Image representation using complete multi-texton histogram

doi:10.1007/s11042-019-08350-1

Image representation using complete multi-texton histogram

2020-01-04

13

PRE

AI

Kherfi Mohammed Lamine

DOI：10.1007/s11042-019-08350-1

Original

Original request for help

Save

Abstract

En 中文

Texture is a fundamental aspect which is often used to represent the visual content of images. Psychologists stated that color and texture have a close relationship via fundamental micro-structures called textons. Textons are considered as atoms for pre-attentive human visual perception. Based on texton theory, many literature works have tried to develop efficient models for image recognition. In this paper, we put forward a texton-based method, called Complete Multi-Texton Histogram (CMTH), that has the ability to discriminate both texture and non-texture color images. CMTH incorporates information about the color, edge orientation and texton distribution within the image. The proposed CMTH has been extensively examined on five publicly available datasets. Three of these datasets were intended to evaluate texture discrimination, namely: Vistex, Outex, and Batik whereas the two others were intended to evaluate heterogeneous image discrimination, namely: Corel10K and UKBench. The proposed method has been evaluated via image classification and retrieval tasks. The obtained results have shown that our proposed descriptor significantly outperforms the state of the art methods in both classification and retrieval.

Keywords:

Texton theory

Texture perception

Image representation

Features extraction

Image representation using complete multi-texton histogram

Abstract

AI Summary

Journal

Organization