Ellipse based local binary pattern for color image retrieval

A novel region-based method for color scene images characterization is proposed. The method exploits the texture features present in grayscale images extracted by LBP. It superposes a set of ellipses which are generated by rotation and contraction and organized for covering the image, where each ellipse delimits a region of the LBP map. One LBP-histogram is calculated for each region. Finally, texture features present in the images are represented by m LBP-histograms, where m is the number of ellipses (regions) used. Each elliptic region in the query image is then matched with the most similar region of its class in the target image. A global distance between the query and the target images is afterward derived. The method has been compared to published works, including LBP and its improved version ILBP, using the Wang dataset. Results show significant higher or comparable performance of our method with regard to the other methods.