JP4380205B2 - Image collation device and image collation method - Google Patents

Description

【００２１】
【数２】

【００２２】
対数変換部２１１２１，２１１２２は、フーリエ変換部２１１１１，２１１１２で生成された、フーリエ画像データＦ１（ｍ，ｎ），Ｆ２（ｍ，ｎ）の振幅成分に基いて対数処理を行う。この振幅成分の対数処理は、画像データの詳細な特徴情報を含む高周波成分を強調する。
【００２３】
詳細には、対数変換部２１１２１は数式（３）に示すように振幅成分Ａ（ｕ，ｖ）に基いて対数処理を行いＡ’（ｕ，ｖ）を生成し対数−極座標変換部２１１３１に出力する。対数変換部２１１２２は数式（４）に示すように振幅成分Ｂ（ｕ，ｖ）に基いて対数処理を行いＢ’（ｕ，ｖ）を生成し対数−極座標変換部２１１３２に出力する。
【００２４】
【数３】

【００２５】
【数４】

【００２６】
対数−極座標変換部２１１３１，２１１３２は、対数変換部２１１２１，２１１２２から出力された信号に基いて、対数−極座標系（例えばｌｏｇ−ｒ，Θ）に変換する。
一般的に例えば点（ｘ，ｙ）に対して数式（５），（６）に示すように定義すると、ｒ＝ｅ^μの場合にはμ＝ｌｏｇ（ｒ）であり、任意の点（ｘ，ｙ）に対応する一義的な（ｌｏｇ（ｒ），Θ）が存在する。この性質により対数−極座標変換部２１１３１，２１１３２は座標変換を行う。
【００２７】
【数５】

【００２８】
【数６】

【００２９】
詳細には、対数−極座標変換部２１１３１，２１１３２は、数式（７）に示す集合（ｒ_ｉ、θ_ｊ）、および数式（８）に示す関数ｆ（ｒ_ｉ、θ_ｊ）を定義する。
【００３０】
【数７】

【００３１】
【数８】

【００３２】
対数−極座標変換部２１１３１，２１１３２は、数式（７），（８）で定義した集合（ｒ_ｉ、θ_ｊ），関数ｆ（ｒ_ｉ、θ_ｊ）を用いて、画像データＡ’（ｕ，ｖ），Ｂ’（ｕ，ｖ）それぞれを、数式（９），（１０）に示すように対数−極座標変換を行い、ｐＡ（ｒ_ｉ、θ_ｊ）、ｐＢ（ｒ_ｉ、θ_ｊ）を生成し、それぞれ信号ＳＡ２１１、信号ＳＲ２１１として、位相限定相関部２１２に出力する。
【００３３】
【数９】

【００３４】
【数１０】

【００３５】
図３は、図２に示したフーリエ・メリン変換部２１１の動作を説明するための図である。
画像ｆ１（ｍ，ｎ），画像ｆ２（ｍ，ｎ）は、例えばｘ，ｙ軸に対して異なる所定角度を持った矩形領域Ｗ１，Ｗ２を含む。
フーリエ・メリン変換部２１１において、例えば図３に示すように、画像ｆ１（ｍ，ｎ）が、フーリエ変換部２１１１１によりフーリエ変換されて、フーリエ画像データＡ’（ｕ，ｖ）が生成され、対数変換部２１１２１および対数−極座標変換部２１１３１により、画像データｐＡ（ｒ，θ）が生成される。
【００３６】
同様に、画像ｆ２（ｍ，ｎ）が、フーリエ変換部２１１１２によりフーリエ変換されて、フーリエ画像データＢ’（ｕ，ｖ）が生成され、対数変換部２１１２２および対数−極座標変換部２１１３２により、画像データｐＢ（ｒ，θ）が生成される。
【００３７】
上述したように、画像ｆ１（ｍ，ｎ），ｆ２（ｍ，ｎ）は、フーリエ変換および対数−極座標変換により、デカルト座標から対数−極座標系（フーリエ・メリン空間とも言う）上に変換される。
フーリエ・メリン空間では、画像のスケーリングに応じて、成分がｌｏｇ−ｒの軸に沿って移動し、画像の回転角度に応じてθ軸に沿って移動する性質がある。
この性質を用いて、画像ｆ１（ｍ，ｎ），ｆ２（ｍ，ｎ）のスケーリング（倍率情報）および回転角度を、フーリエ・メリン空間上のｌｏｇ−ｒの軸に沿った移動量、およびθ軸に沿った移動量に基いて、求めることができる。
【００３８】
位相限定相関部２１２は、例えば位相限定フィルタ（ＳＰＯＭＦ：Ｓｙｍｍｅｔｒｉｃｐｈａｓｅ−ｏｎｌｙ ｍａｔｃｈｅｄ ｆｉｌｔｅｒ）を用いた位相限定相関法により、フーリエ・メリン変換部２１１から出力されたパターンデータを示す信号ＳＡ２１１およびＳＲ２１１に基いて、それぞれの平行移動量を求める。
位相限定相関部２１２は、例えば図１に示すように、フーリエ変換部２１２０，２１２１、合成部２１２２、位相抽出部２１２３、および逆フーリエ変換部２１２４を有する。
【００３９】
フーリエ変換部２１２０，２１２１は、対数−極座標変換部２１１３１，２１１３２から出力された信号ＳＡ２１１（ｐＡ（ｍ，ｎ）），ＳＲ２１１（ｐＢ（ｍ，ｎ））に基いて数式（１１），（１２）により、フーリエ変換を行う。ここで、Ｘ（ｕ，ｖ），Ｙ（ｕ，ｖ）は、フーリエ係数である。フーリエ係数Ｘ（ｕ，ｖ）は数式（１１）に示すように振幅スペクトルＣ（ｕ，ｖ）および位相スペクトルθ（ｕ，ｖ）により構成される。フーリエ係数Ｙ（ｕ，ｖ）は数式（１２）に示すように振幅スペクトルＤ（ｕ，ｖ）および位相スペクトルφ（ｕ，ｖ）により構成される。
【００４０】
【数１１】

【００４１】
【数１２】

【００４２】
合成部２１２２は、フーリエ変換部２１２０，２１２１で生成されたＸ（ｕ，ｖ），Ｙ（ｕ，ｖ）を合成して相関をとる。例えば合成部２１２２は、Ｘ（ｕ，ｖ）・Ｙ^＊（ｕ，ｖ）を生成し、位相抽出部２１２３に出力する。ここで、Ｙ^＊（ｕ，ｖ）は、Ｙ（ｕ，ｖ）の複素共役である。
【００４３】
位相抽出部２１２３は、合成部２１２２から出力された合成信号に基いて振幅成分を除去して位相情報を抽出する。
例えば位相抽出部２１２３は、例えばＸ（ｕ，ｖ）Ｙ^＊（ｕ，ｖ）に基いて、その位相成分Ｚ（ｕ，ｖ）＝ｅ^{ｊ（θ（ｕ，ｖ）−φ（ｕ，ｖ））}を抽出する。
【００４４】
位相情報の抽出は、上述した形態に限られるものではない。例えば、フーリエ変換部２１２０，２１２１の出力、数式（１３），（１４）に基いて位相情報を抽出した後、数式（１５）に示すように位相成分のみ合成を行い、Ｚ（ｕ，ｖ）を生成してもよい。
【００４５】
【数１３】

【００４６】
【数１４】

【００４７】
【数１５】

【００４８】
逆フーリエ変換部２１２４は、位相抽出部２１２３から出力された、位相情報のみの信号Ｚ（ｕ，ｖ）に基いて、逆フーリエ変換処理を行い、相関強度画像を生成する。
詳細には、逆フーリエ変換部２１２４は、数式（１６）に示すように、信号Ｚ（ｕ，ｖ）に基いて逆フーリエ変換処理を行い、相関強度画像Ｇ（ｐ，ｑ）を生成する。
【００４９】
【数１６】

【００５０】
倍率情報−回転情報生成部２１３は、逆フーリエ変換部２１２４により生成された相関強度画像Ｇ（ｐ、ｑ）におけるピーク位置の画像中心からのずれ量が、すなわち登録画像ＡＩＭと、照合画像ＲＩＭに対してフーリエ・メリン変換を行った結果得られたパターンデータ間の平行移動量と等価であるので、このずれ量を検出することにより、登録画像ＡＩＭに対する照合画像ＲＩＭの倍率情報（拡大／縮小率）および回転角度情報を示すデータを含む補正情報Ｓ２１を生成する。
【００５１】
補正部２２は、倍率情報−回転情報部２１の倍率情報−回転情報生成部２１３から出力された補正情報Ｓ２１に基いて、照合画像ＲＩＭの補正を行う。詳細には、補正部２２は、補正情報Ｓ２１に含まれる倍率情報および回転角度情報に基づいて、照合画像ＲＩＭを拡大／縮小処理し、回転処理を行い、位相限定相関部２３に出力する。補正部２２の補正処理により、登録画像ＡＩＭと照合画像ＲＩＭ間のスケーリングおよび回転成分の差異が除去される。
このため、登録画像ＡＩＭと、補正処理された照合画像ＲＩＭとの間には、平行移動成分のみが差異として残っている。
【００５２】
位相限定相関部２３は、上述した登録画像ＡＩＭと、補正処理された照合画像ＲＩＭとの間の平行移動成分、およびその相関値を検出する。この検出は、例えば上述した位相限定フィルタを用いた位相限定相関法により求める。
詳細には、位相限定相関部２３は、フーリエ変換部２３１１，２３１２、合成部２３２、位相抽出部２３３、および逆フーリエ変換部２３４を有する。
【００５３】
フーリエ変換部２３１１，２３１２、合成部２３２、位相抽出部２３３、および逆フーリエ変換部２３４それぞれは、上述した位相限定相関部２１２のフーリエ変換部２１２０，２１２１、合成部２１２２、位相抽出部２１２３、および逆フーリエ変換部２１２４それぞれと同じ機能を有するので簡単に説明する。
【００５４】
フーリエ変換部２３１１は、登録画像ＡＩＭをフーリエ変換し合成部２３２に出力する。この際、予めフーリエ変換部２１１１１でフーリエ変換処理した登録画像ＡＩＭをメモリ１２に記憶しておき、それを合成部２３２に出力してもよい。こうすることにより２重にフーリエ変換処理を行うことがないために、処理が軽減される。
フーリエ変換部２３１２は、補正部２２により、補正された画像Ｓ２２をフーリエ変換を行い、処理結果のフーリエ画像を合成部２３２に出力する。
【００５５】
合成部２３２は、フーリエ変換部２３１１，２３１２から出力されたフーリエ画像Ｓ２３１１，Ｓ２３１２を合成し、合成画像Ｓ２３２を位相抽出部２３３に出力する。
位相抽出部２３３は，合成画像Ｓ２３２に基づいて上述したように位相情報を抽出して信号Ｓ２３３を逆フーリエ変換部２３４に出力する。
逆フーリエ変換部２３４は、信号Ｓ２３３に基づいて逆フーリエ変換を行い相関強度画像（相関画像データ）を生成し、信号Ｓ２３として判別部２４に出力する。
【００５６】
上述の位相限定相関法における平行移動量の検出処理の詳細な説明を行う。
例えば、原画像ｆ１（ｍ，ｎ）、原画像ｆ２（ｍ，ｎ）、および画像ｆ２（ｍ，ｎ）を平行移動した画像ｆ３（ｍ，ｎ）＝ｆ２（ｍ＋α，ｎ＋β）それぞれをフーリエ変換処理し、数式（１７）〜（１９）に示すように、フーリエ係数Ｆ１（ｕ，ｖ），Ｆ２（ｕ，ｖ），Ｆ３（ｕ，ｖ）を生成する。
【００５７】
【数１７】

