基于局部二元模式的电子胃镜图像识别

Electronic Gastroscope Image Recognition Based on LBP

由于受到胃部蠕动、气泡、食物、光照以及图像采集过程中摄像头移动等因素影响,电子胃镜图像存在亮度变化较大等问题,常用的计算机辅助分析方法难以取得理想的效果。针对该问题,在分析电子胃镜图像特点的基础上,提出一种电子胃镜图像病灶良恶性识别方法。在不同颜色通道中使用结合局部二元模式算法,提取其纹理特征向量,分别输入支持向量机进行训练和识别,对不同颜色空间的识别结果采用投票原则确定最终结果。实验结果表明,该方法的识别率达到92.2%。

The electronic gastroscope image, which is sensitively affected by gastric peristalsis, air bubble, food, illumination and camera moving during the capturing, has some defect such as large illumination variety, so common computer aided analysis method can not achieve good result. Based on analyzing the characteristic of electronic gastroscope image, a new method to distinguish between carcinoid electronic gastroscope image and malignancy ones is presented. It uses the Local Binary Pattern（Lt3P） method in RGB channel of images and extracts the texture feature which is subsequently trained and classified by Support Vector Machine（SVM）, gets the classification of all three channels and decides the image carcinoid or malignancy through voting principle to supply the clinical analysis. Experimental result indicates the method gets a recognition rate of 92.2%.