Extended Gaussian-Filtered Local Binary Patterns for Colonoscopy Image Classification

Local Binary Patterns (LBP) and its variants are widely used for texture classification. In this paper we propose a new variant of LBP descriptor called the extended Gaussian filtered Local Binary Patterns (GF-LBP) which is robust to illumination changes, noise and captures more informative edge-like features for classification. Experiments on a colonoscopy image dataset with 2100 images for binary (`normal´ or `abnormal´) classification show that the proposed xGF-LBP descriptor significantly outperforms the standard LBP descriptor and its considered variants.