【００５８】
【数１８】

【００５９】
【数１９】

【００６０】
フーリエ係数Ｆ１（ｕ，ｖ）〜Ｆ３（ｕ，ｖ）に基づいて、数式（２０）〜（２２）に示すように、位相情報のみの位相画像Ｆ’１（ｕ，ｖ）〜Ｆ’３（ｕ，ｖ）を生成する。
【００６１】
【数２０】

【００６２】
【数２１】

【００６３】
【数２２】

【００６４】
位相画像Ｆ’１（ｕ，ｖ）と位相画像Ｆ’２（ｕ，ｖ）の相関の位相画像の相関Ｚ１２（ｕ，ｖ）を数式（２３）に示すように計算し、位相画像Ｆ’１（ｕ，ｖ）と位相画像Ｆ’３（ｕ，ｖ）の相関の位相画像の相関Ｚ１３（ｕ，ｖ）を数式（２４）に示すように計算する。
【００６５】
【数２３】

【００６６】
【数２４】

【００６７】
相関Ｚ１２（ｕ，ｖ）の相関強度画像Ｇ１２（ｒ，ｓ）、および相関Ｚ１３（ｕ，ｖ）の相関強度画像Ｇ１３（ｒ，ｓ）を数式（２５），（２６）に示すように計算する。
【００６８】
【数２５】

【００６９】
【数２６】

【００７０】
数式（２５），（２６）に示すように、画像ｆ３（ｍ，ｎ）が、画像２（ｍ，ｎ）に比べて（＋α、＋β）だけずれている場合には、位相限定相関法では、相関強画像において、（−α，−β）だけずれた位置に相関強度のピークが生成される。この相関強度位置の位置により、２つの画像間の平行移動量を求めることができる。
また、上述したフーリエ・メリン空間上で、この位相限定相関法を用いることにより、フーリエ・メリン空間上の平行移動量が検出できる。この平行移動量は、上述したように実空間では倍率情報および回転角度情報に相当する。
【００７１】
図４は、自己相関法と位相限定相関法の相違点を説明するための図である。
自己相関法では、例えば図４（ａ），（ｂ）に示すように、画像ＩＭ１、および画像ＩＭ１と同じ画像ＩＭ２をフーリエ変換を行い自己相関関数ＳＧ１を生成すると、図４（ｃ）に示すように、相関強度が高いピークと、その周辺部に小さい相関強度を有する相関強度分布が得られる。図４（ｃ）において縦軸は相関強度を示す。
【００７２】
一方、上述した位相限定相関法では、図４（ｄ），（ｅ）に示すように、画像ＩＭ１、および画像ＩＭ１と同じ画像ＩＭ２をフーリエ変換を行い、位相情報のみを相関すると図４（ｆ）に示すように、相関強度が高く鋭いピークのみを有する相関強度分布が得られる。このように位相限定法では自己相関法に比べて相関に関して明確な情報を得ることができる。図４（ｆ）において、縦軸（ｚ軸）は相関強度を示し、ｘ軸，ｙ軸はずれ量を示す。
【００７３】
図５は、位相限定相関法において、２つの画像間で平行移動ずれがある場合の相関強度分布を説明するための図である。
例えば図５（ａ），（ｂ）に示すような画像ＩＭ１、および画像ＩＭ１より数画素平行移動している画像ＩＭ３の画像の位相限定法における相関強度分布は、例えば図５（ｃ）に示すように、相関強度が高く鋭いピークが、図４（ｆ）に示した相関画像データの内のピーク位置から、平行移動量に応じた距離だけずれた位置に分布している。しかし、図５（ｃ）に示すピーク強度は、図４（ｆ）に示したピーク強度に比べて小さい。これは画像ＩＭ１，ＩＭ２に比べて、画像ＩＭ１，ＩＭ３の一致している画素領域が小さいためである。
【００７４】
図６は、位相限定法において、２つの画像間で回転ずれがある場合の相関強度分布を説明するための図である。
図６（ａ），（ｂ）に示すような画像ＩＭ１、および画像ＩＭ１より数度回転している画像４の位相限定法における相関強度分布は、例えば図６に示すように、弱い相関強度の相関強度分布が得られる。単純に位相限定相関法を用いた場合には回転ずれにより、相関を検出することが困難である。
【００７５】
このため、本実施形態に係る画像照合装置１では、登録画像ＡＩＭおよび照合画像ＲＩＭを、フーリエ・メリン変換を行い、フーリエ・メリン空間上で、位相限定相関法を用いることにより、平行移動量を検出し、登録画像ＡＩＭおよび照合画像ＲＩＭの倍率情報、および回転角度情報を検出する。その情報に基づいて、照合画像の倍率および回転角度を補正する。
補正した照合画像ＲＩＭおよび登録画像ＡＩＭ間の平行移動ずれを、位相限定相関法により検出し、同時に相関ピークに基づいて画像間の照合を行う。
【００７６】
図７は、位相限定相関部２３が出力する相関画像データを説明するための図である。図７（ａ）は登録画像ＡＩＭ、図７（ｂ）は照合画像ＲＩＭ、図７（ｃ）は相関画像データの一具体例を示す図である。図８は図７（ｃ）に示した相関画像データを説明するための図である。
【００７７】
登録画像ＡＩＭ，照合画像ＲＩＭとして、例えば手書き文字や血管パターン等の２値化および線形状のパターンを含む場合、例えば図７（ａ），（ｂ）に示すような登録画像ＡＩＭ，照合画像ＲＩＭの場合、位相限定相関部２３は信号Ｓ２３として図７（ｃ）に示すような相関画像データを出力する。
【００７８】
図７（ｃ）において、ｚ軸方向は相関画像データにおける相関ピーク強度を示し、相関ピーク強度は登録画像ＡＩＭと照合画像ＲＩＭ間の相関度に相当する。図７（ｃ）および図８に示すように、ｘ軸方向，ｙ軸方向での相関ピーク位置は、例えば画像中心からのずれ量が登録画像ＡＩＭと照合画像ＲＩＭ間の平行移動量に相当する。
【００７９】
判別部２４は、位相限定相関部２３から出力された信号Ｓ２３に基づいて、登録画像ＡＩＭと照合画像ＲＩＭの照合を行う。
判別部２４は、例えば図２に示すように、候補特定部２４１、類似度生成部２４２、積算部２４３、および照合部２４４を有する。
候補特定部２４１は本発明に係る候補特定手段に相当し、類似度生成部２４２は本発明に係る類似度生成手段に相当し、積算部２４３は本発明に係る積算手段に相当し、照合部２４４は本発明に係る照合手段に相当する。
【００８０】
候補特定部２４１は、位相限定相関部２３から出力された信号Ｓ２３に基づいて、登録画像ＡＩＭと照合画像ＲＩＭと２次元上の位置関係の複数の候補を特定し、信号Ｓ２４１として類似度生成部２４２に出力する。
【００８１】
図９は、図１に示した候補特定部２４１の候補特定処理を説明するための図である。数値は、相関画像データのＸ−Ｙ面上での相関画像データの相関ピーク強度を示す。
例えば図７（ａ），（ｂ）に示すような２値化および線形状のパターンを含む登録画像ＡＩＭ，照合画像ＲＩＭの場合、相関の大きい画像同士でも、相関ピーク強度（相関強度とも言う）が図９に示すように値が小さい。
候補特定部２４１は、例えば図９に示すように、信号Ｓ２３に基づいて相関強度の上位Ｎ個、本実施形態では８個を、登録画像ＡＩＭと照合画像ＲＩＭと２次元上の位置関係の候補として特定し、信号Ｓ２４１として類似度生成部２４２に出力する。
【００８２】
類似度生成部２４２は、候補特定部２４２で特定された複数の位置関係の候補を示す信号Ｓ２４２に基づいて、登録画像ＡＩＭと、補正部２２で補正された照合画像ＲＩＭとの間の２次元上の異なる複数の位置関係それぞれにおいて、画素毎に、登録画像ＡＩＭと、補正部２２で補正された照合画像ＲＩＭを比較して類似度を生成する。
【００８３】
詳細には、類似度生成部２４２は、例えば図７（ａ），（ｂ）に示すように、登録画像ＡＩＭをｆ１（ｍ，ｎ）、補正部２２で補正された照合画像ＲＩＭをｆ２（ｍ，ｎ）のｍ×ｎ画素の２値化および線形状のパターンを含む画像とすると、例えば類似度Ｓｉｍを数式（２７）により算出し、算出結果を積算部２４３および照合部２４４に出力する。
【００８４】
【数２７】

