Efficient design of Local Binary Pattern for image retrieval

Abstract

A novel image description method named Half-LBP has been proposed in this paper for image retrieval applications. Half-LBP transforms the relationship among the intensity levels of image local pixel blocks into a binary pattern in such a way that reduces the dimension of the descriptor significantly while keeping the retrieval accuracy comparable with the state of art methods. The basic operation is quite similar to the texture operator named Local Binary Pattern (LBP). However, the dimension of the LBP method and its variants are too high and take substantial amount of time to retrieve the images. To address this issue, Half-LBP image descriptor has been proposed which takes comparatively less time to find out the matched images from the database.