Local spatial binary pattern: a new feature descriptor for content-based image retrieval

Abstract

In this paper, we propose a novel image retrieval algorithm using local spatial binary patterns (LSBP) for contentbased image retrieval. The traditional local binary pattern (LBP) encodes the relationship between the referenced pixel and its surrounding neighbors by calculating gray-level difference, but LBP lacks the spatial distribution information of texture direction. The proposed method encodes spatial relationship of the referenced pixel and its neighbors, based on the gray-level variation patterns of the horizontal, vertical and oblique directions. Additionally, variation between center pixel and its surrounding neighbors is calculated to reflect the magnitude information of the whole image. We compare our method with LBP, uniform LBP (ULBP), completed LBP (CLBP), local ternary pattern (LTP) and local tetra patterns (LTrP) based on three benchmark image databases including, Brodatz texture database(DB1), Corel database(DB2), and MIT VisTex database(DB3). Experiment analysis shows that the proposed method improves the retrieval results from 70.49%/41.30% to 73.26%/46.26% in terms of average precision/average recall on database DB2, from 79.02% to 85.92% and 82.14% to 90.88% in terms of average precision on databases DB1 and DB3, respectively, as compared with the traditional LBP.