关键词: 维吾尔文文字识别, 连体段, 聚类算法, 等间距法, 有效相似比, 正确率

Abstract: In Uyghur character recognition, the effect of the cluster will affect the recognition rate directly. To improve the clustering result, an improved K-means clustering algorithm based on Uyghur word-part is presented. The first step of the method is to select the center of the clustering by using the equal interval method repeatedly in order to select the best codebook, then adjust the number of clustering classes （noted as K） by using an effective similarity ratio dynamically. Finally, the word-part clustering is completed. The experimental results show that：compared with the traditional K-means algorithm, the improved K-means algorithm gets a better result and the clustering accuracy is more than 90%.

Key words: Uyghur character recognition, word-part, clustering algorithm, equal interval method, effective similarity ratio, accuracy