a result of comparison, an image area that is present on the board but not on the film is detected as a short portion in a portion corresponding to a defect in the wiring pattern on the board, and an image area that is present on the film but not on the board is detected. Is indicated as a disconnection point, and the circuit pattern is inspected for thickness and thinness based on the distance between the outer contour position of the circuit pattern of the film and the substrate and the position of the skeleton line, and a potential defect portion or potential Performs a virtual continuity test to detect short parts.
[0035]
Hereinafter, an outline will be given along the processing procedure.
[Create inspection standard data]
<1. Reading original film image-forming binary (BCID) data>
As shown in FIG. 1A, an original image film (10) used for image exposure onto a photoresist surface for producing a printed circuit board is formed on a background area (12) and a printed wiring board on the printed wiring board. The image area (11) which becomes a circuit pattern area is read, and the image area (11) and the background area (12) adjacent to the area (11) are read and converted into binary data (BCD). That is, for example, as shown in FIG. 1B, in this binary data (BCD) conversion operation, each of the two areas is divided into a large number of fine areas in units of dots, and the positions of these fine areas. Reflected image density data (IDD) for identifying whether it is an image area or a background area is formed and data indicating the position of each fine area and the image area from these two data. Binary data (BCID) is formed from the data indicating whether it is a background area.
[0036]
The operation for converting to binary data (BCID) includes the image information of the image area (11) of the original pattern film (10) used for image exposure on the photoresist surface when the printed circuit board is created, and at least the image area (11). Reading the non-image information of the surrounding background area (12) by dividing it into a large number of fine areas in dot units, and adding position coordinate data (PCD) to each of the read fine areas in dot units; The high light reflectance information or the low image density information of a large number of fine areas in the dot unit of the pattern area (11) taken as one value, for example, (01), and the low light reflectance information or high of the background area (12) And operation for obtaining reflected image density data (IDD) by setting the background area density information to the other value, for example, (00).
[0037]
In the present invention, it is preferable to store the position coordinate data (PCD) after the operation of giving the position coordinate data, and high light reflection of a large number of fine areas in dot units of the read image area. Reflected image density data (IDD) is obtained by setting the rate information or low image density information as one value (01) and the low light reflectance information or high background area density information as the other value (00). It is preferable to store reflected image density data (IDD) after the operation.
This binary data (BCID) formation operation is any operation as long as it forms data consisting of data indicating the position of each fine area and data indicating whether it is an image area or a background area. Of course, there is no problem.
[0038]
In the present invention, the image reading is preferably performed by a scanner. When capturing a film, it is sufficient to capture only the shape and geometric characteristics thereof, so that it is captured as monochrome binary data. In the program used in the present invention, the exchange between the scanner and the personal computer (PC) can be performed in accordance with a general TWAIN standard. Therefore, a TWAIN compatible scanner built on Windows (registered trademark) is basically used. It has a configuration that does not choose any model.
[0039]
<2. Thinning>
Thinning is a process of creating data (hereinafter referred to as conduction data) that guarantees pattern conduction while maintaining the geometric characteristics of the film image. As shown in FIG. 2, thinning of an image to be a circuit pattern on a substrate according to the present invention is performed by cutting an image area into a background area bit by bit from the outside based on binary data (BCD). It is. When the pixel is sufficiently small, of course, it may be cut into the background area in units of 2 bits instead of 1 bit. However, in this specification, in order to facilitate understanding, the case of cutting down 1 bit at a time is described below. explain.
The cutting into the background area is an operation in which the pixels forming the image area used in the secondary correction of the image information are cut from the outside into the background area bit by bit from the original circuit image, and binary data (BCID ) When the image pixel located at the boundary with the background area is set as the center of the matrix of two-dimensionally developed pixels of 3 pixels in the vertical row and 3 pixels in the horizontal row, the background region pixel among the surrounding 8 pixels The line width is obtained by repeating the operation of rewriting to a non-image pixel based on which is an image pixel and the image pixel in an alternating direction along the outer periphery of the pattern until there is no parallel part of the image pixel-image pixel in contact with the side Is performed by repeating in the alternating direction along the outer periphery of the image until there is no portion exceeding one pixel. Various methods have been proposed for thinning, but the following method can be preferably used here.
[0040]
That is, the thinning ((thinning) or skeletonizing) process in the present invention simplifies the description of the line figure for easy handling on the computer. Various algorithms have been proposed for thinning and thinning. Here, an example will be described. As shown in FIG. 3, the symbol of the pixel of the binary image B to be thinned is represented by p as the center pixel.₀, Its neighboring pixels are p_k, K∈N₈= {1, 2,..., 8}, the thinning procedure for the binary image B consists of two steps as follows.
“Step 1” The boundary pixel of the binary image B is examined, and the pixel p₀If the following four conditions are satisfied, p₀A flag is attached to the pixel. The pixel value B (p) marked at the end of the above processing for all boundary pixels is set to zero.
"conditions"
[0041]
[Expression 1]

(Ii) S (p₀) = 1, where S (p₀) Is p₀Value B (p₁), B (p₂), B (p₃), ..., B (p₈), B (p₁) In this order is the number of times the pixel value has changed from 0 to 1. For example, in the case of FIG.₀) = 3, S (p₀) = 2.
(Iii) B (p₁) ・ B (p₃) ・ B (p₇) = 0
(Iv) B (p₁) ・ B (p₅) ・ B (p₇) = 0
“Step 2”: Conditions (iii) and (iv) of Step 1 are changed as follows. Other conditions and processing procedures are the same as in Step 1.
(Iii) ’B (p₁) ・ B (p₃) ・ B (p₅) = 0
(Iv) ’B (p₃) ・ B (p₅) ・ B (p₇) = 0
This thin line processing can be performed by designating the number of times until a certain line width is obtained, or, as described above, can be repeated until a figure having a line width of 1 is finally obtained. It is preferable to specify the number of times until a certain line width is obtained.
[0042]
For example, if the image of the character E in FIG. 5 is thinned, FIG. (A) is the original image, FIG. (B) is the image obtained by performing step 1 once, and FIG. The image performed once, FIG. (D) is a thinned image. In the image thinning operation in the present invention, the start end (p1 point) and the end (p2 point) of the formed thin line are further labeled, and both the end portions (p1 point) (p2 point) and the connection portion are connected. It is preferable to form standard skeleton line data (MSLD) including the feature data.
[0043]
In addition, accompanying the thinning process, the beginning (p1 point) and the end (p2 point) of each formed thin line are labeled, and the both ends (p1 point) (p2 point) and the connecting part Feature data is obtained and added to the thinned data to form thinned standard skeleton line data (MSLD) of the circuit image.
