Research On Methods Of Seal Image Feature Extraction

With unceasing enhancement of the office automation degree, traditional manual recognition seal method is now unable to satisfy the need of massive services. Thus, Seil Recognition Systems based on Information Technology are now the needing systems of the finance and public security department. Feature extraction is the most important part of Seil Recognition Systems. Thus, this dissertation mainly presents researches on feature extraction of seal images.This dissertation first introduces the requirement on Seal Image Recognition Systems. It also presents the main architecture of the main Seal Image Recognition Systems and the difficulty in the recognition process. Then, this dissertation introduces methods in the feature extraction processes. It shows details of the section of image preprocessing. This section includes the color feature extraction, high quality binarization and noise removal of the color images. Then, based on image preprocessing, the paper shows the method of geometry feature extraction. Finally, it proposes the method of the character part segmentation.In the preprocessing process, the most important step is the image binarization. After studying several representative algorithms, we use the LOG arithmetic operator method to do binarization which can enhance the edge data of characters .We use the chain code track algorithm to withdraw the outline of the seal, and through rejecting the smooth spots in the outline to obtain the vertices of the seal. Then, though the number and geometry position of the vertices, we obtain the shape of rectangular, circular, elliptic and triangular seals.It is clear that if we can't find the right center point of circular or elliptic seals, then we can't extract the character part feature successfully. So we come out an inner contour extraction algorithm to rectify the geometry center of the seals. We also come out an adaptive polar coordinate algorithm based on Bresenham algorithm to transform the seal to polar coordinate.It is proved that those algorithms can do shape extraction, geometry feature extraction and character part feature extraction effectively.