A Research Of Biomimetic Recognition On Neighborhood Growth For Off-line Handwritten Chinese Character

Off-line handwritten Chinese character recognition is still considered one of the most difficult issues in the field of pattern recognition, thus is considered one of the main obstacles to handwritten Chinese character information technology. Its research is of great significance for Chinese information processing, office automation and computer intelligent input and so on. As the human cognitive system is a strong multi-dimensional, fault tolerance, development, integrated text recognition system. In this paper, humanoid recognition system is a foundation, nearest neighbor algorithm is integrated. The fracture problem of cross uneven vertical is not straight and strokes of handwritten Chinese characters often appear to get a good solution. It has great theoretical significance and practical value.The following are the major contents in the thesis:1、 Stroke characteristics of Chinese character are analyzed, fault-tolerant characterization methods for offline handwritten Chinese characters topology graph is established. Human cognitive mechanism of Chinese characters is analyzed, Chinese character image is transformed into a fault tolerant Chinese character basic stroke types (horizontal, vertical, write, Na) in different locations of the fold stroke graphics and strokes intersect coincide topology graph.2、 A neighborhood growth covering algorithm based unit elliptical horizontal, vertical, write, flick of the four directions of neurons traverse handwritten Chinese character images is established. Knowledge of the Chinese character image feature based on biomimetic pattern is gained. Handwritten Chinese character multi-dimensional characteristics of the knowledge data structure tables are established.3、 The rule of recognition to match the fault tolerance comparison of off-line handwritten Chinese character is given through the imitation of human learning, memory, comparison, judgment.SCUT-IRAC HCCLIB handwritten Chinese character samples library of the Chinese character image is selected. The recognition method based on neighborhood growth biomimetic pattern is verified by simulations. The expriment is prved to be close to human fault tolerance and veracity.