[0044]
<3. Expansion>
Normally, when the standard skeleton line data of the captured board image and the thin film image are compared, the board image and the standard skeleton line data do not overlap correctly due to problems such as the reading position accuracy of the image (scanner etc.). . Therefore, in order to invalidate the positional deviation due to the reading accuracy of the image (scanner, etc.) in units of several dots, an image with an expanded original pattern line width is prepared so as not to impair the geometric characteristics of the film image. At the time of comparison, the positional deviation is ignored with respect to the substrate image in the expanded image. Further, the number of expansions of the image is processed according to the designated number of dots. When the line spacing is narrow so as not to connect to adjacent graphic elements at the time of expansion, this portion is set so as to guarantee a gap having a minimum width of 1 dot. This minimum width can of course be a gap larger than the dot in some cases.
[0045]
In the present invention, when the standard skeletal line data of a film image captured and thinned by reading an image of the original film is compared with the wiring pattern on the substrate, the reading position accuracy of the image or pattern (by a scanner or the like), etc. Because of this problem, the substrate pattern and the standard skeleton line data do not overlap correctly. Therefore, in order to invalidate the positional deviation due to the reading accuracy of the image (scanner, etc.) in units of several dots, prepare an image with the original pattern line width expanded so as not to impair the geometric characteristics of the film image. At the time of comparison, the positional deviation is ignored with respect to the substrate image in the expanded image. Further, the number of expansions of the image is processed according to the designated number of dots.
[0046]
Further, when the line spacing is narrow so as not to connect to adjacent graphic elements at the time of expansion, this portion is set so as to guarantee a gap with a minimum width of 1 dot, and standard expanded image data (MEID) is obtained. The geometric characteristics of the film image in the expansion operation for obtaining the standard expanded image data (MEID) include, for example, L-shaped curved point, T-shaped intersection, Y-shaped intersection, end thick point, curved portion, shoulder Part of the image, such as X-shaped intersections.
[0047]
<4. Image composition>
Thinned standard skeleton line data (MSLD) and standard dilated image data (MEID) are synthesized. At this time, the thinned standard skeleton line data (MSLD) is converted into a designated color (for example, red) and then superposed on the standard expanded image data (for example, white). Alternatively, the thinned image and the expanded image (background black, expanded white) are superimposed. At this time, the thinned data is superposed after being converted into another color (background black, thin line red). By doing so, it is possible to handle the continuity data that should be guaranteed at the minimum and the expanded image in consideration of the fatness, thinness, and error on the same image. The synthesized image is converted into a hue of 1 pixel = 1 byte (8 bits). This composite image data is composite standard image data (CMID).
[0048]
<5. Reference point detection>
Also, based on this composite standard image data (CMID), the original image is divided into four small areas including image elements from the four corners, and for each divided small area, the respective images included in them. The center point of the element is formed as reference point data.
The reference point separates the graphic elements within the specified range from the four corners of the film data, and the circularity within a certain range included therein (usually about 0.85 to 1.2 can be specified) or the center detection error is small A center point of an image element having a shape is detected. In the present invention, the reference point is used to adjust the rotation direction and horizontal / vertical direction for overlapping the image and the wiring pattern, and expansion / contraction. The range connecting the reference points is the inspection target range. There are the following feature points in the figure related to the center point, and there are various methods for obtaining them.
[0049]
That is, as a feature point of the binary image as shown in FIG.
(A) the area of the object,
(B) the perimeter of the object;
(C) Maximum value R of the distance from the center of gravity G to the edge_maxAnd minimum value R_min,
(D) Average value R for the entire circumference of the distance from the center of gravity G to the edge_mean
(E) Maximum radius R_maxAxis and minimum radius R_minThe angle ψ between the X axis and the X axis (reference axis)_RmaxAnd ψ_Rmin,,
(F) The major axis d of the inertia equivalent ellipse_emaxAnd minor axis d_emin,
(G) Angle ψ between inertial main axis X and axis_mm,
(H) Long side L of a circumscribed rectangle parallel to the inertial main axis_RmAnd short side L_RsLength of,
(I) The position of the center of gravity of the object (X_G, Y_G),
(J) Maximum value L of the projection of the object onto the X axis and the Y axis_PXmAnd L_PYm, Minimum value L_PXsAnd L_PYs,
(K) the sum of the number and area of holes in the object).
[0050]
The circularity is used to know the shape of the object region, and is defined as follows using the area A and the peripheral length L.
[0051]
[Expression 2]

If the region is circular, R will be 1 (maximum), and this value will be smaller than 1 as it deviates from the circle (for example, 0.785 for a square).
In the present invention, by detecting the center point of each image element based on the circularity of each image element, the reference point of the image element in each small region is formed as reference point data (SPOD).
Then, reference point data (SPOD) is merged with the composite standard image data (CMID) to form composite reference image data (CMSID).
These [Create Test Standard Data] operations are summarized as shown in FIG.
[0052]
[Create test data]
<6. Circuit board wiring pattern reading>
According to the present invention, the test board wiring pattern data (SVCD) (B) includes four positions for forming four rectangles that are sufficiently wide to surround the wiring pattern information at the four corners of the test printed board. As described above, it is formed by reading the wiring pattern information of the test substrate and converting it into binary data (BCPD). When binary data (BCPD) is read by reading this wiring pattern information, (f1) median filter processing (f2) of the read wiring pattern is performed in order to remove noise that is indeterminate and undefined in size. It is preferable to include smoothing filter processing and (f3) isolated noise removal filter processing.
[0053]
<7. Smoothing filter processing>
In the case of noise elimination from a single noisy image, the true value of a pixel that is randomly hidden by noise can never be known, but the purpose is to prevent noise from getting in the way. To do this, it can be visually inconspicuous.
As can be seen from FIG. 7 as an enlarged view of an image with noise, there is an abrupt density difference between the density of noise and the density around it. Also, since there is an abrupt difference in density, it is annoying, and a technique for removing noise using this noise property is generally called smoothing. However, since the edge portion of the target image also has a sharp density difference, how to separate the edge portion from the noise portion and remove only the noise is the highlight of smoothing.
The moving average method is the simplest noise removal method, and is a method of replacing the average value of 3 × 3 pixels around a certain pixel with the value of the pixel as shown in FIG. In short, it uses the theory that if the image is blurred, the fine noise disappears. In this method, whether it is a noise or an edge, it blurs without any difference, so even if the noise can be removed, the target image may be blurred.
[0054]
<8. Median filter processing>
Therefore, it has been considered to remove noise without blurring the edge of the target image as much as possible, and a well-known technique called a median filter is advantageous in the present invention.
Assume that there is an image having a density as shown in FIG. Now, in order to obtain the values of the pixels surrounded by a circle, the densities of nine pixels in a 3 × 3 region (region surrounded by a solid line) are examined and arranged in ascending order as follows.
