Research On Official Seal Identification

Seal imprints, which signify the authority of official documents in the legal sense, arewidely used for validating official documents. Many criminals counterfeit seals to engage inillegal activities because of the special significance of seal imprints. However, traditional manualmethod of seal identification is inefficient, restricted by subjective factors and objectiveconditions. Thus, to develop the study of automatic seal identification technology has muchhigher theoretical significance and practical significance.With the application of image processing techniques and pattern recognition techniques, thisdissertation mainly focuses on the research of automatic identification of circle seal, whichdeeply discusses some theories and methods on automatic seal verification technology. Themajor innovations of this dissertation are listed as follows:1,According to the color and structural features of circle seal on official documents, animage pre-processing method for seal identification based on self-adaptive inpainting isproposed. The loss of seal information, which results from the text superimposition in theprocess of image segmentation, is resolved; Furthermore, the structure of seal image isn'tdamaged in the process.2,SIFT feature matching algorithm is introduced to the registration of seal images. In orderto resolve the problem, that the registration precision of existing seal registration algorithmsdepend heavily on the quality of stamping, the SIFT algorithm, which can extract much morefeature points and has good noise immunity, is used to achieve the registration. Moreover, thealgorithm successfully overcomes the problem that there is no information to use, except forthe seal's geometric structure. Good results are obtained in practice.3,Improved identification method, based on Zernike moment feature and Support VectorMachine classification, is proposed. The results of the existing identification algorithm basedon Zernike moment invariants is heavily dependent on the quality of stamping, According tothe actual conditions of seal identification, an improved strategy, which combines Zernikemoment feature with Support Vector Machine classification, achieves the authentication.Comparison test verifies the effectiveness of the improved algorithm.4,Voting algorithm is introduced for integrated decision-making ruling by the results of allidentification algorithms. Experiments show that this measurement can avoid the lack of asingle identification algorithm.