Research On Online Handwriting Verification Based On Chinese Structure

Handwriting has always been an important means for identity authentication.With the development of biometric identification technology,handwriting verification is a hot research topic in the field of pattern recognition.So far,the off-line handwriting verification research is relatively mature.Because the advent of the information age,the stylus pen-based input devices are cheaper than before,the paperless office and paperless trading have become widespread,most writers tend to sign online instance of offline.Online handwriting is the trend of future,but there are still many problems to be solved for online handwriting verification.This paper summarizes the characters of Chinese handwriting.For the instability of feature extraction and poor result of verification,the method of feature extraction is improved and structural feature of Chinese handwriting is extracted in paper,which focus on the text-dependent online handwriting verification and puts forward the method of online handwriting verification based on Chinese character structure.The main work of this paper includes:(1)A two level matching framework which is proposed based on statistical feature and structure feature.In the stage of feature extraction,two kinds of features are extracted: statistical feature and structural feature.We choice seven kinds of statistical features which are classified by majority voting procedure as the rough classification of handwriting;We use the structure feature as the fine classification,which is the second level matching,in this way we can reduce the matching time.(2)In the stage of structural feature extraction,two different structural feature are compared and finally we use the stoke segment instance of stroke as the structural feature.In order to extract the stroke segment feature,we summarize some traditional algorithms and select the segment extraction algorithms based on inner angle and polygonal approximation.With the algorithm,we can find the feature point through inner angle,and two neighboring of which determine a segment.At last we combine the segments to get the segment feature.(3)The Hausdorff distance is used as the similarity metric of structural feature.Hausdorff distance and several traditional modified algorithms are analyzed and we propose a new modified algorithm which is based on the partial Hausdorff distance and the modified Hausdorff distance.The new modified Hausdorff distance removes the maximum and minimum values of the directed Hausdorff distance,and we treat the average value as the new Hausdorff distance.(4)In the process of matching,the paper improves the matching method.First of all,we normalized all the stoke segments to reduce the distance between sample stoke and reference stroke.Then we detected the invalid segments in order to provide the mismatching.What's more,we matched the segments in dynamic way and we deleted some points from the reference segments to get better matching rate.(5)The handwriting verification experiment was carried out in the handwriting database called WHUT-13 DZ,which included 30 writer's handwriting.Every writer has written six times with twelve different of Chinese characters.In the experimental,the new modified Hausdorff distance got the best result and the FRR is 1.1% and the FAR is 1.2%.