Research On Handwritten Chinese Character Recognition Method Based On Rough Set And Genetic Algorithm

Off-line handwritten Chinese character recognition is one of the most difficult problems in character recognition, for its huge quantity, complex structure and various transmutations. Rough Set theory has been proved to be an excellent mathematical tool to deal with uncertain and vague problem. Genetic Algorithm (GA) is an adaptive heuristic search algorithm. In the paper, approach combining the two theories is used to solve problems in the off-line handwritten Chinese character recognition. The major contents in the thesis are in the following:1. Adding Chinese character real attribute as priori information, a off-line handwritten Chinese character recognition decision information system is constructed based on the concept of information system in generalized domain defined by Rough-Set.2. The idea that the essentiality of off-line handwritten Chinese character attribute can be figured by entropy value shed light on the design of an off-line handwritten Chinese character quantity reduction algorithm based on GA. It decreases the redundancy degree of attributes and simplifies the recognition process without information loss.3. Based on rule conflict judgment, a heuristic redundant sample reduction method is suggested to reduce the redundancy degree of multi-sample Chinese character to further simplify the recognition process.4. A kind of off-line handwritten Chinese character ruler fusion recognition method, based on confidence, is designed to recognize the unknown character sample, while failing to find the perfect rule for the unknown character sample in existing rule set. So it can decrease the recognition rejection of the system.The SCUT-IRAC handwritten Chinese character samples are selected as experimental samples for experimental verification. The results of experiment in MATLAB7.0 indicate:The attribute reduction method is available for the reduction of multi-dimension statistical off-line handwritten Chinese character attribute; the heuristic redundant sample reduction method is effectual for the reduction of the redundancy samples; the ruler fusion method based on confidence can increase the recognition accuracy of off-line handwritten Chinese character recognition decision information system.