Off-line Handwriter Identification Technology Research Based On Character Elements

Handwriter Identification (HI) means to match an unknown handwriting script with the most likelihood handwriting sample which has been assured of its writer in the handwriting database to determine its real writer. It belongs to the category of identity authentication. In reality, it is widely used in the fields of forensic evidence, criminal investigation, security guard, and so on. It could contribute a lot to the safety of state, the normal and regular life of people and the healthy development of economy. With the development of social economy, social issues involved with handwriting fraud have become more and more. HI based on computer intelligence can help experts lessen the range of suspicions, mark out the key distinguished samples and offer the objective evidence, and make more scientific and rational identification conclusions.In this paper, we comprehensively reviewed the history of HI and the research status quo, deeply analyzed its move direction, generally discussed the current research methods and unresolved problems in this field. Basing on that, we carried out our own research work.The chief contents of this paper and innovations in the study:(1) Defined character elements (CEs) in virtue of amount of information. proposed new HI frame based on CEs, constructed the method for computing the amount of information preserved in strokes, and gave the discrimination way of CEs and the segmentation criterion for non-CEs.(2) Proposed the method for constructing feature elements template based on ISODATA clustering algorithm and gave the discrimination method based on support vector machine (SVM). Comparing with the conventional HI method, our method has advantages of lower error rate and more stable capability. Moreover, on the basis of CEs, the process of HI becomes more feasible and valid.(3) Proposed the method of constructing feature elements template based on Self-organized feature map (SOFM) clustering algorithm, introduced the idea of contribution scores of CEs and constructed the discrimination method based on the sorting order of contribution scores. Test results showed that our method preserved lower error rate and more robust performance. We hope that our researches could provide a new idea for the research of handwriting identification with computers.