2 2 3 3 (4) 4 4 5 10
The central value at this time (this is called the median), in this case nine in total, so the fifth density 4 from the left is the density of the pixel to be obtained. Assuming that the pixel of 10 is included as noise, the noise is certainly removed. This is because, when the density is extremely different from the surrounding area, when they are arranged in the order of size, they are gathered in the left or right side and are not selected as the median value. As described above, the median filter is a process for obtaining a median value of the density of pixels in an area around a certain pixel and setting it as the density of the target pixel. Then, the pixel in the area surrounded by the dotted line is examined as to what will happen to the next pixel on the right.
2 3 3 4 (4) 4 4 5 10
The median is 4, which is actually 4 but 4. This is the damage caused by the treatment, but it is often not visually understood. The problem is whether or not the edge portion is preserved. FIG. 10A shows an image with an edge. When pixels in a circle are obtained in order, the result is as shown in FIG. You can see that the edges are completely preserved. In the moving average, since the noise component is also included in the calculation of the average, the output is affected by the noise. However, in the median filter, the noise component is not easily selected, and thus the output is not greatly affected. Therefore, when compared with the same 3 × 3, the median filter is superior in noise removal capability. When the effect is seen in an actual image, FIG. 11 shows the results of processing a noisy image by a median filter and a moving average method, respectively, and the median filter preserves edges even when removing noise. It can be seen that this is a very good technique. The calculation time of the median filter is about five times longer than the moving average.
[0055]
<9. Isolated noise removal filter processing>
The output image is a binary image for both region extraction and edge extraction. Before performing these processes, it is important to perform pre-processing that removes noise in the image in advance by smoothing. This facilitates the subsequent processes. However, if there is still a disturbing noise in the output binary image, it must be removed by post-processing. The binary image noise is as shown in FIG. 12 and is called sesame salt noise. Of course, this noise can also be removed by a median filter, but can be removed in the present invention by a process called expansion / contraction using the binary value. In this case, expansion is a process for setting a pixel to 1 if there is at least one in the vicinity of a certain pixel, and 0 for others. Shrinkage is a process in which there is at least one 0 in the vicinity of a certain pixel. This process is to set the pixel to 0, and the others to 1. When this process is applied to expansion → contraction, the resulting image is fattened by expansion and thinned by contraction, and as a result hardly changes, but black isolated noise is removed during expansion, conversely, contraction → When it works with expansion, white isolated noise can be removed during contraction.
<9a. Reflected light noise removal processing>
This noise removal processing includes noise removal processing based on reflected light. This reflected light noise removal operation is performed according to the characteristics of the light source used, the spectral characteristics of the material on the surface of the test substrate, the spectral characteristics of the color filter used for color separation, and the color display characteristics of the color monitor. This is an operation for erasing a portion that is recognized as reflected light noise beyond the hue deviation between the wiring pattern on the substrate surface and the background region. Specifically, in the present invention, for each of the three RGB components of the test substrate pattern, (i) for the red R component, any detected red pixel of any density will be described in detail later. The histogram is converted into a high density pixel of the pixel set part belonging to the maximum density value among the density gradations in the histogram, for example, 256 gradations, and (ii) for the green G color component, The green pixel of any density detected is converted into a density to which the largest pixel group belongs in the region of the intermediate density or higher among the 256 gradation levels in the histogram. For the B color component, the detected blue pixel of any density is converted into a low density pixel of the density value of the pixel set belonging to the lowest density value among the 256 levels of density gradation in the histogram. By performing this pixel density conversion processing, the B color component information, which is the main cause of the non-wiring information in the background area, is cut intensively without mainly cutting the R color pixel component information as an element constituting the wiring pattern. Then, by removing the B color pixel component from the white balance represented by R + G + B = white and losing the color balance, it is possible to remove the reflection noise recognized as white light.
In the present invention, the density gradation is not limited to 256 stages, and may be any other number of stages suitable for image processing by a computer, for example, 64 stages. However, for ease of understanding, the following description describes the case of 256 stages.
[0056]
<10. Density calculation for RGB component separation by threshold>
In the present invention, the test substrate wiring pattern data (SVCD) is calculated for each of the three RGB color components of the substrate pattern and the calculated wiring pattern density distribution and the background portion which is a non-pattern portion of each component substrate. It is preferable that the density is calculated based on a threshold value so that the wiring pattern of the substrate of each RGB component can be smoothly separated from the background portion by contrasting with the density distribution. In other words, the board image is currently captured with a density value of 8 bits (0 to 256) for each color of RGB. Here, since there is a relatively large density difference between the object (circuit pattern) on the substrate and the other background, a discrimination threshold selection method (for example, Noriyuki Otsu: based on judgment and minimum two criteria) so that it can be accurately separated. Using the automatic threshold selection method, refer to the IEICE Transactions), the density value that maximizes the density dispersion is obtained for each RGB color, and the color indicating the pattern is emphasized for the density value of each color obtained. A value weighted to can be calculated as a threshold for binarization.
[0057]
In the present invention, the classification density threshold value detection operation for calculating the density threshold value for classification from the background portion for each of the three RGB components read from the test substrate pattern, specifically, for each color, By classifying all the constituent pixels, a histogram indicating which density gradation, for example, one of 256 density gradations belongs, is automatically created, and a set of pixels belonging to a certain density area in this histogram This is an operation of finding the density of a valley where the number of existing pixels between the body and another pixel group belonging to another density area is small, and detecting the density threshold based on this valley density.
[0058]
As shown in FIG. 30, for example, the density of 256 gradations is plotted on the horizontal axis for each of red, green, and blue, and the number of pixels is plotted on the vertical axis for each gradation density. For example, valleys clearly appear in the graph. If not, the 200th density in red is the maximum density value of the portion where the pixels are gathered, and the other red pixels are also replaced with the 200th density. Convert to In green, the density to which the largest pixel group belongs in the range of 256 gray levels or higher is the 181st density, so other green pixels are also converted to the 200th density. Further, the blue pixel is always converted to one having a zero value density.
[0059]
Continuing the description of the histogram, for example, in FIG. 13, the character “picture” is shown together with another subject, but an automatically selected threshold can be used to extract only the character portion from this. The threshold processing is to set the pixel value of the corresponding output image to 1 when the brightness is equal to or greater than a certain threshold for each pixel of the input screen, and to 0 otherwise. In terms of the formula:
[0060]
[Equation 3]

become that way. Here, f (x, y) and g (x, y) are the density values of the pixels at the location (x, y) in the image before and after the processing, respectively, and t is a threshold value. In the case of the image of FIG. 13, since the characters are darker and the background is lighter, the background is extracted when the above expression is applied. Therefore, the following expression is used to extract those smaller than the threshold. Such calculation of the density threshold value can be automatically performed by the discrimination threshold value selection method in the present invention.
