The seal detection and confirmation is one of the important task in the business process of paper form, because the seal means a promise of commerce and payment. This paper describes a seal detection method from color, gray or monochrome document image. The basic process of seal detection consists of two steps: extract a contour from input image, and then make hough-voting with tracking contour. However, simple hough-voting method is not suitable for realtime processing and it is weak at image disturbance such as rubbing, lacking, oozing of seal image, piling on form background object etc. We have introduced some improvement into hough-processing on document image. First is contour approximation with line segment and some operations on the set of lines in order to compute image smoothing. Second is simplified hough-voting using line segment. Some experiments were performed with collected form samples including various seal type, and this method achieved 98.7% of recognition rates.The seal detection and confirmation is one of the important task in the business process of paper form, because the seal means a promise of commerce and payment. This paper describes a seal detection method from color, gray or monochrome document image. The basic process of seal detection consists of two steps: extract a contour from input image, and then make hough-voting with tracking contour. However, simple hough-voting method is not suitable for realtime processing and it is weak at image disturbance such as rubbing, lacking, oozing of seal image, piling on form background object etc. We have introduced some improvement into hough-processing on document image. First is contour approximation with line segment and some operations on the set of lines in order to compute image smoothing. Second is simplified hough-voting using line segment. Some experiments were performed with collected form samples including various seal type, and this method achieved 98.7% of recognition rates.%入出金や物品取引の証明として発行される伝票類を取り扱う業務では，日付印（領収印）の有無をチェックして当該伝票を分類することがタスクの一つとなっている．本報告では，カラー画像，グレー画像または二値画像として入力された帳票画像から，日付印の有無判定，更に押印された領域を正確に検出する自動認識機能について述べる．本方式は，文書画像から輪郭を抽出し，輪郭追跡によるハフ投票により，所定形状を持つ印影を検出する．押印のかすれ・つぶれ，背景模様と押印の重なりなどの外乱要因に対応し，かつ処理速度の高速化を図るために，輪郭の区分線分列での近似と，区分線分列に対するスムージングなどの輪郭演算，省略化したハフ投票を導入することで改善を行った．自社収集サンプルで評価実験を行った結果，日付印の正検知率98．7％を達成した．
展开▼