【００８５】
積算部２４３は、類似度生成部２４２が生成した類似度Ｓｉｍを積算し、積算結果を信号Ｓ２４３として照合部２４４に出力する。
【００８６】
照合部２４４は、積算部２４３が積算した類似度Ｓｉｍを基に登録画像ＡＩＭと照合画像ＲＩＭの照合を行う。詳細には、例えば照合部２４４は、積算した類似度Ｓｉｍが所定値よりも大きい場合には、登録画像ＡＩＭと照合画像ＲＩＭが一致していると判別する。
【００８７】
また、照合部２４４は、類似度生成部２４２から出力された類似度Ｓｉｍを基に登録画像ＡＩＭと照合画像ＲＩＭの照合を行う。詳細には、例えば照合部２４４は、類似度Ｓｉｍが所定値よりも大きい場合には、登録画像ＡＩＭと照合画像ＲＩＭが一致していると判別する。
また、照合部２４４は、照合処理の結果を示す信号Ｓ２４を出力する。
【００８８】
図１０は、図１に示した画像照合装置１の動作を説明するためのフローチャートである。以上説明した構成の画像照合装置１の動作を図８を参照しながら簡単に説明する。
例えば画像入力部１１により、登録画像ＡＩＭおよび照合画像ＲＩＭが入力され、メモリにそれぞれの画像データが格納される（ＳＴ１）。
ここで、登録画像ＡＩＭに対する照合画像ＲＩＭの倍率（拡大／縮小率）および回転角度情報を求めるために、登録画像ＡＩＭがメモリ１２から読み出され（ＳＴ２）、倍率情報−回転情報部２１のフーリエ変換部２１１１１により、フーリエ変換処理され（ＳＴ３）、フーリエ画像データＳ２１１１１がメモリ１２に格納、記憶される（ＳＴ４）。
フーリエ画像データＳ２１１１１の内の振幅成分は、対数変換部２１１２１により対数処理が行われ、対数−極座標変換部２１１３１により、対数−極座標系に変換される（ＳＴ５）。
【００８９】
照合画像ＲＩＭが、メモリ１２から読み出され（ＳＴ６）、同様にフーリエ変換部２１１１２によりフーリエ変換処理され（ＳＴ７）、フーリエ画像データＳ２１１１２の内の振幅成分が、対数変換部２１１２２により対数処理が行われ、対数−極座標変換部２１１３２により、対数−極座標系に変換される（ＳＴ８）。
【００９０】
上述した登録画像ＡＩＭおよび照合画像ＲＩＭにフーリエ・メリン変換を行った結果得られた、画像信号（パターンデータとも言う）ＳＡ２１１、ＳＲ２１１それぞれは、位相限定相関部２１２のフーリエ変換部２１２０，２１２１によりフーリエ変換処理され（ＳＴ９）、合成部２１２２により合成され（ＳＴ１０）、位相抽出部２１２３により合成信号から振幅成分が除去され（ＳＴ１１）、残りの位相成分が、逆フーリエ変換部２１２４により逆フーリエ変換処理され（ＳＴ１２）、得られた相関画像データのピーク位置の画像中心からのずれ量に基づいて、倍率情報−回転情報生成部２１３により倍率情報および回転情報を含む補正情報が生成される（ＳＴ１３）。
【００９１】
補正部２２では、補正情報に基づいて照合画像の拡大／縮小および回転処理の補正処理が行われ、画像間のスケーリング成分、および回転成分が除去される（ＳＴ１４）。残る差異は平行移動成分のみであり、位相限定相関法を用いて検出される。
【００９２】
補正処理が行われた照合画像ＲＩＭは、位相限定相関部２３のフーリエ変換部２３１２によりフーリエ変換されて（ＳＴ１５）フーリエ画像データＳ２３１２が生成され、メモリ１２に格納されたフーリエ変換された登録画像ＡＩＭが読み出され（ＳＴ１６）、合成部２３２により合成データＳ２３２が生成される（ＳＴ１７）。
【００９３】
この際、登録画像ＡＩＭがフーリエ変換部２３１１によりフーリエ変換されてフーリエ画像データＳ２３１１が生成されて合部部２３２に入力されてもよい。合成データＳ２３２の内の振幅情報が位相抽出部２３３により除去され（ＳＴ１８）、残りの位相情報が逆フーリエ変換部２３４に入力され、相関画像データとして信号Ｓ２３が出力される（ＳＴ１９）。
【００９４】
上述した処理により生成された相関画像データＳ２３に基づいて、候補特定部２４１により相関画像データＳ２３における相関ピーク強度の上位側から、例えばＮ個の候補Ｐ_ｉ（Ｐ_０，Ｐ_１，Ｐ_２，…，Ｐ_ｎ−１）が抽出され、信号Ｓ２４１として類似度生成部２４２に出力される（ＳＴ２０）。
【００９５】
積算部２４３は、積算のための変数を初期化する（ＳＴ２１）。例えば変数ｉを０、積算値Ｓを０に初期化する。
類似度生成部２４２では、各候補（座標）Ｐ_ｉについて、相関画像データの中心からのずれ量をそれぞれ検出することで、登録画像ＡＩＭ，照合画像ＲＩＭの位置合わせを行った後に類似度Ｓｉｍ（ｉ）を算出し信号Ｓ２４２として積算部２４３および判別部２４４に出力する（ＳＴ２２）。
【００９６】
ステップＳＴ２３において、照合部２４４では、類似度Ｓｉｍ（ｉ）と、予め設定した第１の閾値ｔｈ１とを比較し、類似度Ｓｉｍ（ｉ）が第１の閾値より小さい場合には、積算部２４３では、類似度Ｓｉｍ（ｉ）を積算し、詳細には数式Ｓ＝Ｓ＋Ｓｉｍ（ｉ）により積算し照合部２４４に出力する（ＳＴ２４）。
ステップＳＴ２５において、照合部２４４では、積算値Ｓと予め設定した第２の閾値ｔｈ２とを比較し、積算値Ｓが第２の閾値ｔｈ２よりも小さい場合には、変数ｉと値Ｎ−１とが比較され（ＳＴ２６）、変数ｉがＮ−１と一致していない場合には、変数ｉに１加算し（ＳＴ２７）、ステップＳＴ２２の処理に戻る。ステップＳＴ２５において、変数ｉがＮ−１と一致した場合には、画像が不一致であるとする（ＳＴ２８）。
【００９７】
一方、ステップＳＴ２３の比較処理において、照合部２４４では、類似度Ｓｉｍ（ｉ）が第１の閾値以上の場合には、画像が一致していると判別し、また、ステップＳＴ２５の比較処理において、照合部２４４では、積算値Ｓが第２の閾値ｔｈ２以上の場合には、画像が一致しているとし（ＳＴ２９）、例えばセキュリティ分野における静脈パターン照合装置に、本実施形態に係る画像照合装置を適用した場合には、電子錠を解除するといった処理を動作処理部１６が行う。
【００９８】
以上説明したように、登録画像ＡＩＭと照合画像ＲＩＭとの２次元上の位置関係の複数の候補を特定する候補特定部２４１と、登録画像ＡＩＭと照合画像ＲＩＭとの間の２次元上の異なる複数の位置関係それぞれにおいて、登録画像ＡＩＭと照合画像ＲＩＭとを比較して類似度Ｓｉｍを生成する類似度生成部２４２と、類似度生成部２４２が生成した類似度Ｓｉｍを積算する積算部２４３と、積算部２４３が積算した類似度Ｓｉｍを基に照合を行う照合部２４４とを設けたので、例えば、比較対照を行う２枚の画像データ間おの相関が小さい場合であっても、複数の各候補の位置関係それぞれについて算出される類似度を加算することにより、類似度単独で照合を行う場合に比べて、高精度に照合を行うことができる。また、類似度Ｓｉｍが第１の閾値ｔｈ１よりも大きい場合には一致していると判別するので、高速に照合処理を行うことができる。
【００９９】
また、登録画像ＡＩＭと照合画像ＲＩＭとの回転角度補正処理または拡大率補正処理、およびフーリエ変換処理の結果の位相成分に基づいて相関を検出する倍率情報−回転情報部２１、補正部２２、および位相限定相関部２３を設けたので、登録画像ＡＩＭと照合画像ＲＩＭとが回転角度や拡大率が異なる場合であっても、照合処理を行うことができる。
【０１００】
図１１は、本発明に係る画像照合装置の第２実施形態のソフトウェア的な機能ブロック図である。
本実施形態に係る画像照合装置１ａのハード的な機能ブロックは、第１実施形態とほぼ同じであるので説明を省略する。
画像照合装置１ａのＣＰＵ１５ａは、例えば図１１に示すように、倍率情報−回転情報部２１、補正部２２、位相限定相関部２３、および判別部２４ａを有する。
判別部２４ａは、例えば図１１に示すように、候補特定部２４１、類似度生成部２４２、類似度補正部２４２１、積算部２４３、および照合部２４４を有する。
第１実施形態と同様な機能については説明を省略し、相違点のみ説明する。
【０１０１】
類似度補正部２４２１は、類似度生成部２４２が、登録画像ＡＩＭと照合画像ＲＩＭとの２次元上の異なる複数の位置関係それぞれにおいて、登録画像ＡＩＭと照合画像ＲＩＭの画素毎に比較して生成した類似度Ｓｉｍを、当該画素の周囲の画像特性に応じて補正し類似度Ｓｉｍ’を生成し、信号Ｓ２４２１として積算部２４３および照合部２４４に出力する。
積算部２４３および照合部２４４は、類似度Ｓｉｍ’に基づいて第１実施形態と同様な処理を行う。
【０１０２】
図１２は、図１に示した類似度補正部２４２１の機能を説明するための図である。図１２（ａ）は登録画像ＡＩＭの画像パターン、図１２（ｂ）は照合画像ＲＩＭの画像パターン、図１２（ｃ）は類似度生成部２４２が生成する類似度Ｓｉｍの一具体例を説明するための図である。図１３は、図１に示した類似度補正部２４２１の機能を説明するための図である。
【０１０３】
例えば類似度生成部２４２が生成する類似度Ｓｉｍは、例えば図１２（ａ）に示す登録画像ＡＩＭの画像パターンと、図１２（ｂ）示す照合画像ＲＩＭの画像パターンの類似度Ｓｉｍを生成する場合には、図１２（ｃ）に示すように、両画像の線形状のパターンの交点ＣＰに基づいて算出する。
【０１０４】
類似度生成部２４２は、例えば図１３（ａ）に示すように、登録画像ＡＩＭのパターンＰＴ１１と、照合画像ＲＩＭのパターンＰＴ２１との相関パターンＰＴ３１の場合に、線形状のパターンの方向が異なり相関がないにもかかわらず、交点のみにより類似度Ｓｉｍを算出するために、類似度が上昇してしまう。
本実施形態に係る類似度補正部２４２１は、例えば交点ＣＰを検出し、例えばその画素の周囲の９画素において中心部以外にも交点ＣＰがある場合には中心部の交点ＣＰを相関のある点と判別し、中心部以外に交点ＣＰのない場合には中心部の交点ＣＰを相関のない交点として判別する。
【０１０５】
例えば類似度補正部２４２１は、図１３（ａ）に示すように登録画像ＡＩＭのパターンＰＴ１１と、照合画像ＲＩＭのパターンＰＴ２１との相関パターンＰＴ３１の場合に、中心部以外に交点ＣＰがないので、中心部の交点を相関のない交点として判別する。
類似度補正部２４２１は、図１３（ｂ）に示すように登録画像ＡＩＭのパターンＰＴ１２と、照合画像ＲＩＭのパターンＰＴ２２との相関パターンＰＴ３２の場合に、中心部以外に交点ＣＰがないので、中心部の交点を相関のない交点として判別する。
【０１０６】
類似度補正部２４２１は、図１３（ｃ）に示すように登録画像ＡＩＭのパターンＰＴ１２と、照合画像ＲＩＭのパターンＰＴ２２との相関パターンＰＴ３２の場合に、中心部以外に交点ＣＰがあるので、中心部の交点ＣＰを相関のある交点として判別する。
【０１０７】
類似度補正部２４２１は、画像全体について求めた相関のある交点数を、全交点数で割った値を有効率ｅｆｆとし、数式（２８）に示すように、類似度生成部２４２が生成した類似度Ｓｉｍに乗算して補正処理を行い、補正した類似度Ｓｉｍ’を生成し、信号Ｓ２４２１として出力する。
【０１０８】
【数２８】

