Study And Application Of Character Integrated Recognition Based On Subspace

In recent years, many new theories and methods emerged in pattern recognition field, one of which is subspace method. Recognizing small handwritten character set with this new method and constructing an integrated reorganization system with subspace classifier and BP classifier are the main objects in this paper. The background of this paper is the graduate job information input and administration system in universities.The subspace method was initially used in linear feature extraction and data compression. It can compress vector data to energy concentrative axis and map data from high dimension space to low dimension space. In character recognition, the key function of subspace method is mapping character feature vector from high dimension space to low dimension space, finding out every class's concentrative energy axis and constructing every class's subspace with the axis. By projecting feature vector to every class subspace, the character can be determined to one class in accordance with the projecting length. This is the difference between subspace method and other statistic methods. Since it is a new pattern recognition method, some fundament knowledge was discussed in this paper, such as the basic notion of subspace method, the construct of subspace classifier, the classing rules, the rejecting rules, etc. At last the representative BLSM and ALSM classifier were tested with multiple character sets. Every method has its merits and shortcomings. The direction of character recognition is integrated recognition. In this paper, the effects, the layers, the basic composing modes and the integrated algorithms of integrated recognition are deeply discussed. A mix model with the subspace classifier and BP neural network classifier was realized, which is used in handwritten English letter and number recognition. Preprocessing and feature extraction are important stages in character recognition system, related closely with the classifier precision. In the first place, character images were processed with smooth, contour, nonlinear normalization algorithms, and then different character featureswere extracted according to different objects. Font recognition based on wavelet package texture analysis was discussed in detail in chapter 2. It's another feature of this paper.The research results in this paper were used in the graduate job information input and administration system in universities. At first, the task was analyzed distinctly, and then the general project design and corresponding solving methods were put forward in detail. With character recognition technique, object-oriented programming technique and modern database technique, we help job administration department in universities fulfill this task successfully.