Saliency Aware Locality Constrained Linear Coding for WCE Multi Abnormality Detection based on SURF

Abstract

Wireless Capsule Endoscopy (WCE) has become a widely used diagnostic technique for the gastrointestinal tract at the cost of a huge quantity of data that wants to be analysed. To find a solution to this problem a new computer aided system using novel features is proposed in this paper to automatically detect the multi abnormalities in GI tract from WCE videos. In the feature learning stage to obtain the representative visual words, the training images of polyp, ulcer, bleeding, and normal images are sampled and represented by Speeded up Robust Features (SURF) descriptor, and is constructed by K-means clustering algorithm. These four types of visual words are combined to composite the representative visual words for classifying the WCE images. In the feature coding stage we propose a locality constrained linear coding (LLC) algorithm to encode the images. Moreover, it includes the patch saliency constrained on the feature coding stage to highlight the important information in the images. LLC uses the locality constraints to project each descriptor into its local-coordinate system, and the projected coordinates are concatenated by max pooling to create the final representation, and classified using SVM The experimental results exhibit a higher accuracy, sensitivity, specificity and lower processing time, validates the acceptance of the proposed method.