【摘要】 提出了一种基于胃镜图像的计算机辅助病灶检测方法。首先,引入超像素理论,将胃镜图像分割成大小均匀且包含相似像素的若干区域;然后,分别提取颜色特征和纹理特征,并将其融合作为特征描述符;最后,采用二级串联分类器进行胃镜图像内干扰区域的去除以及病灶区域的识别。实验结果表明,本方法病灶检测正确率(AUC)可达到91.588%。更多还原

【Abstract】 This paper proposes a computer-aided detection method based on gastroscope images. Firstly, segment the gastroscope images based on the theory of superpixels to obtain the areas of uniform pixels. Then, extract color histogram and combine with the LBP texture as the feature descriptor. Finally, two-stage tandem classifier is utilized to wipe off the interference and recognize the lesion region. This method performs well with the 91.588% accuracy.更多还原