关键词: 图像识别,食管癌,内镜,聚类 中图法分类号TP391.7文献标识码A

Abstract: Videoendoscope is an important tool for the diagnosis of esophageal cancer in early stage．Computer aided diagnosis could improve the efficiency．However,there is lack of effective recognition algorithms．Therefore,the paper proposed an image recognition algorithm for esophageal endoscopy (IRAFEE) based on esophageal endoscopic image texture features．Firstly,the IRAFEE algorithm divides the endoscopic image into sub-areas,extracts gray level co-occurrence matrix for each sub-area．Secondly,the IRAFEE algorithm calculates four features of gray level co-occurrence matrix:angle second moment,contrast ratio,inverse difference moment and degree of association,constructs feature vectors．Finally,the IRAFEE algorithm clusters the feature vectors for several times,filters the result clusters based on expert rules and identifies the potential lesion areas．The experiments show that the proposed IRAFE algorithm is feasible and effective.

Key words: Image recognition,Esophageal cancer,Videoendoscope,Cluster