The machine recognition of handwritten Chinese characters has been studied for nearly three decades. However, there still is a considerable gap between the theoretical research and practical applications. It was once considered to be a very hard problem and regarded as one of the ultimate goals of character recognition research. It has four inherent difficulties: (1) There are a large number of Chinese characters. (2) Chinese characters are more complicated than other characters. (3) There are many similar Chinese characters. (4) The shape variation of Chinese characters caused by different writers is very large. Among these difficulties, the shape variation is the most serious problem. However, the shapes of characters written by one person are relatively stable. Based on this consideration, it is expected that a handwritten Chinese character recognition system will perform better if it is at first adapted to the writing style of one person and then recognizes the characters just written by the same person. This is so-called personal or writer-dependent handwritten Chinese character recognition. It was introduced as a feasible way to bring handwritten Chinese character recognition into practical use. The research work described in this dissertation is mainly centered about this topic. After a careful investigation on various methods for the recognition of handwritten Chinese characters, several simple and efficient algorithms are proposed and a personal handwritten Chinese character recognition system is developed. First, a linear size normalization algorithm is proposed, which is especially suitable for personal handwritten Chinese character recognition. It can maintain the relative position of the centroid as well as the ratio of the width and height of a character image, thus highly reflects the personal handwriting characteristics. Its operation is simpler and its performance is better than that of other traditional linear and nonlinear size normalization algorithms. Second, a flexible template matching algorithm for handwritten Chinese character recognition is proposed on the basis of a dynamic model introduced by Burr in 1980. It can deform a temple Chinese character by using the dynamic model so that the template is closer to an unknown character before performing the similarity measure between these two characters. It can also average the images of several samples of a Chinese character to produce an intermediate image as the standard template of this character. Third, based on the consideration that handwritten Chinese characters are complex two-dimensional shapes, a character recognition algorithm using the generalized Hough transform is proposed, in which features extracted from the parameter space are used to calculate the similarity measure between two character images. Thealgorithm is finally implemented in a look-up-table scheme which can be extended to other applications of pattern
修改评论