Research On New Method For Seal Imprint Verification

Seal imprints are one of the most important evidence to identify the authenticity of financial instruments in China. They also are widely used in Japan, Korea and some other East Asia countries. With the development of science and technology, quality of counterfeit seal imprints is becoming higher and higher. Financial scams by using counterfeit seal imprints are repeated. During traditional manual seal imprint verification, some problems have come out on accuracy and speed. Additionally, most of existing automatic verification methods aim at hand-carved fake seal imprints, and they are not able to verify high-simulations accurately. Therefore, researches on automatic verification methods for counterfeit seal imprints with high qualities are of great theoretical and practical significance. This dissertation proposed a complete automatic verification method for high simulation seal imprints, including seal imprint image segmentation, feature extraction and registration, and discrimination analysis. A large number of seal imprints were verified. The results demonstrated the accuracy and effectiveness of the proposed algorithm. At last, the algorithm was transplanted from computer to a DSP based seal imprint verification system to test its performance and verification speed.The major innovations of the dissertation are summarized as follows:1. An adaptive morphological segmentation algorithm is proposed to obtain a binary seal imprint. Different Chinese characters have different stroke features and background evenness. A seal imprint is divided into some sub-regions according to the distribution of characters. The background across each grayscale sub-region is smoothed by top-hat transformation. The size of structuring element in top-hat transformation might have a great influence on the segmentation result. The optimal size of the structuring element for the top-hat transformation on each sub-region is iteratively estimated according to the local foreground area. Each top-hat processed sub-region is binarized by Otsu's method. Experiments showed that the proposed algorithm can effectively reduce adhesion and incompleteness distortions in the segmentation results, even when the original seal had a poor quality.2. An algorithm combined similarities on SIFT features with spatial relations is proposed to register a SS(sample seal) and a MS(model seal). In the binarized images of the two seal imprints, SIFT features are extracted and matched. To enhance the accuracy and stability of the matching result, RANSAC is used to reject the mismatched SIFT feature pairs. Then the homography matrix is computed according to the corresponding relationship between space locations of matched SIFT pairs. Due to the large quantity and rotational invariance of SIFT features, the proposed registration algorithm can eliminates the influences of deflection angles between seal images on the registration result. Moreover, the algorithm has no constraints on the shapes of seal imprints and structures of strokes.3. An algorithm based on quantified edge difference between MS and SS was proposed to automatically verify seal imprints. The edge difference between MS and a deliberately fake SS was slight, while that between MS and a genuine SS was also small due to the variety of imprinting conditions. To evaluate similarities between MS and SS, edge difference was quantified as two parameters, the distance between non-overlapped corresponding edges and their lengths. According to the two parameters, an edge difference histogram is constructed. With the histogram being the input feature vector, SVM is used to classify the SS as true or false. Experiments showed that not only could the proposed algorithm effectively verify the high-simulations, but also it had good tolerance of differences between genuine seal imprints. The correct recognition rate was over 99%. When both of the false accept rate and the false rejection rate are close to zero, the rejection rate was about 3%.4. The entire verification algorithm was implemented on an embedded seal imprint detector. With a piece of high-speed DSP(Digital Signal Processor) as the processing core, the detector was designed to integrate color image acquisition, data processing, man-machine interaction and network transmission. The entire proposed algorithm was transplanted from PC to the DSP-based system, and optimized. Experiment results proved that the proposed algorithm could verify seal imprints effectively in the DSP-based system. The verification time for a square seal was about 3 seconds, while that for a circle seal imprint was about 5 seconds.