Study On Seal Identification

With the development of the computer techniques, difficulties of forging seal decline. Therefore economic crimes caused by forgery seals are hard to be forbidden. The purpose of seal identification is to provide a system that can identify tested seal images to be genuine or forgery, and it also can offer an automatic, efficient and accurate documentation authority identification way to avoid crimes.Genuine and forgery seal identification techniques are studied in this paper. Firstly, the complex characteristics of the seal images caused in the process of producing conditions are analyzed. To solve these problems respectively, the circularity clusters and the Ostu method are firstly used to realize the shape classification and threshold processing of different seal images. Then the image denoise is performed well by scanning beam seed filling and labeling algorithm. After the processes above, the good binary seal image can be obtained.On the base of good binary seal images, two identification means are applied to extract seal features. One means is to extract feature from the tested seals after registration to the standard seal image. This means has advantage in describing the detail features. Another means is to extract rotation-invariant and translation-invariant features without registration. The purpose of this means is to avoid error cause by registration.According to the first means and disadvantage of classical seal image registration methods, a double-level registration method , which includes a projection convolution in polar coordinates in rough registration level and a heuristic search in precise registration level, is used to complement the precise registration. After image registration, two methods of featuresextraction are proposed. The first method is to analyze the characteristics of diff-image from the viewpoint of structure analysis to obtain features that can reflect difference between genuine-difference and forgery-difference. The second method is to extract frequency features by multiple Gabor filtering.According to the second means, the usability of common invariant features is analyzed and the invariance of algebraic features is discussed. After that, a SVD of polar seal image matrix method is mentioned to extract vSV features.In this way multiple features are extracted according to the characteristics of structure, frequency domain and matrix speciality. These three features can well describe the characteristics of seal image. With these features, fisher discriminant classifier and support vector machine are imposed to carry on efficient classifier selection and multiple classifiers combination. Then a voting rule is utilized for decision fusion of multiple efficient classifiers.In this paper, image process and pattern recognition techniques are used (n identify the forgery seals from the genuine seals in a seal identification system, which facilitates image segmentation. multiple supplement. feature extraction, multi-classifiers decision fusion algorithm. Experimental results show that the approach presented has good robustness to noise sealing conditions and satisfactory tolerance of difference within class.