摘要: 受消化道蠕动和光照等因素的影响,常用计算机辅助分析方法对胶囊内窥图像的识别结果不佳。为此,提出一种基于模糊纹理谱的胶囊内窥图像识别方法。在图像的各个分量上引入模糊纹理谱,分别提取特征向量,并利用BP人工神经网络进行训练和识别,对不同分量图像采用投票原则确定最终识别结果。实验结果表明,该方法对内窥图像中正常图像和肿物图像的识别率达到92%,可有效辅助临床医生对胶囊内窥图像的筛查工作。
关键词:
胶囊内窥图像,
纹理特征,
模糊纹理谱,
多BP神经网,
投票原则
Abstract: Capsule endoscope image is sensitively affected by factors such as digestive tube peristalsis and illumination, so common computer aided analysis methods can not achieve good recognition result. Aiming at this problem, a capsule endoscope image recognition method based on Fuzzy Texture Spectrum(FTS) is proposed in this paper. It composes feature extraction of FTS of each image channel and voting decision using multiple BP artificial neural network. Experimental result indicates that the proposed method gets a recognition rate of 92%, and it can be used as a supplementary diagnostic tool in screening of capsule endoscope image for clinicians.
Key words:
capsule endoscope image,
texture feature,
Fuzzy Texture Spectrum(FTS),
multiple BP neural network,
voting principle
中图分类号:
李凯旋, 孙宇千, 刘哲星, 刘思德, 石锦平, 吕庆文. 基于模糊纹理谱的胶囊内窥图像识别[J]. 计算机工程, 2012, 38(19): 229-232.
LI Kai-Xuan, SUN Yu-Qian, LIU Zhe-Xing, LIU Sai-De, DAN Jin-Beng, LV Qiang-Wen. Capsule Endoscope Image Recognition Based on Fuzzy Texture Spectrum[J]. Computer Engineering, 2012, 38(19): 229-232.