Factors Affecting The Recognition Rate Of Printed Uyghur And The Research Of Countermeasures

The printed Uyghur recognition technology has made certain achievements. But as there are still problems on dealing with some local details, that causes the recognition rate further increased more difficult. This paper carefully tests and analyzes a lot of identification samples, from which we summarize the prone to erroneous identification character forms and combinations. We find a part of false consciousness reason of common erroneous identification strings combined with analyzing the original algorithm we used, and put forward new improved algorithms for segmentation and recognition. The concrete work done in this paper is as follows:1. Through the test of the existing software, we analyze the main factors affecting the recognition rate in printed Uyghur recognition system in detail. The segmentation method and over segmentation combination method as well as character recognition method the project team used before combining with examples are discussed specifically. We make a detailed analysis of their advantages and disadvantages in the practical application of the process, and put forward the corresponding countermeasures in view of the limitations.2. On the basis of the original vertical projection segmentation method, and for the overlap problem between two conjoined sections, an improved drop fall algorithm is proposed to achieve that. On the basis of over segmentation based on vertical projection method which the baseline is set white, an improved moving window method combined with the number of characters information is proposed to be used to evaluate character segmentation position synthetically. We use this recognition feedback to guide segmenting, so as to realize the over segmentation combination.3. For the problem that there are many similar Uyghur letters indistinguishable, a recognition method based on the separation of the main and subsidiary stroke is proposed. First, according to the letter writing feature and the subsidiary stroke position feature, letters are classified into 12 subcategories coarsely. For the letters which have subsidiary strokes among them, the main and subsidiary stroke should be separated in advance, and the MQDF classifier is used to recognize them respectively. Then the two parts are merged as the final recognition result.The experimental results show that the segmentation and recognition algorithm has achieved certain results in the utilization, and the segmentation and recognition accuracy of the system has been improved.