| Handwriting identification deals with the techniques of identifying a person by analyzing her/his handwriting styles. It has been applied to a lot of areas and become one active research subject on computer vision and pattern recognition. Many researchers have studied the online handwriting identification. However, the offline handwriting identification is seldom explored in that the latter is much more difficult than the former. Offline handwriting images can easily be imitated and copied. In addition, it is nearly impossible to detect from them the dynamic information utilized by the online systems, such as the writing speed, acceleration, and pressure and so on. This paper is concerned with the popular texture analysis technique in the traditional image recognition, the novel pattern recognition approach, namely synergetic neural network pattern recognition, and their applications. Furthermore, it proposes and implements an offline handwriting identification multi-classifier. It is organized as follows. Firstly, it introduces the basic concepts and principles of texture analysis technique. Secondly, it discusses the algorithm framework of texture segmentation based on texture spectrum and improves the corresponding algorithms. Thirdly, it studies the concepts and principles of synergetic neural network (SNN) and explores several its several models and algorithms. It also presents a hybrid approach on the basis of the synergetic network and texture spectrum, which utilizes the advantages of both the top-down synergetics and the traditional down-to-top classifying methods to construct the multi-classifier. The first level classifier is based on SNN, in which a novel prototype selection algorithm on the basis of texture features instead of pixels is presented and the second level the texture segmentation algorithm based on the global texture spectrum. Finally, it tests the multi-classifier with a greatnumber of handwriting images and discusses the experimental results.
|
PDF Full Text Request