【０１０９】
例えば、図１２に示したような場合には、有効率ｅｆｆが０であり、その結果補正した類似度Ｓｉｍ’は０である。
【０１１０】
図１４は、図１１に示した画像照合装置１ａの動作を説明するためのフローチャートである。以上説明した構成の画像照合装置１の動作を図１４を参照しながら、第１実施形態との相違点のみ説明する。ステップＳＴ１〜ステップＳＴ２０の処理は第１実施形態と同様であり説明を省略する。
【０１１１】
積算部２４３は、積算のための変数を初期化する（ＳＴ２１）。例えば変数ｉを０、積算値Ｓを０に初期化する。
類似度生成部２４２では、各候補（座標）Ｐ_ｉについて、相関画像データの中心からのずれ量をそれぞれ検出することで、登録画像ＡＩＭ，照合画像ＲＩＭの位置合わせを行った後に類似度Ｓｉｍ（ｉ）を算出し信号Ｓ２４２として、類似度補正部２４２１に出力する（ＳＴ２２）。
【０１１２】
ステップＳＴ２２１において、類似度補正部２４２１は、類似度生成部２４２から出力された類似度Ｓｉｍとしての補正信号Ｓ２４２に基づいて、画像全体について求めた相関のある交点数を、全交点数で割った値を有効率ｅｆｆとし、数式（２８）に示すように、類似度生成部２４２が生成した類似度Ｓｉｍに乗算して補正処理を行い、補正した類似度Ｓｉｍ’を生成し、信号Ｓ２４２１として、照合部２４４および積算部２４３に出力する。
【０１１３】
ステップＳＴ２３において、照合部２４４では、類似度Ｓｉｍ’（ｉ）と、予め設定した第１の閾値ｔｈ１とを比較し、類似度Ｓｉｍ’（ｉ）が第１の閾値より小さい場合には、積算部２４３では、類似度Ｓｉｍ’（ｉ）を積算し、詳細には数式Ｓ＝Ｓ＋Ｓｉｍ’（ｉ）により積算し照合部２４４に出力する（ＳＴ２４）。
【０１１４】
ステップＳＴ２５において、照合部２４４では、積算値Ｓと予め設定した第２の閾値ｔｈ２とを比較し、積算値Ｓが第２の閾値ｔｈ２よりも小さい場合には、変数ｉと値Ｎ−１とが比較され（ＳＴ２６）、変数ｉがＮ−１と一致していない場合には、変数ｉに１加算し（ＳＴ２７）、ステップＳＴ２２の処理に戻る。ステップＳＴ２５において、変数ｉがＮ−１と一致した場合には、画像が不一致であるとする（ＳＴ２８）。
【０１１５】
一方、ステップＳＴ２３の比較処理において、照合部２４４では、類似度Ｓｉｍ’（ｉ）が第１の閾値以上の場合には、画像が一致していると判別し、また、ステップＳＴ２５の比較処理において、照合部２４４では、積算値Ｓが第２の閾値ｔｈ２以上の場合には、画像が一致しているとし（ＳＴ２９）、例えばセキュリティ分野における静脈パターン照合装置に、本実施形態に係る画像照合装置を適用した場合には、電子錠を解除するといった処理を動作処理部１６が行う。
【０１１６】
以上、説明したように、本実施形態では、類似度生成部２４２が、登録画像ＡＩＭと照合画像ＲＩＭとの２次元上の異なる複数の位置関係それぞれにおいて、登録画像ＡＩＭと照合画像ＲＩＭの画素毎に比較して生成した類似度Ｓｉｍを、当該画素の周囲の画像特性に応じて補正し類似度Ｓｉｍ’を生成する類似度補正部２４２１を設け、照合部２４４は、補正した類似度Ｓｉｍ’に基づいて照合を行うので、第１実施形態に比べて、より高精度の照合処理を行うことができる。
【０１１７】
また、角度ずれの大きい、つまり相関の小さい線分同士の交点がある場合であっても、補正処理により類似度を低くするので、その結果、相関の大きい画像同士から求めた相関と、小さい画像同士から求めた相関度との差異が大きくなり、より高精度の照合処理を行うことができる。
【０１１８】
なお、本発明は本実施の形態に限られるものではなく、任意好適な種々の改変が可能である。
例えば、本実施形態では、類似度生成部は、数式（２７）により類似度を算出したが、この形態に限られるものではない。例えば類似度生成部は、線形状のパターンの相関に適した類似度を算出する処理を行えればよい。
【０１１９】
本実施形態では、フーリエ変換処理、対数変換処理、および対数−局座標変換を行い、フーリエ・メリン空間上での平行移動量を算出することにより、倍率情報および回転角度情報を生成したが、この形態に限られるものではない。例えば、倍率情報および回転角度情報が検出可能な空間に、座標変換を行ってもよい。
【０１２０】
また、第１の閾値ｔｈ１と、第２の閾値ｔｈ２を固定値にしたが、この形態に限られるものではない。例えば閾値それぞれを、画像パターンにより可変にすることで、より高精度の照合を行うことができる。
【０１２１】
【発明の効果】
本発明によれば、画像間の照合を高精度に行うことができる画像照合装置、画像照合方法、およびプログラムを提供することができる。
【図面の簡単な説明】
【図１】図１は、本発明に係る画像照合装置の第１実施形態の機能ブロック図である。
【図２】図２は、図１に示した画像照合装置のソフトウェア的な機能ブロック図である。
【図３】図３は、図２に示したフーリエ・メリン変換部の動作を説明するための図である。
【図４】図４は、自己相関法と位相限定相関法の相違点を説明するための図である。
【図５】図５は、位相限定相関法において、２つの画像間で平行移動ずれがある場合の相関強度分布を説明するための図である。
【図６】図６は、位相限定法において、２つの画像間で回転ずれがある場合の相関強度分布を説明するための図である。
【図７】図７は、位相限定相関部２３が出力する相関画像データを説明するための図である。（ａ）は登録画像ＡＩＭ、（ｂ）は照合画像ＲＩＭ、（ｃ）は相関画像データの一具体例を示す図である。
【図８】図８は図７（ｃ）に示した相関画像データを説明するための図である。
【図９】図９は、図１に示した候補特定部２４１の候補特定処理を説明するための図である。
【図１０】図１０は、図１に示した画像照合装置１の動作を説明するためのフローチャートである。
【図１１】図１１は、本発明に係る画像照合装置の第２実施形態のソフトウェア的な機能ブロック図である。
【図１２】図１２は、図１に示した類似度補正部２４２１の機能を説明するための図である。
【図１３】図１３は、図１に示した類似度補正部の機能を説明するための図である。
【図１４】図１４は、図１１に示した画像照合装置１ａの動作を説明するためのフローチャートである。
【符号の説明】
１，１ａ…画像照合装置、１１…画像入力部、１２…メモリ、１３…ＦＦＴ処理部、１４…座標変換部、１５…ＣＰＵ、１６…動作処理部、２１…倍率情報−回転情報部、２２…補正部、２３…平行移動部、２４…判別部、２１１…フーリエ・メリン変換部、２１２…位相限定相関部、２１３…倍率情報−回転情報生成部、２３２…合成部、２３３…位相抽出部、２３４…逆フーリエ変換部、２４１…候補特定部、２４２…類似度生成部、２４３…積算部、２４４…照合部、２１２０，２１２１…フーリエ変換部、２１２２…合成部、２１２３…位相抽出部、２１２４…逆フーリエ変換部、２３１１，２３１２…フーリエ変換部、２４２１…類似度補正部、２１１１１，２１１１２…フーリエ変換部、２１１２１，２１１２２…対数変換部、２１１３１，２１１３２…対数−極座標変換部。[0001]
BACKGROUND OF THE INVENTION
  The present invention relates to an image collation apparatus that performs collation based on image information such as a fingerprint image, a still image, and a moving image.andImage verification methodTo the lawIt is related.
[0002]
[Prior art]
Conventionally, various pattern matching devices are known as methods for matching based on image information. For example, as a pattern matching device, a registered image and a matching image to be compared are compared only in a predetermined positional relationship, a correlation value is calculated, and the registered image and the matching image are matched based on the correlation value.
[0003]
[Problems to be solved by the invention]
However, in the conventional pattern matching device, depending on the type of image pattern, the correlation value may not be high even for almost the same image, and even if only the peak is used, matching may not be performed with high accuracy. Improvement is desired.
[0004]
  The present invention has been made in view of such circumstances, and an object of the present invention is to provide an image collation apparatus capable of performing collation between images with high accuracy.andImage verification methodThe lawIt is to provide.
[0005]
[Means for Solving the Problems]
  In order to achieve the above object, a first aspect of the present invention is to provide the first image and the first image in each of a plurality of different two-dimensional positional relationships between the first image and the second image. A Fourier transform unit for Fourier transforming the two images, a coordinate transform unit for obtaining a phase from polar coordinates by performing a Fourier transform after Fourier transforming the first image and the second image, and the first obtained by the coordinate transform unit. A first phase-only correlation unit that synthesizes phases corresponding to the first and second images and inverse Fourier transforms the synthesized phase to obtain a peak position in the correlation intensity image; and an image of the peak position in the correlation intensity image Detect the amount of deviation from the center,A correction unit that corrects the verification image, which is the second image, by enlarging, reducing, and rotating the registered image, which is the first image, according to the shift amount, the first image, and the correction A second phase-only correlation unit that obtains a peak position by subjecting the corrected second image to Fourier transform and combining to extract a phase and performing inverse Fourier transform on the extracted phase; Candidate specification for specifying a plurality of candidates for the two-dimensional positional relationship between the first image and the corrected second image based on the plurality of peak positions output from the phase-only correlation unit And the first image and the correction in each of a plurality of different two-dimensional positional relationships between the first image specified by the candidate specifying means and the corrected second image. The second image is compared to generate similarity The first image and the corrected second image are generated based on the similarity generated by the similarity generation unit, the integration unit that integrates the similarity generated by the similarity generation unit, and the similarity acquired by the integration unit. Collating means for collating images withHave
[0006]
  According to the first aspect of the present invention, the image collating device further includes a logarithmic conversion unit that performs logarithmic conversion on the Fourier-transformed image before the coordinate conversion unit.
  Also,The image collating apparatus further includes a similarity correction unit, and the similarity correction unit includes linear patterns in different directions of the first image and the second image corrected by the similarity generation unit. The similarity is obtained from the intersection, the correlation of the intersection of the central portion is determined based on the presence or absence of an intersection other than the central portion in a predetermined number of pixels around the intersection, and the similarity is corrected using the correlation and the integration is performed. Output to the means.
  Further, the similarity correction unit calculates the number of intersecting intersection points obtained for the entire image composed of linear patterns in different directions of the first image and the corrected second image of the entire image. A value divided by the total number of intersections is used as an effective rate, and the effective rate is multiplied by the similarity generated by the similarity generation unit to perform correction processing, thereby generating a corrected similarity.
[0007]
  Furthermore, in order to achieve the above object, a second aspect of the present invention is to provide a first image and the second image in each of a plurality of different two-dimensional positional relationships between the first image and the second image. A first step of Fourier transforming the second image; a second step of Fourier transforming the first image and the second image and then obtaining a phase from the polar coordinates; and A third step of synthesizing phases corresponding to the first and second images and inverse Fourier transforming the synthesized phase to obtain a peak position in the correlation intensity image; and an image of the peak position in the correlation intensity image. Detect the amount of deviation from the center,A fourth step of correcting the verification image, which is the second image, by enlarging, reducing, and rotating the registered image, which is the first image, based on the shift amount; and the first image, The corrected second image is subjected to Fourier transform and then combined to extract a phase, and the extracted phase is subjected to inverse Fourier transform to obtain a peak position. Based on the plurality of peak positions , A sixth step of specifying a plurality of candidates for the two-dimensional positional relationship between the first image and the corrected second image, and the first image in which the plurality of candidates are specified The first image and the corrected second image are compared to generate a similarity in each of a plurality of different two-dimensional positional relationships between the image and the corrected second image. A seventh step and an eighth step of integrating the generated similarities; Based on the accumulated the degree of similarity, a ninth step of collating the second image which is the corrected and said first imageHave
[0009]
DETAILED DESCRIPTION OF THE INVENTION
The image collation apparatus according to the first embodiment of the present invention performs a rotation angle correction process or an enlargement ratio correction process between a registered image and a collated image, and performs registration and image registration in each of a plurality of different two-dimensional positional relationships. Similarity is generated by comparing with the verification image, the similarity is integrated, and verification is performed based on the integrated similarity.
Hereinafter, description will be given with reference to the drawings.
[0010]
FIG. 1 is a functional block diagram of a first embodiment of an image collating apparatus according to the present invention. For example, as illustrated in FIG. 1, the image matching device 1 according to the present embodiment includes an image input unit 11, a memory 12, an FFT processing unit 13, a coordinate conversion unit 14, a CPU 15, and an operation processing unit 16.
The image input unit 11, the memory 12, the FFT processing unit 13, the coordinate conversion unit 14, and the CPU 15 are connected by a bus BS. The operation processing unit 16 is connected to the CPU 15.
[0011]
The image input unit 11 is an input unit for inputting an image from the outside. For example, the registered image AIM and an image to be compared with the registered image AIM (referred to as a collation image RIM) are input to the image input unit 11.
The memory 12 stores an image input from the image input unit. For example, the memory 12 stores a registered image AIM, a collation image RIM, and a program P.
The program P includes, for example, procedures related to correlation processing, correction processing, similarity generation processing, integration processing, collation processing, and the like according to the present invention that are executed by the CPU 15.
The FFT processing unit 13 performs a two-dimensional Fourier transform process based on the image stored in the memory 12 under the control of the CPU 15, for example, and outputs the processing result to the coordinate conversion unit 14 and the CPU 15.
[0012]
The coordinate conversion unit 14 converts, for example, logarithm-polar coordinates based on the result of the two-dimensional Fourier transform process processed by the FFT processing unit 13 under the control of the CPU 15, and outputs the coordinate conversion result to the CPU 15.
The operation processing unit 16 performs a predetermined process such as releasing an electronic key when two images match, for example, based on a result of a collation process of the CPU 15 described later.
[0013]
CPU15 performs the collation process which concerns on embodiment of this invention, for example based on the program P memorize | stored in the memory 12, the registration image AIM, and the collation image RIM.
Further, the CPU 15 controls the image input unit 11, the memory 12, the FFT processing unit 13, the coordinate conversion unit 14, the operation processing unit 16, and the like, and executes processing according to the present embodiment.
[0014]
FIG. 2 is a software functional block diagram of the image collating apparatus shown in FIG. The CPU 15 controls the FFT processing unit 13, the coordinate conversion unit 14, and the like, and performs processing based on, for example, functional blocks as shown in FIG.
As shown in FIG. 2, the CPU 15 includes a magnification information / rotation information unit 21, a correction unit 22, a phase only correlation unit 23, and a determination unit 24.
The magnification information-rotation information unit 21, the correction unit 22, and the phase only correlation unit 23 correspond to a correlation detection unit according to the present invention.
[0015]
The magnification information-rotation information unit 21 and the correction unit 22 perform a rotation angle correction process or an enlargement ratio correction process on the registered image AIM and the collation image RIM.
The magnification information-rotation information unit 21 generates magnification information and / or rotation information based on the registered image AIM and the collation image RIM and outputs the magnification information and / or rotation information to the correction unit 22.
The magnification information includes information indicating the enlargement / reduction ratio of the registered image AIM and the collation image RIM. The rotation information includes information indicating the rotation angle of the registered image AIM and the collation image RIM.
[0016]
Specifically, for example, the magnification information-rotation information unit 21 includes a Fourier-Melin transform unit 211, a phase only correlation unit 212, and a magnification information-rotation information generation unit 213.
The Fourier-Merlin transform unit 211 performs Fourier-Merlin transform, which will be described later, based on the respective image information, and outputs signals SA211 and SR211 indicating the respective transformation results to the phase-only correlation unit 212.
[0017]
Specifically, the Fourier-Melin transform unit 211 includes Fourier transform units 2111, 21112, logarithmic transform units 21121, 2112, and logarithmic-polar coordinate transform units 2111, 211132.
[0018]
For example, when the registered image AIM is an N × N image and the registered image AIM is f1 (m, n), the Fourier transform unit 21111 performs Fourier transform as shown in Equation (1), and the Fourier image Data F <b> 1 (m, n) is generated and output to the logarithmic conversion unit 21121. For example, when the collation image RIM is an N × N image and the collation image RIM is f2 (m, n), the Fourier transform unit 21112 performs Fourier transform as shown in Equation (2), and the Fourier image Data F2 (m, n) is generated and output to the logarithmic conversion unit 21122.
[0019]
The Fourier image data F1 (m, n) is composed of an amplitude spectrum A (u, v) and a phase spectrum Θ (u, v) as shown in Equation (1), and the Fourier image data F2 (m, n) is As shown in Equation (2), it is composed of an amplitude spectrum B (u, v) and a phase spectrum Φ (u, v).
[0020]
[Expression 1]

