WCE video abstracting based on novel color and texture features

Wireless Capsule Endoscopy (WCE) is a novel imaging technique for the investigation of the entire small bowel. It produces over 50,000 images per examination. It is a highly time-consuming and labor tedious task to review such a large amount of images. Recent research has focused on developing computer aided diagnosis (CAD) systems to reduce the heavy burden of physicians. In this paper, we propose a novel strategy to abstract WCE videos based on abnormality instead of topography. Our scheme utilizes multiple novel features and linear discriminant analysis (LDA) to extract the most representative images in each suspected video clip. Experiments are performed on the current data sets to evaluate our method. The results demonstrate that our strategy achieves significant reduction of the number of frames without losing critical information.