一种用于胶囊内窥图像的小肠出血智能识别算法

A Novel Method of Small Bowel Bleeding Intelligent Recognition in Capsule Endoscope Image

针对胶囊内窥图像的出血检测提出了一种新的智能识别算法。该算法首先将彩色内窥图像在RGB空间进行分水岭分割，再将图像转换至CIEluv彩色空间，在前分水岭分割的基础上，在CIEluv空间上对各分块区域依据欧氏距离判别彩色相似度，进行小肠出血的识别与出血区域提取的工作。实验结果表明该算法灵敏度为94．3％，特异度为98．4％。

In this paper, a new intelligent recognition method is proposed for bleeding recognition in capsule endoscope image. First, watershed segmentation in RGB color space is applied to the original capsule endoscope color image. Then the original image is transformed into CIEtuv color space, last small bowel bleeding recognition and extraction of bleeding region are accomplished by color similarity measurement according to Euler distance in CIEluv color space. Experiments show that the sensitivity is 94.3% and the specificity is 98.4%.