[0021]
[Expression 2]

[0022]
Logarithmic conversion units 21121, 21122 perform logarithmic processing based on the amplitude components of the Fourier image data F1 (m, n) and F2 (m, n) generated by the Fourier transformation units 2111, 21112. This logarithmic processing of amplitude components emphasizes high frequency components including detailed feature information of image data.
[0023]
Specifically, the logarithmic conversion unit 21121 performs logarithmic processing based on the amplitude component A (u, v) as shown in Equation (3) to generate A ′ (u, v) and outputs it to the logarithmic-polar coordinate conversion unit 21131. To do. The logarithmic conversion unit 21122 performs logarithmic processing based on the amplitude component B (u, v) as shown in Equation (4) to generate B ′ (u, v) and outputs it to the logarithmic-polar coordinate conversion unit 21132.
[0024]
[Equation 3]

[0025]
[Expression 4]

[0026]
The logarithm-polar coordinate converters 21131, 21132 convert the logarithm-polar coordinate system (for example, log-r, Θ) based on the signals output from the logarithmic converters 21121, 2112.
In general, for example, if the point (x, y) is defined as shown in equations (5) and (6), r = e^μIn this case, μ = log (r), and there is a unique (log (r), Θ) corresponding to an arbitrary point (x, y). Due to this property, the logarithm-polar coordinate conversion units 21131, 21132 perform coordinate conversion.
[0027]
[Equation 5]

[0028]
[Formula 6]

[0029]
Specifically, the logarithm-polar coordinate converters 21131, 21132 are set to the set (r_i, Θ_j), And the function f (r shown in Equation (8)_i, Θ_j) Is defined.
[0030]
[Expression 7]

[0031]
[Equation 8]

[0032]
The logarithm-polar coordinate converters 21131, 21132 are the sets defined by the mathematical formulas (7), (8) (r_i, Θ_j), Function f (r_i, Θ_j), Log-polar coordinate conversion is performed on each of the image data A ′ (u, v) and B ′ (u, v) as shown in equations (9) and (10), and pA (r_i, Θ_j), PB (r_i, Θ_j) And output to the phase only correlation unit 212 as the signal SA211 and the signal SR211 respectively.
[0033]
[Equation 9]

[0034]
[Expression 10]

[0035]
FIG. 3 is a diagram for explaining the operation of the Fourier-Melin transform unit 211 shown in FIG.
The image f1 (m, n) and the image f2 (m, n) include rectangular regions W1 and W2 having different predetermined angles with respect to the x and y axes, for example.
In the Fourier-Melin transform unit 211, for example, as shown in FIG. 3, the image f1 (m, n) is Fourier transformed by the Fourier transform unit 21111 to generate Fourier image data A ′ (u, v), and the logarithm Image data pA (r, θ) is generated by the conversion unit 21121 and the logarithmic-polar coordinate conversion unit 21131.
[0036]
Similarly, the image f2 (m, n) is Fourier-transformed by the Fourier transform unit 21112 to generate Fourier image data B ′ (u, v). The logarithmic transform unit 21122 and the log-polar coordinate transform unit 21132 Data pB (r, θ) is generated.
[0037]
As described above, the images f1 (m, n) and f2 (m, n) are transformed from Cartesian coordinates onto a logarithmic-polar coordinate system (also referred to as Fourier-Melin space) by Fourier transformation and logarithmic-polar coordinate transformation. .
In the Fourier-Melin space, the component moves along the log-r axis according to the scaling of the image, and moves along the θ axis according to the rotation angle of the image.
Using this property, the scaling (magnification information) and the rotation angle of the images f1 (m, n) and f2 (m, n) are converted into the amount of movement along the log-r axis in the Fourier-Merlin space, and θ It can be determined based on the amount of movement along the axis.
[0038]
The phase-only correlation unit 212 is based on the signals SA211 and SR211 indicating the pattern data output from the Fourier-Melin transform unit 211 by a phase-only correlation method using, for example, a phase-only filter (SPOMF: Symmetric phase-only matched filter). Then, the respective parallel movement amounts are obtained.
For example, as illustrated in FIG. 1, the phase only correlation unit 212 includes Fourier transform units 2120 and 2121, a synthesis unit 2122, a phase extraction unit 2123, and an inverse Fourier transform unit 2124.
[0039]
The Fourier transform units 2120 and 2121 are based on the signals SA211 (pA (m, n)) and SR211 (pB (m, n)) output from the logarithmic-polar coordinate conversion units 21131 and 21132. ) To perform a Fourier transform. Here, X (u, v) and Y (u, v) are Fourier coefficients. The Fourier coefficient X (u, v) is composed of an amplitude spectrum C (u, v) and a phase spectrum θ (u, v) as shown in Equation (11). The Fourier coefficient Y (u, v) is composed of an amplitude spectrum D (u, v) and a phase spectrum φ (u, v) as shown in Equation (12).
[0040]
## EQU11 ##

[0041]
[Expression 12]

[0042]
The combining unit 2122 combines X (u, v) and Y (u, v) generated by the Fourier transform units 2120 and 2121 to obtain a correlation. For example, the synthesizing unit 2122 uses X (u, v) · Y^*(U, v) is generated and output to the phase extraction unit 2123. Where Y^*(U, v) is a complex conjugate of Y (u, v).
[0043]
The phase extraction unit 2123 extracts phase information by removing the amplitude component based on the combined signal output from the combining unit 2122.
For example, the phase extraction unit 2123 can perform, for example, X (u, v) Y^*Based on (u, v), its phase component Z (u, v) = e^{j (θ (u, v) −φ (u, v))}To extract.
[0044]
The extraction of the phase information is not limited to the above-described form. For example, after extracting phase information based on the outputs of the Fourier transform units 2120 and 2121 and the equations (13) and (14), only the phase component is synthesized as shown in the equation (15), and Z (u, v) May be generated.
[0045]
[Formula 13]

