Font Size: a A A

A Wavelet Transform Based Chinese Handwriting Automatic Signature Identification System

Posted on:2006-09-09Degree:MasterType:Thesis
Country:ChinaCandidate:H LiFull Text:PDF
GTID:2208360155966782Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
Signature is one of the biological features used abroad to verify identity and the research into signature verification is indeed significant both to practical application and the development of science. But to date, it is still far from satisfactorily solved. In china, research in such field has just begun.The automatic signature verification could be off-line or on-line. Comparatively speaking, the former is less limited in equipment involvement and applied in more fields. Nevertheless, it is harder than the latter due to losing the dynamic information during the writing process.In this paper, a wavelet-based off-line handwritten signature verification system is proposed. We attempt to use the coordinate data of the traced closed contours as bases to extract features. As the closed contours can trustily reflect structural information of original signature. The proposed system starts with a closed-contour tracing algorithm. If we use the second derivative of a smoothing function as mother wavelet, the zero-crossings of the transformed highpass data naturally indicate sharper variation points. According to this theory, we select Marr wavelet as mother wavelet in this paper. The coordinate data of the traced closed contours are decomposed into multi-resolutional signals using wavelet transforms. Then the zero-crossings corresponding to the coordinate data are extracted as features for matching.Due to different personal writing styles, it is impossible to uniquely determine a global threshold value that fits all writers. A statistical measurement is devised to decide systematically which closed contours and their associated frequency data of a writer are most stable and discriminating. Based on these data,the optimal threshold value which controls the accuracy of the feature extraction process is calculated. Moreover, a simple grade verification method is adopted to improve the testing speed in this paper.To quantitatively evaluate the generalization capabilities of our .approach, we gathered 25 writers' signatures, each writer providing 20 genuine signatures and 10 forged signatures. And we group these signatures for testing or used as database data by different combinations. Experimental results show that this verification system is effective and feasible.
Keywords/Search Tags:off-line handwritten signature verification, wavelet transform, zero-crossing, dissimilarity degree, dynamic time warping
PDF Full Text Request
Related items