The Research Of Handwriting Identification Based On Feature Fusion

Handwriting can reflect a person’s writing style. Each person’s handwriting has its own unique physical characteristics, which is formed by the long-time different writing habits. Handwriting identification, because of its high collection and acceptability, has become a very active research topic in the field of identity authentication.The text structure is mainly used as the research object in the traditional feature-extraction methods of the handwriting. When the text content exist much difference, it will increase the clas s changes to a certain extent and reduce the identification rate. The strokes, as a basic part of the text, still have a great reproducibility when it has low text related degree or exists a large number of typos.According to the problem that the feature-extraction method which use word structure as the research object of the off- line handwriting, cannot obtain a stable feature with low relativity, a method for extracting the strokes as the research object of feature fusion is put forward. Introducing the probability statistics theory and using the grid window to extract the brush strokes, width variation trend and curvature feature of the edge. It uses Visual Studio 2010 as the platform to realize the system.In this experiment, the pretreatment of handwr iting image mainly contains the removal of noise and background, gray, binaryzation, the extraction of edge and skeleton. Comparing the four classic edge extraction algorithm of handwriting, it shows that the Sobel operator has a better effect in edge extraction of handwriting through the analysis of experiment. It can obtain a much ideal skeleton image by using the thinning algorithm which is based on the improved traditional thinning algorithm.The text improves the micro-structure characteristics method. Taking strokes as the research object, getting rid of the constraints of body dependence and using the grid window to extract the brush strokes, width variation trend and curvature feature of the edge. Introducing the idea of the principal component ana lysis, merging data of multi feature and generating feature vector.In matching operation of the handwriting feature, this paper uses weighted Euclidean distance, weighted chi square distance and weighted Manhattan distance to calculate the similarity of handwriting.An offline handwriting identification system is realized. The library contains handwriting of 120 people. Compared with the existed identification method in the experiment, the correct rates of top-ten selected identification are 90.7% and 98.3% when correlation degree is lower. The experiment shows that, the handwriting identification method which is based on the feature fusion and use strokes as the object can still achieve better results under lower correlation degree.