[0046]
[Expression 14]

[0047]
[Expression 15]

[0048]
The inverse Fourier transform unit 2124 performs an inverse Fourier transform process based on the signal Z (u, v) having only phase information output from the phase extraction unit 2123 to generate a correlation strength image.
Specifically, the inverse Fourier transform unit 2124 performs an inverse Fourier transform process based on the signal Z (u, v) as shown in Equation (16), and generates a correlation strength image G (p, q).
[0049]
[Expression 16]

[0050]
The magnification information-rotation information generation unit 213 determines the shift amount from the image center of the peak position in the correlation strength image G (p, q) generated by the inverse Fourier transform unit 2124, that is, the registered image AIM and the collation image RIM. Since this is equivalent to the amount of translation between the pattern data obtained as a result of the Fourier-Melin transformation, the magnification information (enlargement / reduction ratio) of the matching image RIM with respect to the registered image AIM is detected by detecting this deviation amount. ) And correction information S21 including data indicating the rotation angle information is generated.
[0051]
The correcting unit 22 corrects the collation image RIM based on the correction information S21 output from the magnification information-rotation information-rotation information unit 21 magnification information-rotation information generation unit 213. Specifically, the correction unit 22 performs an enlargement / reduction process on the collation image RIM based on the magnification information and the rotation angle information included in the correction information S21, performs the rotation process, and outputs the result to the phase-only correlation unit 23. By the correction process of the correction unit 22, the difference in scaling and rotation components between the registered image AIM and the collation image RIM is removed.
For this reason, only the translation component remains as a difference between the registered image AIM and the corrected collation image RIM.
[0052]
The phase only correlation unit 23 detects a translation component between the above-described registered image AIM and the corrected collation image RIM, and a correlation value thereof. This detection is obtained, for example, by the phase only correlation method using the above-described phase only filter.
Specifically, the phase only correlation unit 23 includes Fourier transform units 2311 and 2312, a synthesis unit 232, a phase extraction unit 233, and an inverse Fourier transform unit 234.
[0053]
The Fourier transform units 2311 and 2312, the synthesis unit 232, the phase extraction unit 233, and the inverse Fourier transform unit 234 are respectively the Fourier transform units 2120 and 2121, the synthesis unit 2122, the phase extraction unit 2123, and the phase-only correlation unit 212 described above. Since it has the same function as each of the inverse Fourier transform units 2124, it will be briefly described.
[0054]
The Fourier transform unit 2311 performs Fourier transform on the registered image AIM and outputs the result to the synthesis unit 232. At this time, the registered image AIM subjected to the Fourier transform process by the Fourier transform unit 21111 may be stored in the memory 12 and output to the combining unit 232. By doing so, the Fourier transform processing is not performed twice, so that the processing is reduced.
The Fourier transform unit 2312 performs Fourier transform on the image S22 that has been corrected by the correction unit 22 and outputs a Fourier image as a processing result to the synthesis unit 232.
[0055]
The synthesizing unit 232 synthesizes the Fourier images S2311, S2312 output from the Fourier transform units 2311, 2312 and outputs the synthesized image S232 to the phase extraction unit 233.
The phase extraction unit 233 extracts the phase information as described above based on the composite image S232 and outputs the signal S233 to the inverse Fourier transform unit 234.
The inverse Fourier transform unit 234 performs inverse Fourier transform based on the signal S233, generates a correlation strength image (correlation image data), and outputs the correlation strength image to the determination unit 24 as a signal S23.
[0056]
A detailed description of the parallel displacement detection process in the above-described phase-only correlation method will be given.
For example, the original image f1 (m, n), the original image f2 (m, n), and the image f3 (m, n) = f2 (m + α, n + β) obtained by translating the image f2 (m, n) are Fourier transformed. Processing is performed to generate Fourier coefficients F1 (u, v), F2 (u, v), and F3 (u, v) as shown in equations (17) to (19).
[0057]
[Expression 17]

[0058]
[Formula 18]

[0059]
[Equation 19]

[0060]
Based on the Fourier coefficients F1 (u, v) to F3 (u, v), as shown in the equations (20) to (22), the phase images F′1 (u, v) to F′3 having only the phase information. (U, v) is generated.
[0061]
[Expression 20]

[0062]
[Expression 21]

[0063]
[Expression 22]

[0064]
The correlation Z12 (u, v) of the phase image of the correlation between the phase image F′1 (u, v) and the phase image F′2 (u, v) is calculated as shown in Equation (23), and the phase image F ′ The correlation Z13 (u, v) of the phase image of the correlation between 1 (u, v) and the phase image F′3 (u, v) is calculated as shown in equation (24).
[0065]
[Expression 23]

[0066]
[Expression 24]

[0067]
The correlation strength image G12 (r, s) of the correlation Z12 (u, v) and the correlation strength image G13 (r, s) of the correlation Z13 (u, v) are calculated as shown in equations (25) and (26). To do.
[0068]
[Expression 25]

[0069]
[Equation 26]

[0070]
As shown in equations (25) and (26), when the image f3 (m, n) is shifted by (+ α, + β) compared to the image 2 (m, n), the phase-only correlation method is used. In the strong correlation image, a correlation strength peak is generated at a position shifted by (−α, −β). Based on the position of the correlation strength position, the amount of parallel movement between the two images can be obtained.
In addition, by using this phase-only correlation method on the above-described Fourier-Melin space, the amount of parallel movement on the Fourier-Melin space can be detected. This parallel movement amount corresponds to magnification information and rotation angle information in the real space as described above.
[0071]
FIG. 4 is a diagram for explaining the difference between the autocorrelation method and the phase only correlation method.
In the autocorrelation method, for example, as shown in FIGS. 4A and 4B, when the image IM1 and the same image IM2 as the image IM1 are subjected to Fourier transform to generate the autocorrelation function SG1, the result shown in FIG. As described above, a correlation strength distribution having a peak having a high correlation strength and a small correlation strength at the periphery thereof is obtained. In FIG.4 (c), a vertical axis | shaft shows correlation strength.
[0072]
On the other hand, in the above-described phase-only correlation method, as shown in FIGS. 4D and 4E, when the image IM1 and the image IM2 that is the same as the image IM1 are subjected to Fourier transform and only the phase information is correlated, FIG. ), A correlation strength distribution having a high correlation strength and only a sharp peak is obtained. Thus, the phase-only method can obtain clear information regarding the correlation as compared with the autocorrelation method. In FIG. 4F, the vertical axis (z-axis) indicates the correlation strength, and the x-axis and the y-axis indicate the shift amount.
[0073]
FIG. 5 is a diagram for explaining the correlation intensity distribution when there is a parallel displacement between two images in the phase-only correlation method.
For example, the correlation intensity distribution in the phase limiting method of the image IM1 as shown in FIGS. 5A and 5B and the image IM3 moved several pixels parallel to the image IM1 is shown in FIG. 5C, for example. Thus, sharp peaks with high correlation strength are distributed at positions shifted from the peak positions in the correlation image data shown in FIG. 4 (f) by a distance corresponding to the amount of parallel movement. However, the peak intensity shown in FIG. 5C is smaller than the peak intensity shown in FIG. This is because the matching pixel regions of the images IM1 and IM3 are smaller than those of the images IM1 and IM2.
[0074]
FIG. 6 is a diagram for explaining the correlation strength distribution when there is a rotational shift between two images in the phase only method.
The correlation intensity distribution in the phase limiting method of the image IM1 as shown in FIGS. 6A and 6B and the image 4 rotated several degrees from the image IM1 is, for example, as shown in FIG. A correlation strength distribution is obtained. If the phase-only correlation method is simply used, it is difficult to detect the correlation due to rotational deviation.
[0075]
For this reason, in the image collation device 1 according to the present embodiment, the registered image AIM and the collation image RIM are subjected to Fourier-Melin transformation, and the phase-only correlation method is used in the Fourier-Melin space to obtain the translation amount. And detecting magnification information and rotation angle information of the registered image AIM and the collation image RIM. Based on the information, the magnification and rotation angle of the verification image are corrected.
A translation shift between the corrected collation image RIM and the registered image AIM is detected by the phase only correlation method, and at the same time, collation between the images is performed based on the correlation peak.
[0076]
FIG. 7 is a diagram for explaining the correlation image data output by the phase only correlation unit 23. FIG. 7A is a registered image AIM, FIG. 7B is a collation image RIM, and FIG. 7C is a diagram showing a specific example of correlation image data. FIG. 8 is a diagram for explaining the correlation image data shown in FIG.
[0077]
When the registered image AIM and the collation image RIM include, for example, binarized and linear patterns such as handwritten characters and blood vessel patterns, for example, the registered image AIM and the collation image RIM as shown in FIGS. In this case, the phase only correlation unit 23 outputs correlation image data as shown in FIG.
[0078]
In FIG. 7C, the z-axis direction indicates the correlation peak intensity in the correlation image data, and the correlation peak intensity corresponds to the correlation degree between the registered image AIM and the collation image RIM. As shown in FIGS. 7C and 8, the correlation peak positions in the x-axis direction and the y-axis direction are, for example, the amount of deviation from the image center corresponds to the parallel movement amount between the registered image AIM and the collation image RIM. .
[0079]
  The discriminating unit 24 collates the registered image AIM and the collation image RIM based on the signal S23 output from the phase only correlation unit 23.
  DiscriminatorFor example, FIG.2As shown in FIG. 5, the candidate specifying unit 241, the similarity generation unit 242, the integrating unit 243, and the collating unit 244 are included.
  The candidate identification unit 241 corresponds to the candidate identification unit according to the present invention, the similarity generation unit 242 corresponds to the similarity generation unit according to the present invention, the integration unit 243 corresponds to the integration unit according to the present invention, and the collation unit Reference numeral 244 corresponds to collation means according to the present invention.
[0080]
The candidate specifying unit 241 specifies a plurality of candidates for the two-dimensional positional relationship between the registered image AIM, the matching image RIM, and the similarity generation unit as the signal S241 based on the signal S23 output from the phase only correlation unit 23. It outputs to 242.
[0081]
FIG. 9 is a diagram for explaining candidate specifying processing of the candidate specifying unit 241 shown in FIG. The numerical value indicates the correlation peak intensity of the correlation image data on the XY plane of the correlation image data.
For example, in the case of a registered image AIM and a collation image RIM including a binarized and linear pattern as shown in FIGS. 7A and 7B, correlation peak intensities (also referred to as correlation intensities) between images having a large correlation. However, the value is small as shown in FIG.
For example, as illustrated in FIG. 9, the candidate specifying unit 241 selects the top N correlation strengths based on the signal S23, that is, eight in the present embodiment, the registered image AIM, the matching image RIM, and the two-dimensional positional relationship candidates. And output to the similarity generation unit 242 as a signal S241.
[0082]
The similarity generation unit 242 generates a two-dimensional image between the registered image AIM and the matching image RIM corrected by the correction unit 22 based on the signal S242 indicating a plurality of positional relationship candidates specified by the candidate specifying unit 242. In each of the plurality of different positional relationships above, for each pixel, the registered image AIM and the collation image RIM corrected by the correction unit 22 are compared to generate a similarity.
[0083]
Specifically, as shown in FIGS. 7A and 7B, for example, the similarity generation unit 242 sets the registered image AIM to f1 (m, n) and the matching image RIM corrected by the correction unit 22 to f2 ( Assuming that the image includes a binarization of m × n pixels of m, n) and a linear pattern, for example, the similarity Sim is calculated by Equation (27), and the calculation result is output to the accumulating unit 243 and the matching unit 244. .
[0084]
[Expression 27]

