Seal Extraction And Recognition Technology

In the oriental countries, such as China, Japan and Korea, seals are extensively used for the purpose of entity identification. In China, seals play a more important role. With the rapid step of office automation, manual seal identification has become the bottleneck. Thus, automatic seal identification has been an active topic in the field of document analysis and processing.Automatic seal identification involves image processing and analysis, pattern recognition etc. Many methods of automatic seal identification by computer have been proposed with lot of results. The main disadvantages are: Firstly, most of the methods can't be applied directly to real application. Because they deal with an image in which only a seal exists. Secondly, the recovery of missed stroke is unsatisfied. Thirdly, much restrict is put on the seals such as the seals used are square in shape, the seals can not be set upside down and can't be rotated morethan 45°.In this paper, we discuss in detail about the automatic seal identification from the aspects of theoretical researching and practical realizing. A whole method is proposed using image processing and analysis, pattern recognition theory. The proposed method involves pre-processing, location by projection, recovery of missed stroke based on overlapped point, outer-most frame tracing and seal segmentation, extraction of one-pixel-wide edge, registration based on contour analysis and circle fitting, similarity measure based on distance-weighted correlation. Firstly, the region including seal is located and segmented by projection. Secondly, the seal with missed stroke is extracted from the saturation channel image and the missed stroke is recovered. Thirdly, registration is accomplished based on contour analysis and circle fitting. At last, the similarity is computed based on distance-weighted correlation.Experimental results show that the method is stable and fast. It can process seals with variant shape and size. It approaches the level of real application.