Unsupervised redundant image deletion for wireless capsule endoscopy examination
胶囊内窥镜冗余图像数据自动筛除方法

This paper proposed an unsupervised algorithm to delete the redundant WCE images, which was based on the analysis of the normalized mutual information and normalized cross-correlation coefficient between the successive frames. The algorithm firstly conducted quantification and clustering in HSV color space. Then, it calculated the similarity metrics between the successive frames. Finally, it iteratively applied deletion procedure according to the prescribed deletion rate. The pathology retaining rate, which was defined as the percentage of the remaining images bearing pathological changes from the total ones was almost 100% with very low mis-deletion rate for 70% prescribed deletion rate of 49 patients. Experimental results show that the method based on the analysis of the normalized mutual information is effective to delete redundancy images and greatly reduces diagnosis time.