Study On Segmentation Of Unconstrained Handwritten Chinese Words And Procession Of Multi-sort Notes

The segmentation of handwritten Chinese words and the classification of multi-sort financial notes images is the basis of implementation of notes automatic procession. The segmentation of unconstrained handwritten Chinese words is the key problem of handwritten Chinese words recognition. It is also the difficulty of current segmentation-recognition truss.In this paper, a method of notes classification which is based on template matching and image substraction is provided. The available image is found by template matching in original bill image. After the elimination of noise by filter, the image substraction is used to detecting the existence of template in available image. The validity of method is proved by the experimental result of classification on bill images.The betterment projection method is provided in the paper. It is used to pre-segmentation of unconstrained handwritten Chinese word image. Based on the material character of unconstrained handwritten Chinese word, the classical projection method is complemented and expanded. The transcendent width of the character is obtained by minimal twin multiplex method. The projection segmentation point is acquired by use of certain rules. In addition, the program which corresponds to the algorithm is completed and applied in the cheque verification system.The method which is the combination of structure analysis and upper recognition is used to segment touching and overlapping Chinese words. The primary segmentation point is obtained by analysis of whole Chinese word image connection area. The bogus segmentation point is eliminated by analysis of contour difference and projection. By use of the combination of segmentation and recognition method, the 12,000 touching and overlapping Chinese words are segmented. The accuracy is 92.4%.Finally, a post supervision cheque automatic procession system is discussed. The segmentation of cheque factor , segmentation of Chinese money word , segmentation and noise elimination of numerical money word, recognition of unconstrained handwritten Chinese money word and recognition of unconstrained handwritten numeric money word is included in the system. The system is put in use. The effect of the system is better.