Content-based image retrieval using color and a novel texture descriptor: orientational-based local binary pattern

The increasing use of images and multimedia has led, in the past years, to a major growth of data and images to store. With this, comes the importance of having a way to facilitate the indexing and the retrieval of the images. In this article, we present a new method for content-based image retrieval. The proposed approach is composed of two phases: first, a preselection of relevant images from the initial database based on HSV color space and using color moment information. The second phase relies on using the selected images from the first phase as a new database and extracting the texture feature using a novel descriptor called orientational-based local binary pattern (OB-LBP) derived from the traditional LBP texture feature technique. In the proposed descriptor (OB-LBP) we compare each neighbor of the center pixel (of 3 x 3 window) with its three closest neighbors and form a binary code of three components for each pixel, then based on their directions we generate three images and produce the histogram to extract the final feature descriptor. Corel-10 k and Wang's databases were used to evaluate the proposed method. The results show a significant improvement in the proposed method over existing methods.