Abstract: Wireless Capsule Endoscopy（WCE） has been introduced as a noninvasive color imaging technique for the inspection of the small intestine. In this paper, a method for detecting intestinal villi images in the WCE examination video is proposed. It is based on texture features of the image. The potential villi regions in the image are segmented, and reformed to rectangular areas by a geometric transform. Frequency of small stripes in the rectangle is applied as the intrinsic feature for villi bands, and classified the images by them. The experimental results reveal the proposed method performs better than the traditional texture operators, and is more suitable for clinical applications.

Key words: capsule endoscopy, intestinal villi images, texture features, villi regions

摘要： 人体小肠绒毛特征与多种疾病有重要关系，使用胶囊内镜分析小肠绒毛是医学界新出现的研究方法。为了帮助医生检索胶囊内镜图像序列中的绒毛图像，提出了一种基于纹理的自动检测方法。分割出图像中可能包含绒毛的带状区域，利用几何变换把带状区域转化成方便计算的矩形区域，使用边缘、颜色、条纹频率等特征构建一个分层算法判断该图像是否为绒毛图像。实验部分比较了提出的方法与传统纹理描述算子的检测效果，结果表明提出的方法刻画了绒毛带的本质特征，能更好地检查出序列中的绒毛图像。

关键词: 胶囊内镜, 绒毛图像, 纹理特征, 绒毛带