[0061]
[Expression 4]

An example of a program that performs threshold processing is shown in the following list.
[0062]
[Table 1]

[0063]
For the modes in this list, [Equation 3] is selected when mode = 1, and [Equation 4] is selected when mode = 2. Also, the value after the threshold processing is not 1 and 0 but HIGH and LOW. When the same display program as the original image is used, HIGH = 255 and LOW = 0.
An example of threshold processing using this program is shown in FIG. Depending on how the threshold is given, the degree of omission changes considerably. As is clear from this example, too large threshold values are taken out to the excess, while too small threshold values are discarded as necessary.
[0064]
When describing how the optimum threshold should be determined, the brightness of the object to be extracted and the background should be different. Otherwise, it cannot be identified by the human eye. Therefore, check the brightness of the part you want to extract and the part you want to cut out by specifying some positions. In the example of FIG. 13, the character portion is about 140 and the background portion is about 160. Therefore, if the threshold value is selected to be about 150, the character and the background are likely to be separated. However, there is a method using a histogram (frequency distribution) as a slightly smarter method without checking the brightness of the pixel for each point. As shown in FIG. 15, the histogram indicates the frequency of how many pixels of density value i, and can be obtained by the program shown in the above list. Since the histogram cannot be viewed with only the above list, the histogram is printed out by a program for printing the histogram or printed by a program for imaging the histogram. A clear histogram as shown in FIG. 16 can be obtained.
According to this histogram, the mountains around the density value 140 correspond to the pixels of the character portion, and the mountains around the density value 160 correspond to the background pixels. It seems that the boundary between these two peaks, that is, the valleys, can be well separated, but this histogram is very bumpy and the valleys are not well understood. Therefore, as a method for making it easy to find a valley, it is often performed to display the histogram by averaging the neighborhoods on the histogram to reduce the bumps, and this method can also be suitably used in the present invention. In this way, it becomes easy to find a valley and it can be determined that it is around the density value 148. Although it is not completely extracted due to reasons such as uneven lighting and blurred characters, it can be extracted quite well compared to other threshold values, and the threshold value can be set using the valley of the histogram in this way. The method of determining is called the mode method.
In addition to the mode method, there are a P tile method, a discriminant analysis method, a variable threshold method, and the like as methods for determining the threshold. The P tile method is used when the proportion of objects in the entire image is known (for example, p%). This is a method in which the threshold value is p% of the total frequencies on the histogram from the darkest (or from the brightest), and the discriminant analysis method is when the histogram is divided into two groups of objects and background. In other words, the threshold value is determined so that the statistics of the two groups are different. The variable threshold method is an effective method when the background is not uniform in brightness, and is a method of changing the threshold for each place.
[0065]
<11. Reference point detection>
In the reference point search, the image area within the specified range is cut out from the four corners in the same manner as in the case of the film, and then binarization processing is performed, and the circularity within a certain range included therein and a shape with little center error are included. The center point of the image element is detected as a reference point. This reference point is a film replica and is therefore formed at the same position as the film during pattern formation. Synthetic test data is obtained through the processing from reading the substrate wiring pattern to detecting the reference point.
The outline of these [create test data] operations is shown in FIG.
[0066]
[Inspection process]
Image processing for inspection is performed in two stages. The first stage is a process of superimposing the composite test data (test board wiring image data (SVCD)) and the composite reference data (standard virtual continuity data (MVCD)), and the second stage is effective from the superposed image. This is a process for extracting a difference and discriminating disconnection, short-circuiting, fatness, and thinning.
[0067]
(1) Binarization and superposition
<12. Binarization processing>
The board image is binarized and only the pattern figure is cut out as an effective figure. At that time, in order to reduce the influence of minute noise included in the substrate image, local color difference due to light hitting at the time of scanning, blurring of the edge portion of the image (unclear part of the boundary pixel), etc. As the pre-value processing, a median filter, an averaging filter, an isolated noise removal filter, or the like is applied by user selection. In an image with a lot of noise, since there are a lot of missing images and minute noise after binarization, it is difficult to accurately discriminate between disconnection, short circuit, thickness and thinness.
By applying the preprocessing as described above to the image before binarization, it is necessary to reduce the influence of noise while preserving the edge of the pattern as much as possible. The binarized image is converted into an image of 1 pixel = 1 byte (8 bits), and the portion where the pattern exists is displayed in blue.
[0068]
<13. Image overlay>
An image is created by superimposing the created to-be-synthesized inspection data and synthesis reference data. At this time, the rotation angle and the expansion / contraction rate of the combined inspection data and the combined reference data are calculated using the coordinates of the reference point detected previously, and the images are superimposed while taking this into consideration. Specifically, an exclusive logical ring of the pixels of the film image existing at the position of the target pixel on the substrate is taken, and the value is output as a superimposed image. In this way, the binarized board image is output as it is in the binarized blue color as it is, and the deficient part is output as it is in the red color that was colored when creating the synthesis reference data. can get. The black part that is the other part is a non-inspection area, and the part where the images overlap is displayed in a color other than the original red / blue. Further, after the above processing, the entire image is divided into small areas (for example, divided into 64 dots × 64 dots) using the coordinates of the reference point, and the small areas are rotated and stretched to superimpose the images. Specifically, the divided small area is provided with an overlap pixel (for example, 3 dots × 3 dots) corresponding to several vertical and horizontal pixels, and is rotated and expanded / contracted (for example, a 3 × 3 dot range within the range of the pixels). 36 times geometric correction), the overlapping rate of each small region is calculated, and the portion with the highest rate is selected and fixed. This is performed for each small area. By processing such two-stage overlay, large overlay shifts in the entire image are distributed to each small area, and by further rotating and stretching correction, the shift widths in each small area are dispersed. It becomes smaller and the superposition process can be speeded up, and the accuracy and speed are improved in each stage.
[0069]
(2) Detection of differences (the second stage of the inspection process)
In the superimposing and comparing operations in the present invention, at least one of the standard virtual continuity data of the circuit original and the test board wiring image data of the printed circuit board is subjected to partial rotation or tilt correction, expansion / contraction correction, and / or Alternatively, it is preferable to superimpose after coordinate confirmation adjustment, and it is particularly preferable to superimpose the standard virtual continuity data of the circuit original by partial rotation or tilt correction, expansion / contraction correction and / or coordinate confirmation adjustment.
[0070]
<14. Detection of disconnection and thinning>
As described above, the overlapped image is displayed in red for the deficient portion and in blue for the excess area. Here, the lacking portion is detected as a disconnection or thinning, but what is displayed in red is a film image that has been linearized, so the line width is partially 1 pixel. For this reason, when the images are accurately overlaid, the disconnection can be determined as a disconnection even if one pixel has red color.
