Research And Application System Design For Handwriting Chinese Character Recognition

Handwriting Chinese Character Recognition is an important part of patternrecognition, and it has been widely noticed. Chinese character has a large amount ofclasses and a much more complicated structure, and handwriting character varies indifferent styles. All of these facts add much difficulty for recognition. Despite all of these,the approaches of handwriting Chinese character recognition have still achieved greatsuccess and even have been commercialized.In all these approaches the selection and extraction of features is an important issue.Due to the structure of Chinese character, direction of the strokes contains very importantinformation. So strokes features can be a useful feature for recognition. Directionhistogram features extract gradient direction from edge of the character, then get regionalstatistical histogram as features. In the recent years, sophisticated classification techniqueshave been well studied, which yield higher recognition accuracies. Support VectorMachines find a optimal classification plane in sample space, which the nearest samplepoint to the plane has the largest distance. Since Support Vector Machines allow thenumber of support vectors to be small compared to the total number of training examples,Support Vector Machines can be expended for incremental learning.Here we compare the effect of direction histogram features by experiment, anddesign a handwriting Chinese character recognition system for a specific application. Thissystem extracts handwriting Chinese character from specific application video, thenextracts direction histogram features from handwriting character images, uses SupportVector Machines for training and classification, and makes it suitable for incrementallearning. We will use1-v-1SVMs for multi-classification, and make improvement onvoting strategy, and use the result of voting for self incremental learning and artificiallyadd priori knowledge as rules to adjust recognition. The experimental results show thatthis system gets a good accuracy for this specific application.