Research And Application Of Signature Extraction And Recognition

As a way of document signing and material certificating in people’s daily lives, seal has a history of several thousand years. It is widely used to this day cause it’s convenient to use, simple to verify and easy to preserve. At present, seal is still widely used as a major document signing means in the oriental countries.With the improvement of science and technology, the seemingly sound liable seal is indeed being challenged. Those seals counterfeited by the outlaws through hi-tech means are extremely similar to the genuine one. While the simulation degrees of those counterfeit seals are becoming higher, they possess a potentially larger threat to social security and stability. Every year, counterfeit seals result in huge economic losses for banking and financial industries, governments, enterprises and public institutions, disturb economic order. The traditional verify mode is folding a piece seal impression on the paper, and then contrasting and stitching the folding seal impression with the reserved seals. This traditional verify mode could identify real or fake, but it’s complicate to perform, and its manual operation process is inefficient, thus it’s hard to deal with massive of document seal verification.In view of current situation and problems, scholars of both home and abroad studied and improved the algorithm of seal auto-authentication, and made great achievement on this issue. In this paper, through comparing and learning domestic and international advanced algorithm, the author systematically analyzed and studied the main solutions. Based on the main solutions, the author combined the algorithms organically and improved them, thus better finish the seal extracting and identification task of the test image.In the content of this paper, the author firstly compared the difference and similarity between RGB color model and HIS color model, and then extracted the seal picture from the test image according to their color characteristics, and then binarized, de-noised and repaired the extracted seals. The interference information was minimized after processing. After this, we used SIFT algorithm to extract the SIFT proper vector from the image, and conducted a rough recognition to those seals based on their SIFT characteristics, thus to filtrate those test seals which differs greatly from reserve seals. While the seals passed the rough recognition, program would conduct images registration on seals, got the edge difference between the teat seals and reserved seals, quantified the edge difference, and determined authenticity according to the quantitative result, thus finished the fine-grained identification of the seals.The method proposed in this paper can accurately perform the extraction and identification task of the seal. What’s more, this paper adopted the combination mode of rough identification and fine-grained identification. Using the method, not only the identification task was better performed, but also the program execution was more efficient and was of more practical value as it differentiated different inputting seals.