[0071]
<15. Short circuit and fat detection>
Next, in order to take into account the deviation of the substrate image and the like, the film image is overlaid by applying an expansion process. Since the film image is prevented from being connected to an adjacent graphic element at the time of expansion, when the distance between the lines is narrow, this portion has only an opening with a minimum width of 1 dot. Therefore, two types of judgment criteria are set for short and fat detection.
[0072]
In the present invention, for example, (i) the disconnection in the wiring on the test substrate in (D) includes the binary data (BCID) according to (a) in (A) and the test substrate in (B). When the wiring pattern data is compared for overlaying, it is detected as a surplus portion of the binary data (BCID) according to (a),
(Ii) The short circuit in the wiring on the test substrate in (E) is performed by using the binary data (BCID) according to (a) in (A) and the test substrate wiring pattern data in (B). When comparison for superposition is performed, it is detected as a missing portion of the binary data (BCID) according to (a),
(Iii) The potential disconnection in the wiring on the test substrate in (F) includes the standard skeleton line data (MSLD) according to (b) in (A) and the test substrate wiring pattern in (B). When the comparison with the data is performed for superposition, it is detected as an extra portion of the standard skeletal line data (MSLD) according to (b),
(Iv) The potential short circuit in the wiring on the test substrate in (G) is the standard expansion image data (MEID) according to (c) in (A) and the test substrate wiring pattern data in (B). Are compared for overlaying, they are detected as missing portions of the standard dilated image data (MEID) according to (c).
[0073]
As described above, in the present invention, among the pixels converted to binary data (BCID), an image pixel located at the boundary with the background area is represented as a two-dimensionally developed pixel of 3 pixels in the vertical row and 3 pixels in the horizontal row. When the center of the matrix is used, the rewriting operation to the non-image pixel is performed based on which of the surrounding pixels is the background region pixel and which is the image pixel. More specifically, the case of using a monitor capable of displaying in three colors of RGB will be described. As shown in the following (I), in a 3 × 3 dot image area, on a vertical, horizontal, and diagonal continuous line including the central pixel. If there are three blue pixels side by side (FIG. 19), or the image area of 2 × 2 dots as shown in (II), all of them are blue pixels (FIG. 20). Due to short and fat existing in the gap of line width 1 by processing A portion that, precisely sorted all the defective area present in wide space on (points that do not exist in the film) can be detected.
21 and 22 show an outline of such an [inspection processing] operation.
[0074]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, more specific embodiments of the present invention will be described.
23 and 24 show (i) original reading processing, (ii) thinning processing of a film image, (iii) expansion processing of a film image, and (iv) thinning of a film image in one specific example of the present invention. (A) Standard virtual continuity data (MVCD) creation comprising image synthesis processing of data and expanded data, and (v) reference point calculation processing of synthesis reference data will be described. FIG. 25 and FIG. Image reading processing, (vii) RGB three-color filter is applied as pre-binarization processing of the substrate image, isolated noise removal, smoothing filter, median filter processing, (viii) substrate image binarization processing, ( ix) (B) Preparation of the test board wiring image data (SVCD) consisting of the reference point calculation of the test composite wiring data will be described. FIGS. 27, 28, and 29 show the (X) composite reference data and the test. Comparison of (C) to (G) comprising wiring data image overlay processing and (xi) difference detection processing This explains the constant operation.
[0075]
<Creation of composite reference data>
(I) In reading the film image in this example, the image is taken in as black and white binarized data. In this example, transmission / reception between the scanner and the PC is performed in accordance with the TWAIN standard. The uppermost figure shows the original film image of the wiring with the end enlarged.
(ii) In the thinning process of a film image, the number of times of linearization is specified in dot units, and the specified number of times of linearization processing is performed. The connected element of the lined dots becomes a criterion for determining the minimum guaranteed width region for disconnection and thinning. The example of this figure shows a state after the number of times of thinning is one.
In (iii), the number of expansions is designated in dot units, and the designated number of expansions is performed. The expanded dot connecting element is a criterion for determining whether or not the positional displacement invalidation area, the fat maximum guaranteed width area, and the short circuit are short-circuited. The example of this figure shows a state after the number of expansions 1.
(iV) The data synthesized by the image synthesizing process of the thinned data and the expanded data of the film image is the synthesis reference data that is a criterion for determining the continuity of dots, thinning, and thickening.
(v) In this example, four points are calculated at each corner of the original image as reference points for the composition reference data in the image composition process. Although the color is not shown in this figure, in this example, the result of the thinning data of the film image of (ii) is displayed in red, and the result of expansion of (iii) is displayed in white. Therefore, identification becomes easy.
Although not shown in the figure, in the present invention, in order to improve the processing speed of thinning, the pixels are cut by cutting down to the minimum guaranteed width without cutting the line from the entire contour to one pixel. The number of times can be reduced. Further, the linearized data lined up to the minimum guaranteed width can be inspected simultaneously for the guaranteed conduction route and the minimum guaranteed width. Furthermore, the expansion process can obtain an invalid range and a maximum guaranteed width for eliminating inspection misidentification due to misalignment when images are superimposed.
[0076]
<Create synthetic test pattern data>
(Vi) Substrate image reading is performed in 24-bit color with the same resolution as film image reading. Currently, data transmission / reception is based on the TWAIN standard.
(Vii) As pre-binarization processing of a substrate image, three types of filters, an isolated noise removal filter, a smoothing filter, and a median filter, are selectively applied to the captured substrate image.
(Viii) The binarization processing of the substrate image is calculated as a binarization threshold with a value that emphasizes the circuit pattern portion by using a discrimination threshold selection method. Although no color is shown in this figure, in this example, the circuit pattern portion is displayed in blue and the background area is displayed in black, so that identification is easy. The obtained data is used as synthesis inspection data. The isolated noise removal is performed by a method of removing a blue pixel of one dot, considering a blue pixel not connected immediately before display after the binarization process as noise. The order of use when using three types of filters is as follows: 1. Median filter, 2. Smoothing filter It was in the order of isolated noise removal filter. In the present invention, each filter can be selected as necessary in this way, but the binarization processing speed can be increased by using only the necessary filters.
(ix) In the calculation of the reference point of the composite inspection data, the composite inspection data includes the accuracy of detecting the defective part (disconnection, short, thick, thin part shape, identification of unnecessary deposits, adhesion between the pattern forming member and the substrate. It is important to obtain a binarized image that is as close as possible to the raw image of the inspection board in order to improve (identification based on the color of the defective part). Therefore, more accurate information can be obtained by not performing processing other than the above-described binarized image processing of the substrate image (eg, thinning, expansion, etc.).
