Texture Classification Based On Directional Local Binary Pattern Approach

Texture analysis attempts to quantify qualities such as rough, smooth, silky, or bumpy as a function of the spatial variation in pixel intensities. In this regard, LBP operator acts as one of the best local texture descriptors which is used in texture classification. In spite of all its characteristics, using the directional features information has not gained much attention in which it provides more more accurate information for texture classification. So in this paper, directional LBP based on Gaussian filtering approach is studied and summarized. Firstly the Gaussian filtering and traditional LBP operator was implemented on images in order to extract sign information, secondly all neighboring pixels of Gaussian filtering arranged along predefined directions and thirdly the local differences on each direction were calculated to extract directional information. Finally all extracted features concatenated together to form a final histogram for the classification process by using nearest neighbor classifier. The performance of the method was evaluated through comparing with some existing state-of-the-art LBP algorithms on Brodatz database, and the results demonstrates that the new DLBPG descriptor is more extensive, leading to a superior performance.