[0085]
The integration unit 243 integrates the similarity Sim generated by the similarity generation unit 242 and outputs the integration result to the collation unit 244 as a signal S243.
[0086]
The collation unit 244 collates the registered image AIM and the collation image RIM based on the similarity Sim accumulated by the accumulation unit 243. Specifically, for example, the collation unit 244 determines that the registered image AIM matches the collation image RIM when the integrated similarity Sim is greater than a predetermined value.
[0087]
The collation unit 244 collates the registered image AIM with the collation image RIM based on the similarity Sim output from the similarity generation unit 242. Specifically, for example, the collation unit 244 determines that the registered image AIM matches the collation image RIM when the similarity Sim is greater than a predetermined value.
Moreover, the collation part 244 outputs the signal S24 which shows the result of collation processing.
[0088]
FIG. 10 is a flowchart for explaining the operation of the image collating apparatus 1 shown in FIG. The operation of the image collating apparatus 1 having the above-described configuration will be briefly described with reference to FIG.
For example, the registered image AIM and the collation image RIM are input by the image input unit 11, and each image data is stored in the memory (ST1).
Here, in order to obtain the magnification (enlargement / reduction ratio) and rotation angle information of the collation image RIM with respect to the registration image AIM, the registration image AIM is read from the memory 12 (ST2), and the magnification information-rotation information unit 21 Fourier The transform unit 21111 performs Fourier transform processing (ST3), and Fourier image data S21111 is stored and stored in the memory 12 (ST4).
The amplitude component in the Fourier image data S21111 is subjected to logarithmic processing by the logarithmic conversion unit 21121 and converted to a logarithmic-polar coordinate system by the logarithmic-polar coordinate conversion unit 21131 (ST5).
[0089]
The collation image RIM is read from the memory 12 (ST6), similarly subjected to Fourier transform processing by the Fourier transform unit 21112 (ST7), and the amplitude component in the Fourier image data S21112 is subjected to logarithmic processing by the logarithmic transform unit 21122. Then, the logarithmic-polar coordinate conversion unit 21132 converts the logarithmic-polar coordinate system (ST8).
[0090]
The image signals (also referred to as pattern data) SA211 and SR211 obtained as a result of performing the Fourier-Merin transform on the registered image AIM and the collation image RIM described above are Fourier transformed by the Fourier transform units 2120 and 2121 of the phase-only correlation unit 212, respectively. It is transformed (ST9), synthesized by the synthesizer 2122 (ST10), the amplitude component is removed from the synthesized signal by the phase extractor 2123 (ST11), and the remaining phase component is inverse Fourier transformed by the inverse Fourier transformer 2124. (ST12) Based on the shift amount of the peak position of the obtained correlation image data from the image center, correction information including the magnification information and the rotation information is generated by the magnification information-rotation information generation unit 213 (ST13). .
[0091]
The correction unit 22 performs correction processing for enlargement / reduction of the collation image and rotation processing based on the correction information, and removes the scaling component and the rotation component between the images (ST14). The remaining difference is only the translation component and is detected using the phase only correlation method.
[0092]
The collated image RIM subjected to the correction process is Fourier transformed by the Fourier transform unit 2312 of the phase only correlation unit 23 (ST15) to generate Fourier image data S2312, and the Fourier transformed registered image AIM stored in the memory 12 Is read (ST16), and the combining unit 232 generates combined data S232 (ST17).
[0093]
At this time, the registered image AIM may be Fourier transformed by the Fourier transform unit 2311 to generate Fourier image data S2311 and input to the combining unit 232. The amplitude information in the synthesized data S232 is removed by the phase extraction unit 233 (ST18), the remaining phase information is input to the inverse Fourier transform unit 234, and the signal S23 is output as correlation image data (ST19).
[0094]
Based on the correlation image data S23 generated by the above-described processing, for example, N candidates P are detected by the candidate specifying unit 241 from the upper side of the correlation peak intensity in the correlation image data S23._i(P₀, P₁, P₂, ..., P_n-1) Are extracted and output to the similarity generation unit 242 as a signal S241 (ST20).
[0095]
Integration unit 243 initializes variables for integration (ST21). For example, the variable i is initialized to 0 and the integrated value S is initialized to 0.
In the similarity generation unit 242, each candidate (coordinate) P_i, By detecting the amount of deviation from the center of the correlation image data, the registered image AIM and the collation image RIM are aligned, and the similarity Sim (i) is calculated. It outputs to 244 (ST22).
[0096]
In step ST23, the collation unit 244 compares the similarity Sim (i) with a preset first threshold th1, and when the similarity Sim (i) is smaller than the first threshold, the integration unit 243 Then, the degree of similarity Sim (i) is accumulated, and in detail, it is accumulated by the formula S = S + Sim (i) and is output to the collation unit 244 (ST24).
In step ST25, the collation unit 244 compares the integrated value S with a preset second threshold th2, and if the integrated value S is smaller than the second threshold th2, the variable i and the value N-1 are set. Are compared (ST26), and if the variable i does not match N-1, 1 is added to the variable i (ST27), and the process returns to step ST22. In step ST25, if the variable i matches N-1, it is assumed that the images do not match (ST28).
[0097]
On the other hand, in the comparison process in step ST23, the collation unit 244 determines that the images match when the similarity Sim (i) is equal to or greater than the first threshold value. In the comparison process in step ST25, In the matching unit 244, when the integrated value S is equal to or greater than the second threshold th2, it is assumed that the images match (ST29). For example, the image matching device according to the present embodiment is applied to the vein pattern matching device in the security field. When applied, the operation processing unit 16 performs processing such as releasing the electronic lock.
[0098]
As described above, the candidate specifying unit 241 that specifies a plurality of candidates for the two-dimensional positional relationship between the registered image AIM and the matching image RIM, and the two-dimensional difference between the registered image AIM and the matching image RIM. In each of a plurality of positional relationships, a similarity generation unit 242 that generates a similarity Sim by comparing the registered image AIM and the matching image RIM, and an integration unit 243 that integrates the similarity Sim generated by the similarity generation unit 242 Since the collation unit 244 that collates based on the similarity Sim accumulated by the accumulation unit 243 is provided, for example, even if the correlation between two image data to be compared is small, By adding the similarities calculated for the respective positional relationships of the candidates, it is possible to perform collation with higher accuracy than in the case where collation is performed using the similarity alone. In addition, when the similarity Sim is larger than the first threshold th1, it is determined that they match, so that the matching process can be performed at high speed.
[0099]
Further, a magnification information-rotation information unit 21, a correction unit 22, and a correction unit 22 for detecting a correlation based on a phase component as a result of a rotation angle correction process or an enlargement ratio correction process between the registered image AIM and the matching image RIM, and a Fourier transform process; Since the phase only correlation unit 23 is provided, the matching process can be performed even when the registered image AIM and the matching image RIM have different rotation angles and enlargement ratios.
[0100]
FIG. 11 is a software functional block diagram of the second embodiment of the image collating apparatus according to the present invention.
The hardware functional blocks of the image collating apparatus 1a according to the present embodiment are substantially the same as those in the first embodiment, and thus description thereof is omitted.
As shown in FIG. 11, for example, the CPU 15a of the image collating apparatus 1a includes a magnification information-rotation information unit 21, a correction unit 22, a phase only correlation unit 23, and a determination unit 24a.
For example, as illustrated in FIG. 11, the determination unit 24 a includes a candidate specifying unit 241, a similarity generation unit 242, a similarity correction unit 2421, an integration unit 243, and a collation unit 244.
Description of functions similar to those of the first embodiment will be omitted, and only differences will be described.
[0101]
The similarity correction unit 2421 is generated by the similarity generation unit 242 by comparing each pixel of the registered image AIM and the matching image RIM in each of a plurality of different two-dimensional positional relationships between the registered image AIM and the matching image RIM. The similarity Sim is corrected according to the image characteristics around the pixel to generate a similarity Sim ′, which is output to the integrating unit 243 and the collating unit 244 as a signal S2421.
The accumulating unit 243 and the collating unit 244 perform the same processing as in the first embodiment based on the similarity Sim ′.
[0102]
FIG. 12 is a diagram for explaining the function of the similarity correction unit 2421 shown in FIG. 12A illustrates an image pattern of the registered image AIM, FIG. 12B illustrates an image pattern of the matching image RIM, and FIG. 12C illustrates a specific example of the similarity Sim generated by the similarity generation unit 242. FIG. FIG. 13 is a diagram for explaining the function of the similarity correction unit 2421 shown in FIG.
[0103]
For example, the similarity Sim generated by the similarity generation unit 242 generates, for example, the similarity Sim between the image pattern of the registered image AIM illustrated in FIG. 12A and the image pattern of the matching image RIM illustrated in FIG. As shown in FIG. 12C, the calculation is based on the intersection CP of the linear patterns of both images.
[0104]
For example, as illustrated in FIG. 13A, the similarity generation unit 242 correlates the linear pattern with different directions in the case of the correlation pattern PT31 between the pattern PT11 of the registered image AIM and the pattern PT21 of the matching image RIM. In spite of the absence, the similarity is increased because the similarity Sim is calculated only from the intersection.
The similarity correction unit 2421 according to the present embodiment detects, for example, the intersection point CP. For example, in the nine pixels around the pixel, if there is an intersection point CP other than the center portion, the intersection point CP of the center portion is correlated. If there is no intersection CP other than the center, the center CP is determined as an uncorrelated intersection.
[0105]
For example, in the case of the correlation pattern PT31 between the pattern PT11 of the registered image AIM and the pattern PT21 of the collation image RIM as shown in FIG. The intersection at the center is determined as an intersection with no correlation.
In the case of the correlation pattern PT32 between the pattern PT12 of the registered image AIM and the pattern PT22 of the collation image RIM as shown in FIG. The intersection of the parts is determined as an uncorrelated intersection.
[0106]
In the case of the correlation pattern PT32 between the pattern PT12 of the registered image AIM and the pattern PT22 of the collation image RIM as shown in FIG. The intersection CP of the part is determined as a correlated intersection.
[0107]
The similarity correction unit 2421 uses the value obtained by dividing the number of intersecting intersections obtained for the entire image by the total number of intersections as the effective rate eff, and the similarity generated by the similarity generation unit 242 as shown in Equation (28). The degree Sim is multiplied to perform correction processing to generate a corrected similarity Sim ′ and output it as a signal S2421.
[0108]
[Expression 28]

