WCE Polyp Detection Based On Locality-Constrained Linear Coding With A Shared Codebook

Locality-constrained Linear Coding (LLC) has been proven to have good results in image classification. However, the codebook used is simply obtained by K-means without any other distinctive properties which limits classification performance. Our paper targetsWireless Capsule Endoscopy (WCE) images classification by learning a shared codebook in the LLC. The main idea of this work is try to use the part of codebook that represent private features as much as possible. The proposed method rearrange the columns of the codebook that arrange these rarely used atoms in the bottom of the updated codebook which called shared codebook. Then we encode the local features through the shared codebook. Finally, these codes could be trained and classified by the Support Vector Machine (SVM). The experiment results indicate a better performance of proposed method compared with some existing approaches.