<Overlay of both data>
(x) Furthermore, an image overlay process is performed on the composite reference data and the composite inspection data,
(xi) A difference was detected. Good test results were obtained.
[0077]
The pre-inspection processing and inspection time (board inspection) in this example were as follows.
(I) Inspection range
B5 size: About 180x250mm
Captured data: Approximately 50MB
(A) Pre-inspection processing (scanner scanning-filter application-binarization processing) = about 55 sec
(B) Inspection execution / display = about 25 sec
(C) Total inspection time: 55 + 25 = about 75 sec
(Ii) Inspection range
A4 size: about 210 x 300mm
Captured data: About 100MB
(A) Pre-inspection processing (scanner scanning-filter application-binarization processing) = about 105 sec
(B) Inspection execution / display = about 25 sec
(C) Total inspection time: 105 + 25 = about 130 sec
(Iii) Inspection range
A3 size: about 420 x 300mm
Captured data: Approximately 180MB
(A) Pre-inspection processing (scanner scanning-filter application-binarization processing) = about 180 sec
(B) Inspection execution / display = about 30 sec
(C) Total inspection time: 180 + 30 = about 210 sec
Standard data creation time (film)
(I) Film standard range
B5 size, imported data about 2.5MB
(When the number of thinnings is 2 and the number of fats is 3)
Standard data creation time (scanner capture-linearization / expansion processing-display) = approx. 80 sec
(ii) Film standard range
A4 size, captured data about 4.5MB
(When the number of thinnings is 2 and the number of fats is 3)
Standard data creation time (scanner capture-linearization / expansion processing-display) = about 120 sec
(iii) Film standard range
A3 size, imported data about 6.5MB
(When the number of thinnings is 2 and the number of fats is 3)
Standard data creation time (scanner capture-linearization / expansion processing-display) = about 200 sec
[0078]
FIG. 31 is a diagram for explaining the outline of a printed wiring board inspection method and inspection apparatus according to the present invention. In the figure, a circuit image (11) of an original image film (10) on a Schatten base (10a) is a scanner as an image information reading means having an image optical reading unit (13a) and an image information processing unit (131). (13), is converted into binary data via the CPU (16) of the computer (15), and is stored in the memory (18) including the RAM (1) and the RAM (2). The scanner (13) as an image information reading unit has an image information processing unit (131), and the image information processing unit (131) separates the read image information into RGB three-color information (131a). ), A shift register (131b) for RGB three-color information, a data latch unit (131c), a decoder (131d), a control device (131e) for controlling these and controlling the driving of the optical reading unit (13a), For this purpose, a memory (131f) for storing a control program in a callable manner and a memory (131g) for temporarily storing image information delivered from the decoder (131d) are provided.
In the drawing, the memory (18) is shown divided into RAM (1) and RAM (2), but of course, it is not necessary to have two, and data of a large number of three color pixels having 256 gradations is used. In view of the large-scale calculation for converting these large numbers of data one after another and the large number of repeated calculations, it is only necessary to be able to store and call the appropriate capacity. The same applies to the ROM (1) and the ROM (2) described as the memory (19) that mainly store the image processing program in a readable manner.
[0079]
Therefore, this inspection apparatus includes (A) standard virtual continuity data (MVCD) of the circuit image of the original image film used for image exposure on the photoresist surface for creating the printed circuit board, and (B) the printed circuit board to be inspected. (C) comparison with the test substrate wiring pattern data (SVCD) of (D) disconnection and (E) short circuit in the wiring on the test substrate, and (F) potential disconnection in the wiring and (G) A printed wiring board inspection apparatus for determining a potential short circuit,
(A) For the standard virtual continuity data (MVCD) of the circuit image of the original image film of (A), (a) Image information of the circuit image of the original image film used for image exposure on the photoresist surface for creating a printed circuit board In addition to image information reading means (13) for reading binary data (BCID), (b) CPU (16), (c) standard virtual of the circuit image of the binary film (BCID) and the original film Memory means (18) for storing continuity data (MVCD) in a callable manner, (d) Standard virtual continuity data (MVCD) of the circuit image of the original film, and binary data (BCID) called from the memory means Based on the reference point detection program for detecting the reference point of the read image, the thinning program for thinning the binary data (BCID) called from the memory means, the expansion for the expansion Program, and memory means for storing a merge program for merging binary data is expanded of the thinned binary data (19) can be reduced to have a (e) monitoring means.
The image information reading means (13) of (a) reads (f) the wiring pattern information for the test board wiring image data (SVCD) of the printed circuit board and converts it into binary data (BCPD). Reading means.
The CPU (16) of (b) compares the standard virtual continuity data (MVCD) of the circuit image of the original film with the test board wiring pattern data (SVCD), and disconnects and shorts the test board wiring. And comparison means for determining potential disconnection and potential short circuit in the wiring.
The memory means (18) of (c) is also a memory means for storing the read binary data (BCPD) of the test substrate pattern and the test substrate wiring pattern data (SVCD) in a callable manner in the CPU. .
The memory means (19) of (d) stores a classification density threshold detection program for calculating a density threshold for classifying the three RGB components of the read test substrate pattern from the background portion. It is also a memory means for storing the callable data in the CPU.
[0080]
【The invention's effect】
As described above, as is clear from the detailed and specific description, according to the present invention, wiring pattern defects in printed circuit boards and LSIs (large scale integrated circuits) and photomask replicas therefor are found by wiring appearance inspection, Not only can you make good and bad judgments easily and quickly, but also virtual conducting to find continuity defects caused by thick and thin wiring that cannot be found by visual inspection while maintaining certain quantitative criteria. The (virtual continuity) test can be performed in the manufacturing process of the printed circuit board, and an excellent pattern matching inspection method and inspection apparatus can be provided.
[Brief description of the drawings]
FIG. 1 is a diagram showing an outline of a binarization method of pattern information of a printed wiring board according to the present invention.
FIG. 2 is a diagram schematically showing a standard skeleton data forming method through thinning of the present invention.
FIG. 3 is an explanatory diagram of a thinning operation using a 3 × 3 logical mask in the present invention.
FIG. 4 is an explanatory diagram of a thinning operation using a 3 × 3 logical mask in the present invention.
FIG. 5 is a detailed explanatory diagram of a thinning operation using a 3 × 3 logical mask in the present invention.
FIG. 6 is a diagram for explaining feature point and reference point detection in the present invention.
FIG. 7 is a diagram illustrating an example of an image including typical noise.
FIG. 8 is a diagram illustrating noise removal processing by a moving average method.
FIG. 9 is a diagram illustrating noise removal processing using a median filter according to the present invention.
FIG. 10 is a diagram illustrating noise removal processing using a median filter according to the present invention.
