Automatic Seal Identification In The Seal Verification System Of Bank

With economic development, seal has been applied to people's daily life as a means of identity authentication. The traditional manual seal identification way has many disadvantages such as complicated process, slow speed,easy affection by artificial factor, and inability to deposit and withdraw at any allied bank, so it can't meet the demands of actual life. Under such circumstances, the seal verification system of bank is brought forward. The seal verification system of bank consists of image gather , seal identification and database management. Automatic seal identification utilizes digital image process and mode verification technology to verify the seal identification automatically, which demands it has ability of identifying mock seals and tolerating the difference of real seals. The process consists of seal's extracting and fixing, parameters calculation, match and seal identification. The seal's extracting and fixing is the basis of the whole system. Owing to the influence of the complexity of the seal background, the uncertainty of artificial factor, and the difference of the gathering condition, it's difficult to extract the seal instances. The thesis has proposed to a new color extracting seal model with consideration to the advantages and disadvantages of RGB and HSI color space, which can extract the seal image distinctly and completely. As to the deformity of the seal image, the thesis brings forward the two-level iterative filling arithmetic of symmetry distribution and equality distribution to restore the seal image. This method can restore the deformity information quite well without overfilling at the stroke connections. The precision of the seal parameter calculation determines the match degree and the seal verification rate. Because the seal frame may come up the phenomenon of rupture and blur. the traditional used method, for example outer rectangle method, moment method, and curve-fitting method can bring obvious error, and Hough transform checking curve method has too much calculation. This thesis proposes the parameter calculation method based on the seal geometry shape. Thus the real time quality is guaranteed and the calculation precision is increased with the error only in one or two pixels. Then according to the deflection angle of the acquired geometrical seal parameter, the swatch seal and the model seal are matched by way of converse transform method of image rotate. As a result, the match speed and precision are both enhanced. Considering the speed and security of identification arithmetic, the thesis adopts three-level automatic identification and one-level simulation artificial identification strategy. The primary identification of the seal is invariant moment method with the smallest calculation. The second step, The dubious seal needs to be verified by circle-projection. The automatic match verification and identification based on the difference image can be used to the remained dubious seal in third step. At last a few left dubious seals can be dealt by the simulation artificial verification. The verification strategy adopted by the thesis not only guarantees the speed of the whole system, but also enhances the verification rate of the system.