胶囊内窥图像出血检测中颜色向量相似系数分类器的设计

Design of the Color Vector Similarity Coefficient Classifier for Bleeding Detection in Wireless Capsule Endoscopy Images

为了提高胶囊内窥图像出血智能识别的识别率,将颜色表示与向量运算相结合,提出了使用颜色向量相似系数度量颜色相似性的新方法,推导了颜色向量相似性系数的定义式.在此基础上,设计了应用于RGB颜色空间的颜色向量相似系数分类器,并结合种子区域生长算法实现了内窥图像出血智能识别的新算法.通过实验验证,该算法的出血检测灵敏度和特异度分别达97%和90%,基本实现了胶囊内窥图像出血智能识别.

In order to improve the recognition rate of the bleeding intelligent recognition, the new concept of color vector similarity coefficient was brought forth to measure the color similarity, the calculation formula of the similarity coefficient was also derived. Based on this similarity coefficient the classifier was designed which can be applied in the RGB color space; combining the classifier with the algorithm of seededregion-grow the new algorithm of wireless capsule endoscopy（WCE） image bleeding intelligent recognition was implemented. The experiments show that the sensitivity and the specificity of this algorithm reach 97%, 90% respectively. The WCE image bleeding intelligent recognition is basically realized and will be applied in the WCE images detection to help the clinician.