FIG. 11 is a diagram for explaining an example of using the median filter and the moving average method in the present invention.
FIG. 12 is a diagram illustrating an example of an image including typical sesame salt noise.
FIG. 13 is a diagram conceptually illustrating an example of a processing target image in the present invention.
FIG. 14 is a diagram conceptually illustrating threshold processing of a processing target image in the present invention.
FIG. 15 is a diagram showing a threshold histogram of a processing target image in the present invention.
FIG. 16 is a diagram illustrating a histogram in which the threshold value of the processing target image is adjusted according to the present invention.
FIG. 17 is a diagram showing a process of creating synthesis reference data in the present invention.
FIG. 18 is a diagram showing a synthetic test data creation process according to the present invention.
FIG. 19 is a diagram for explaining a determination of a short circuit or a thickness of a wiring on a test substrate according to the present invention.
FIG. 20 is a diagram for explaining another short circuit / thickness determination of the wiring of the test substrate in the present invention.
FIG. 21 is a diagram showing an inspection first stage process in the present invention.
FIG. 22 is a diagram showing an inspection second stage process in the present invention;
FIG. 23 is a diagram illustrating creation of synthesis reference data according to the present invention.
FIG. 24 is a diagram showing calculation of reference points for composite reference data in the present invention.
FIG. 25 is a diagram for explaining the creation of synthetic test data in the present invention.
FIG. 26 is a diagram showing reference point calculation of composite test data in the present invention.
FIG. 27 is a diagram for explaining difference detection by superimposing the synthesis reference data of the present invention and the synthesis test data.
FIG. 28 is a partially enlarged view of a superimposed image of the synthesis reference data of the present invention and synthesized test data.
FIG. 29 is a diagram for explaining another short circuit / thickness determination of the wiring of the test substrate in the present invention.
FIG. 30 is an example of a histogram for reference threshold values of density standards for three colors of RGB for removing reflected light noise from a test substrate in the present invention.
FIG. 31 is a diagram showing an outline of an example of a printed wiring board inspection apparatus according to the present invention.
[Explanation of symbols]
10 Original pattern film
10a Schatendai
11 Circuit pattern area (image area)
12 Background area
13 Image information reader
13a Optical reading unit
15 computer
16 CPU
17 Monitor
18 Storage device (memory)
19 Storage device (memory)
20 Input means (keyboard)
13a Optical reading unit
131 Image information processing unit
131a Color separation filter
131b Shift register
131c Data latch part
131d decoder
131e control device
131f memory
131g memory

Claims (20)

(A) Standard virtual continuity data (MVCD) of the circuit image of the original image film used for image exposure on the photoresist surface for creating a printed circuit board, and (B) Test board wiring image data of the printed board to be inspected. (C) Comparison operation with (SVCD) determines (D) disconnection and (E) short circuit in the wiring on the test substrate, and (F) potential disconnection and (G) potential short circuit in the wiring A method for inspecting a printed wiring board for determining
(A) The standard virtual continuity data (MVCD) of the circuit image of the original image film (A) includes (a) image information of the circuit image of the original image film used for image exposure on the photoresist surface for creating a printed circuit board. image information reading operation and to binary data (BCID) of the circuit image as a pilot line data read, is thinned in (b) said binary data (BCID) was the circuit image processing thinning circuit image Image thinning operation to form standard skeleton line data (MSLD), and (c) dilation processing of the binary image (BCID) circuit image to form standard dilation image data (MEID) of the circuit image And (d) composite standard image data (CMID) of the circuit image is formed by superposing and synthesizing the standard dilated image data (MEID) on the standard skeleton line data (MSLD). ,composite Based on the quasi image data (CMID), the circuit image, divided from its four corners in each small area including the image elements, in each small region each divided, each of the image elements contained in them An image reference point detection operation in which a center point detected based on circularity is used as a reference point, and a reference point of an image element in each small region is formed as reference point data (SPOD); (e) the composite standard image Obtained by a process including a standard composite reference image data forming operation for forming a composite reference image data (CMSID) by superimposing and combining the reference point data (SPOD) with the data (CMID) of
The test board wiring image data (SVCD) in (B) is (f) a wiring pattern reading operation for reading the wiring pattern information of the printed circuit board to be inspected and converting it into binary data (BCPD); Printed wiring characterized by being obtained by a process including a density threshold value detection operation for calculating a density threshold value for classifying each of the three RGB components of the substrate pattern converted into binary data from the background portion Substrate inspection method.   2. The method for inspecting a printed wiring board according to claim 1, wherein the image reading operation for converting to binary data (BCID) in (a) includes an operation of reading as binary data by a TWAIN compatible scanner.   The image thinning operation for forming the standard skeleton line data (MSLD) in (b) further labels the starting end (p1 point) and the terminal end (p2 point) of the formed thin line, and both ends (p1 point). 2. The printed wiring board inspection method according to claim 1, wherein standard skeleton line data (MSLD) including (p2 point) and the characteristic data of the connected portion is formed. Image thinning operation to form a standard skeleton line data (MSLD) of said (b) is 1 to claim, characterized in that binary data of the circuit image information (BC I D) is for thinning 4. The printed wiring board inspection method according to any one of 3 above. The image thinning operation for forming the standard skeleton line data (MSLD) in (b) forms an image area based on binary data (BCID) obtained by reading from the circuit image of the original image film. 5. The method for inspecting a printed wiring board according to claim 1, further comprising an operation of trimming pixels from the outside to the background area one pixel at a time .   The operation of cutting the pixels forming the image area into the background area bit by bit from the outside is to read the image pixel located at the boundary with the background area from the binary data (BCID) pixels read from the original circuit image. When the center of the matrix of two-dimensionally developed pixels of 3 pixels in a vertical row and 3 pixels in a horizontal row is used as a center, a non-image pixel is determined based on which of the surrounding pixels is the background region pixel and which is the image pixel The rewriting operation is repeated in the alternate direction along the outer periphery of the image until there is no portion whose line width exceeds one pixel by repeating the operation in the alternate direction along the outer periphery of the pattern until there is no parallel part of the image pixel-image pixel in contact with the side The method for inspecting a printed wiring board according to claim 5, further comprising an operation.   The image expansion operation for forming the standard expanded image data (MEID) of the circuit image of (c) recognizes the geometric feature portion of the image in the original of the circuit, and determines the geometry of the image in the original. The method for inspecting a printed wiring board according to claim 1, wherein the inspection is performed without impairing the characteristic features.   The image reference point detection operation of (d) includes detecting the center point of each image element included in each divided small area based on the circularity of each image element. The printed wiring board inspection method according to claim 1, wherein the method is an operation.   (F) The wiring pattern reading operation for converting the wiring pattern of the test substrate into binary data (BCPD) is further performed. (F1) Median filter processing of the read wiring pattern, (f2) Smoothing filter processing, (f3 9. The method for inspecting a printed wiring board according to claim 1, comprising three processes of an isolated noise removing filter process.   (G) For each of the three RGB components of the read substrate pattern, a classification density threshold value detection operation for calculating a density threshold value for classification from the background portion sorts all the pixels constituting each color. Thus, a histogram indicating which of the 256 gradation levels belongs to which density belongs is automatically created, and in this histogram, a pixel aggregate belonging to a certain density area and another pixel aggregate belonging to another density area. The printed wiring board inspection according to any one of claims 1 to 9, wherein the operation is to find a density of a valley having a small number of pixels between and a density threshold value based on the density of the valley. Method. As noise processing, in addition to the three processing operations of (f1) median filter processing of the read wiring pattern, (f2) smoothing filter processing, and (f3) isolated noise removal filter processing, the reflected light of the test substrate is further added. The reflected light noise removing operation includes the operation of removing noise, and the reflected light noise removing operation is performed for any of the detected red pixels of any density for the RGB three-color components of the substrate pattern and (i) for the red R-color component. , Converting to a high density pixel of the pixel set portion belonging to the maximum density value among the 256 levels of density gradation in the histogram, and (ii) for the green G color component, the detected green pixel of any density In the histogram, the green pixel is converted to the density to which the largest pixel group belongs in the region of the intermediate density or higher among the density gradations in the 256 levels in the histogram, while (iii) the blue B color component is detected. By converting the blue pixels of any density into low density pixels of the density value of the pixel set belonging to the lowest density value among the 256 levels of density gradation in the histogram,
The white balance represented by R + G + B = white, which mainly cuts the B color component information, which is the main cause of the non-wiring information in the background area, without cutting the R color pixel component information as an element constituting the wiring pattern. 10. The printed wiring board according to claim 9, wherein the operation is performed to remove reflection noise recognized as white light by removing the white color by removing the B color pixel component from the color balance and destroying the color balance. Inspection method. The standard virtual continuity data (MVCD) of the original circuit image (A) and / or the test board wiring image data (SVCD) of (B) are the four corners of the circuit image of the original image film and / or the test board wiring. image 4 in the corner, as fourth reference point four positions forming a rectangular necessary wide enough to surround the wire image and / or wiring pattern, 1 to claim characterized in that it comprises in binary data 11. The printed wiring board inspection method according to any one of 11 above.   13. The printed wiring board inspection according to claim 12, further comprising the step of determining whether or not there is a defective point other than the wiring formed of the same material as the wiring in a rectangular area having a width surrounding the wiring pattern. Method.   In the comparison operation of (C), at least one of the standard virtual continuity data of the circuit original of (A) and the test board wiring image data of the printed circuit board of (B) is partially rotated or tilted. 13. The method for inspecting a printed wiring board according to claim 1, further comprising a process of correcting, expanding / contracting correction, and / or superimposing the coordinates by adjusting.   The disconnection in the wiring on the test substrate in (D) is caused by superimposing the binary data (BCID) according to (a) in (A) and the test substrate wiring pattern data in (B). 15. The method for inspecting a printed wiring board according to claim 14, wherein when the comparison is made, it is detected as a surplus portion of the binary data (BCID) according to (a).   The short circuit in the wiring on the test substrate in (E) causes the binary data (BCID) according to (a) in (A) to overlap the test substrate wiring pattern data in (B). The printed wiring board inspection method according to claim 14, wherein when the comparison is made, the binary data (BCID) according to (a) is detected as a missing portion.   The potential disconnection in the wiring on the test substrate in (F) includes the standard skeleton line data (MSLD) according to (b) in (A) and the test substrate wiring pattern data in (B). 15. The method for inspecting a printed wiring board according to claim 14, wherein when the comparison for superimposition is performed, it is detected as an extra portion of the standard skeleton line data (MSLD) according to (b).   The potential short circuit in the wiring on the test substrate in (G) overlaps the standard expansion image data (MEID) according to (c) in (A) and the test substrate wiring pattern data in (B). 15. The method for inspecting a printed wiring board according to claim 14, wherein when the comparison for matching is performed, a defect portion of the standard expansion image data (MEID) according to (c) is detected.   A comparison for superimposing the standard expansion image data (MEPD) according to (c) in (A) and the test substrate wiring pattern data in (B) is performed by comparing the geometry of the image in the original of the circuit. Including matching between the characteristic features (q1, q2,... Qn) and corresponding points (q1 ′, q2 ′,... Qn ′) in the test substrate wiring pattern data of (B). The printed wiring board inspection method according to claim 18. (A) Standard virtual continuity data (MVCD) of the circuit image of the original image film used for image exposure on the photoresist surface for creating a printed circuit board, and (B) Test board wiring pattern data of the printed circuit board to be inspected (C) Comparison with (SVCD) determines (D) disconnection and (E) short circuit in the wiring on the test substrate, and (F) potential disconnection and (G) potential short circuit in the wiring An inspection apparatus for a printed wiring board for judging,
(A) For the standard virtual continuity data (MVCD) of the circuit image of the original image film of (A), (a) The image of the circuit image of the original image film used for image exposure on the photoresist surface for the production of a printed circuit board Image information reading means (13) for reading information into binary data (BCID), (b) CPU, (c), binary data (BCID), standard virtual continuity data of circuit image of the original film Memory means (18) for storing (MVCD) in a callable manner, (d) Standard virtual continuity data (MVCD) of the circuit image of the original film, binary data (BCID) called from the memory means (18 ) based on, the read image reference point detecting program for detecting a reference point, said memory means (18) binary data (BCID) thinning program for thinning called from the memory Storage expansion programs for expanding the binary data (BCID) called from stage (18), a merge program for superposing synthesized thinned binary data is expanded of the binary data has been Memory means (19), and (e) monitoring means,
The image information reading means (13) of (a) reads (f) test board wiring pattern information for binary circuit data (BCPD) for the test board wiring image data (SVCD) of the printed circuit board . It is also the image information reading means for,
The CPU in (b) compares the standard virtual continuity data (MVCD) of the circuit image of the original film with the test board wiring pattern data (SVCD) to determine disconnection and short circuit of the test board wiring. And a comparison means for determining a potential disconnection and a potential short circuit in the wiring,
The memory means (18) of (c) is a memory means for storing the read binary data (BCPD) of the test substrate pattern and the test substrate wiring pattern data (SVCD) in a callable manner in the CPU. ,
The memory means (19) of (d) stores a classification density threshold detection program for calculating a density threshold for classifying the three RGB components of the read test substrate pattern from the background portion. A printed circuit board inspection apparatus, characterized in that it is a memory means that is stored in a callable manner in a CPU.

Images