Research On The Image Extraction And Registration Method Of Round Seals

As an effective proof for notes and documents, seals are widely applied in modern economic and social life. The crime of counterfeiting seals has become a major hazard to violate the dignity of law, to affect credibility of the government, to disrupt the economic order and to damage social honest environment. Using such technologies as image processing and pattern recognition, automatically identifying authenticity of seals has been becoming an increasingly important subject.The purpose of this paper is to explore and research in algorithms of all stages on seals identification system. Based on the analysis of verification technologies of seals'authenticity in recent years, this paper presents an image registration method and a difference image identification method for round seals.The main contents are listed as follow:(1) In the extraction and segmentation of seal images, it converts the images from RGB space into that of HSI, segments images in color, saturation and brightness, extracts the red round seal, and then grays the seal using the method of weighted average values. Against the disadvantages of the binarization method, thresholds of which are usually too large and which has large areas of black pixels, to get ideal seal binary image it proposes a method which makes histogram on seal images, calculates the cumulative histograms, determines threshold region by the proportion of the seal in images, gets the threshold through experiments and then eliminates noise interference in seal images..(2) To solve problems of low accuracy and slow speed in seal registration, this paper puts forward a method which uses radial templates to register round seals. The method transforms seal images from two-dimensional data into one-dimensional data. And it designs 8 different radial templates of various frequencies; using radial templates it does convolution operation separately with reserved seals and ones to be verified, devises a criterion function by convolution results, and then gains optimum registration angle via minimization of the criterion function.(3) On the problem to distinguish differences of the text area in seals, this paper proposes the round seal registration method which is based on round-rectangle transformation. This method converts round seals to rectangular ones, proceeds longitudinal one-dimensional projection on the transformed rectangular seals to get projected curves, analyzes average relative errors of projected curve values under different angles between reserved seals and ones to be verified, determines the optimal deflection angles by means of circulating shift of projected curves. Then it can make round seal images from two-dimensional to one-dimensional data, and can accumulate the projection value of multiple pixels into a value in every angle. Afterwards using the difference image method it carries out xor (exclusive) operation to reserved seals and seals to be verified after registration, gets matching similarity through statistical number of black spots after applied the difference image method, and then receives identification results by comparing matching similarity and thresholds...