摘要
Verifying a seal imprint to authenticate the identity is a general and important task in financial industries. The seal imprint attached to a document is often used as a form of authenticity. Therefore, in financial applications, it is often necessary to recognize whether a seal is genuine or forged. In order to simplify the verification, we propose an effective method based on regression analysis to approximate the imprint borders, and use the geometric transformation to adjust the perspective of the detected imprint image. To align the detected image with the original image, the first step is to find four borders and cross each border to obtain four vertices of the rectangular imprints. The Ordinary Least Squares regression is then applied on the outer strong feature points, and the estimated regression line would be taken as the borders. The regression process is repeated several times to estimate border lines as precise as possible. Besides, considering the practical application in taking seal imprints by camera, the imprint is subject to geometric distortion, due to camera perspectives, comparing to the stored genuine imprint image. Therefore, the detected image is geometrically transformed to align its perspective with the genuine imprint. Finally, the SSIM and PSNR are calculated as the decision metrics to check whether the two indexes exceed the predetermined thresholds. Then the imprint is determined as a genuine or forged imprint. The experiments are conducted among various stamping imprints, and the results show the efficiency and effectiveness of the proposed identification approach.
•An effective method based on regression analysis to approximate the imprint borders is proposed.•Geometric transformation is used to adjust the perspective of the detected imprint image.•The experiments are conducted among various stamping imprints, and the results show the efficiency and effectiveness of the proposed identification approach.