Content-based image retrieval using local texture-based color histogram

This paper presents a novel image feature representation method, called local texture-based color histogram (LTCH), for content-based image retrieval. The LTCH can describe the color distribution under a mask, which is defined as a micro-structure image with a near-uniform texture. The near-uniform texture is exacted by center symmetric local trinary pattern (CS-LTP) and micro-structure map. The CS-LTP is coding on a quantized HSV image, and the micro-structure map is defined with the same as CS-LTP code. The LTCH can be considered as a novel visual attribute descriptor combining local texture, color and spatial layout, without any image segmentation and model training. The proposed LTCH method is evaluated on Corel-1000 database and Corel-5000 database with the standard performance evaluation method, for image retrieval. The experimental results demonstrate that the proposed method has a better performance than representative image feature descriptors, such as color difference histogram (CDH), microstructure descriptor (MSD), multi-texton histogram (MTH) and structure elements´ descriptor (SED).