Research On Limited-set Defaced Chinese Character Recognition

Limited-set defaced Chinese Character Recognition is the research with important meaning of Chinese Character Recognition, which plays great part in License Plate Recognition System and ID card Characters Recognition. The integrity Printed Characters Recognition has achieved very good results, but little research has been done in the Limited-set Defaced Characters Recognition, and therefore it has great practical significance and research space.The objects of this system is about 100 Chinese characters with a certain extent defaced by scanned. The entire system is divided into the following steps:1. In Chinese Character image's preprocessing, Because of deference of illumination intensity and the angle of image acquisition, the character image scanned may vary in magnitude and gray-scale, so the image should be smoothed, binarized and normalized . The characters selected for experiments in system are relatively clear, with a little noise. so the linear file and algorithm of globally binarization , as following the linear normalization method was implemented in accordance with the characteristics of graphics.2. The characters feature extraction, first introduce the commonly feature extraction algorithm simply, bring forward sub-stroke feature extraction based on eight pixel block targeting the study of the Limited-set defaced Character subject, then , we obtain the sub-stroke merger algorithm which possibly bring us the basic stoke of Chinese characters eventually.3. In recognition phase, this paper presents a plan to improve the two-tier serial classification structure, which divide the limited-set characters into three categories, named left-right, up-down, and the other three groups. In order to identify a subset of Chinese characters, we need make the characters of left-right and the up-down structure for a rough classification according to standard parts. In particularly classification stage, the characters or sub-characters are matched based on character stroke information link matching algorithm which are classified by three cases in the character depot.We select about 100 samples, some have a certain degree of pollution. The rate of recognition is about 92%.