[0109]
For example, in the case shown in FIG. 12, the effectiveness rate eff is 0, and as a result, the corrected similarity Sim ′ is 0.
[0110]
FIG. 14 is a flowchart for explaining the operation of the image collating apparatus 1a shown in FIG. Only the differences from the first embodiment will be described with reference to FIG. 14 for the operation of the image collating apparatus 1 having the configuration described above. The processing from step ST1 to step ST20 is the same as that in the first embodiment, and a description thereof will be omitted.
[0111]
Integration unit 243 initializes variables for integration (ST21). For example, the variable i is initialized to 0 and the integrated value S is initialized to 0.
In the similarity generation unit 242, each candidate (coordinate) P_i, The degree of similarity Sim (i) is calculated after the registration image AIM and the matching image RIM are aligned by detecting the amount of deviation from the center of the correlation image data, and the similarity correction unit 2421 is used as the signal S242. (ST22).
[0112]
In step ST221, the similarity correction unit 2421 divides the number of correlated intersections obtained for the entire image based on the correction signal S242 as the similarity Sim output from the similarity generation unit 242 by the total number of intersections. The value is an effective rate eff, and as shown in Expression (28), the similarity Sim generated by the similarity generation unit 242 is multiplied to perform correction processing to generate a corrected similarity Sim ′, and a signal S2421 The data is output to the collation unit 244 and the integration unit 243.
[0113]
In step ST23, the collation unit 244 compares the similarity Sim ′ (i) with a preset first threshold th1, and if the similarity Sim ′ (i) is smaller than the first threshold, integration is performed. The unit 243 accumulates the similarity Sim ′ (i), specifically, accumulates by the formula S = S + Sim ′ (i), and outputs it to the collation unit 244 (ST24).
[0114]
In step ST25, the collation unit 244 compares the integrated value S with a preset second threshold th2, and if the integrated value S is smaller than the second threshold th2, the variable i and the value N-1 are set. Are compared (ST26), and if the variable i does not match N-1, 1 is added to the variable i (ST27), and the process returns to step ST22. In step ST25, if the variable i matches N-1, it is assumed that the images do not match (ST28).
[0115]
On the other hand, in the comparison process of step ST23, the collation unit 244 determines that the images match when the similarity Sim ′ (i) is equal to or greater than the first threshold, and also in the comparison process of step ST25. The collation unit 244 assumes that the images match if the integrated value S is equal to or greater than the second threshold th2 (ST29). For example, the image collation device according to the present embodiment is applied to the vein pattern collation device in the security field. Is applied, the operation processing unit 16 performs processing such as releasing the electronic lock.
[0116]
As described above, in the present embodiment, the similarity generation unit 242 performs pixel-by-pixel registration of the registered image AIM and the matching image RIM in each of a plurality of different two-dimensional positional relationships between the registered image AIM and the matching image RIM. A similarity correction unit 2421 is provided that corrects the similarity Sim generated in comparison with the image characteristics according to the image characteristics around the pixel to generate similarity Sim ′, and the collation unit 244 adds the similarity Sim ′ to the corrected similarity Sim ′. Since collation is performed based on this, it is possible to perform collation processing with higher accuracy than in the first embodiment.
[0117]
Even if there is an intersection between line segments with a large angle deviation, that is, a small correlation, the similarity is lowered by the correction process. As a result, a correlation obtained from images with a large correlation and a small image are obtained. The difference with the correlation degree calculated | required from each other becomes large, and more highly accurate collation processing can be performed.
[0118]
Note that the present invention is not limited to the present embodiment, and various suitable modifications can be made.
For example, in the present embodiment, the similarity generation unit calculates the similarity using Equation (27), but the present invention is not limited to this form. For example, the similarity generation unit only needs to be able to perform a process of calculating a similarity suitable for the correlation of the linear pattern.
[0119]
In the present embodiment, Fourier transform processing, logarithmic transformation processing, and logarithm-station coordinate transformation are performed, and the magnification information and the rotation angle information are generated by calculating the parallel movement amount in the Fourier-Merlin space. It is not limited to form. For example, coordinate conversion may be performed in a space where magnification information and rotation angle information can be detected.
[0120]
In addition, the first threshold th1 and the second threshold th2 are fixed values, but the present invention is not limited to this form. For example, by making each of the threshold values variable depending on the image pattern, it is possible to perform more accurate collation.
[0121]
【The invention's effect】
ADVANTAGE OF THE INVENTION According to this invention, the image collation apparatus, the image collation method, and program which can perform collation between images with high precision can be provided.
[Brief description of the drawings]
FIG. 1 is a functional block diagram of a first embodiment of an image collating apparatus according to the present invention.
FIG. 2 is a software functional block diagram of the image collating apparatus shown in FIG. 1;
FIG. 3 is a diagram for explaining the operation of the Fourier-Melin transform unit shown in FIG. 2;
FIG. 4 is a diagram for explaining a difference between an autocorrelation method and a phase-only correlation method;
FIG. 5 is a diagram for explaining a correlation intensity distribution when there is a parallel shift between two images in the phase-only correlation method;
FIG. 6 is a diagram for explaining a correlation intensity distribution when there is a rotational shift between two images in the phase limiting method.
FIG. 7 is a diagram for explaining correlation image data output by a phase-only correlation unit 23; (A) is a registered image AIM, (b) is a collation image RIM, and (c) is a diagram showing a specific example of correlation image data.
FIG. 8 is a diagram for explaining the correlation image data shown in FIG.
FIG. 9 is a diagram for explaining candidate specifying processing of the candidate specifying unit 241 illustrated in FIG. 1;
FIG. 10 is a flowchart for explaining the operation of the image collating apparatus 1 shown in FIG. 1;
FIG. 11 is a software functional block diagram of the second embodiment of the image collating apparatus according to the present invention.
FIG. 12 is a diagram for explaining a function of the similarity correction unit 2421 shown in FIG. 1;
FIG. 13 is a diagram for explaining the function of the similarity correction unit shown in FIG. 1;
FIG. 14 is a flowchart for explaining the operation of the image collating apparatus 1a shown in FIG.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1,1a ... Image collation apparatus, 11 ... Image input part, 12 ... Memory, 13 ... FFT processing part, 14 ... Coordinate conversion part, 15 ... CPU, 16 ... Operation processing part, 21 ... Magnification information-rotation information part, 22 ... correction unit, 23 ... parallel movement unit, 24 ... discriminating unit, 211 ... Fourier-Merin transform unit, 212 ... phase-only correlation unit, 213 ... magnification information-rotation information generation unit, 232 ... combination unit, 233 ... phase extraction unit 234 ... inverse Fourier transform unit, 241 ... candidate identification unit, 242 ... similarity generation unit, 243 ... integration unit, 244 ... collation unit, 2120, 2121 ... Fourier transform unit, 2122 ... synthesis unit, 2123 ... phase extraction unit, 2124 ... Inverse Fourier transform unit, 2311, 2112 ... Fourier transform unit, 2421 ... Similarity correction unit, 21111, 11112 ... Fourier transform unit, 21121, 2122 ... Logarithmic transform unit, 1131,21132 ... the log - polar coordinate conversion unit.

Claims (5)

A Fourier transform unit that Fourier transforms the first image and the second image in each of a plurality of different two-dimensional positional relationships between the first image and the second image;
A coordinate conversion unit that performs a Fourier transform on the first image and the second image and then converts the polar coordinate to obtain a phase from the polar coordinate;
A first phase-only correlation unit that synthesizes phases corresponding to the first and second images obtained by the coordinate transformation unit, and inverse Fourier transforms the synthesized phase to obtain a peak position in the correlation intensity image;
The amount of deviation of the peak position in the correlation intensity image from the image center is detected, and the collation image as the second image is enlarged, reduced, or rotated with respect to the registered image as the first image based on the amount of deviation. A correction unit for processing and correcting ;
The first image and the second image corrected by the correction unit are respectively Fourier-transformed and combined to extract a phase, and the extracted phase is subjected to inverse Fourier transform to obtain a peak position. A phase only correlation section ;
Identifying a plurality of candidates for the two-dimensional positional relationship between the first image and the corrected second image based on the plurality of peak positions output from the second phase-only correlation unit Candidate identification means to
In each of a plurality of different two-dimensional positional relationships between the first image specified by the candidate specifying means and the corrected second image, the first image and the corrected first A similarity generation means for generating a similarity by comparing the two images ;
Integrating means for integrating the similarities generated by the similarity generating means ;
An image collating apparatus comprising: a collating unit that collates the first image with the corrected second image based on the similarity accumulated by the accumulating unit. The image collating apparatus according to claim 1, further comprising a logarithmic conversion unit that logarithmically converts the Fourier transformed image before the coordinate conversion unit. The image collating apparatus further includes a similarity correction unit, and the similarity correction unit includes linear patterns in different directions of the first image and the second image corrected by the similarity generation unit. The similarity is obtained from the intersection, the correlation of the intersection of the central portion is determined based on the presence or absence of an intersection other than the central portion in a predetermined number of pixels around the intersection, and the similarity is corrected using the correlation and the integration is performed. Output to the means,
The image collation apparatus according to claim 1 . The similarity correction means calculates the total number of intersecting points obtained for the entire image composed of linear patterns in different directions of the first image and the corrected second image. The value divided by the score is used as an effective rate, and the effective rate is multiplied by the similarity generated by the similarity generation unit to perform correction processing, thereby generating a corrected similarity.
The image collation apparatus according to claim 3 . A first step of Fourier transforming the first image and the second image in each of a plurality of different two-dimensional positional relationships between the first image and the second image;
A second step in which the first image and the second image are subjected to Fourier transformation and then subjected to polar coordinate transformation to obtain a phase from the polar coordinate;
A third step of synthesizing phases corresponding to the first and second images obtained by coordinate transformation, and inverse Fourier transforming the synthesized phase to obtain a peak position in the correlation intensity image;
The amount of deviation of the peak position in the correlation intensity image from the image center is detected, and the collation image as the second image is enlarged, reduced, or rotated with respect to the registered image as the first image based on the amount of deviation. A fourth step of processing and correcting ;
A fifth step in which the first image and the corrected second image are respectively subjected to Fourier transform and then combined to extract a phase, and the extracted phase is subjected to inverse Fourier transform to obtain a peak position ;
A sixth step of identifying a plurality of candidates for the two-dimensional positional relationship between the first image and the corrected second image based on a plurality of the peak positions ;
In each of a plurality of different two-dimensional positional relationships between the first image in which the plurality of candidates are specified and the corrected second image, the first image and the corrected first A seventh step of generating similarity by comparing the two images ;
An eighth step of integrating the generated similarities ;
An image collating method comprising: a ninth step of collating the first image with the corrected second image based on the integrated similarity .

Images