Offline Signature Verification Based On LBPC And LCPC Features

Offline signatures have been widely used as a traditional identity authentication method.For the authenticity identification of signatures,the method of manual identification is still mainly used,which is inefficient.The use of computers to realize the authenticity of offline signatures has a wide range of application requirements and theoretical research significance.The LBP feature can reflect the texture features of the signature handwriting,but lacks a description of the geometric characteristics of the signature contour.This paper studies the offline signature identification method based on LBPC(Local Binary Patterns on Contour)features and LCPC(Local Contour Pattern Co-occurrence)features.The major tasks of this article are as follows:(1)In order to improve the validity of LBP features on the signature image,the LBP features were combined with the contour feature.By analyzing the LBP image of the offline signature,the ratio of background points was found to be very high,and local binary patterns were the same on the background.In addition,the change of texture on LBP image was mainly reflected on signature's contour.Therefore,a statistical histogram feature of LBPC was proposed in this paper.In order to extract the information of the local position,the LBPC feature was fusion with the mesh feature.(2)Based on the analysis of traditional contour directional features,a histogram feature based on local contour pattern co-occurrence(LCPC)was proposed.The traditional contour direction chain code feature only counts all changes of three consecutive pixels,which has limitations in describing stroke habits.So,in this paper,the basic contour patterns were mentioned to characterize signature's stroke and the concept of pattern co-occurrence was introduced.And the LCPC-based statistical histogram feature was further proposed.In order to extract information of the local position,the LCPC feature was also fusion with the mesh features.(3)In order to prove the validity of the proposed LBPC and LCPC features,experiments were conducted on two open offline signature databases.The LBPC features were tested on the MCYT-75 and GPDS-160 datasets.Using the Chi-square distance discrimination method,the equal error rates obtained were 13.49% and 11.31%,respectively.And the average error rates obtained were 13.73% and 15.3% when the SVM classifier was applied.Besides,the LCPC features were also tested on the MCYT-75 and GPDS-160 datasets.Using Chi-square distance discrimination method,the equal error rates obtained were 11.66% and 9.89%,respectively.And the average error rate obtained were 11.59% and 13.58% when the SVM classifier was applied.Finally,considering the complementarity of these two contour features,the two features were combined to further improve the performance of the system.