Bank Check Print Chinese Character String Identifying The China-africa Amount

Research on Printed Chinese character string recognition not only helps further improve the automatic procession degree of financial notes recognition system, but also is conducive to bring Printed Chinese character string recognition in financial notes recognition toward the practical. Besides, it has a broad application prospects. So, it is a valuable research topic.In this dissertation, Printed Chinese character string which is non-amount in check image of CCB is selected as the research object. Several key techniques which are pre-processing, segmentation, recognition and post-processing are discussed.Change color image into gray image. For the red seal in some check image, effective gray processing strategy is raised. First, judge the color of Chinese character string. For the blue and black Chinese character string image, an algorithm on the basis of keeping the image information of Chinese character string region is raised to remove the effect of the red seal. For the red Chinese character string image, we try to solove the obvious gap situation between Chinese character string and seal. Using the weighted average average method to gray, and then look for a blank area based on gray projection of its binary image to separate Chinese character string image and seal.Tilt correction of Chinese character string. Tilt often appears when Chinese character string is printed. It needs to be corrected in order to the subsequent segmentation and feature extraction. First the method of straight fit is used to calculate the inclination angle. This method can solve most of the inclined situation. For the remaining situation which Chinese character string is not the whole inclined, the method of segmented polyline fit is proposed and has achived very good results.Two classifiers is used to Chinese character recognition. First grade coarse clssification is key to improve recognition speed. A coarse clssification scheme based on the class center and nearest neighbor method is presented in the dissertation. An improved nearest neighbor clssification is raised in fine classification. Experiment results show that proposed method has excellent performance on recognition speed and recognition rate.Recognition post-processing. A method which combines N-gram language model based on statistical with Chinese character recognition is discussed in this dissertation. For the particularity of Chinese character string to be recognition, a post-processing method based on finding and matching Chinese character string is raised in this paper. Finally, the two post-processing is combined. The experiment proved that the recognition of the whole string is86.72%.