No-reference Image Quality Assessment Based on Regional Information

Abstract

Human vision can quickly assess the quality of images because of the identity in a region and the divisibility between different regions. According to human visual characteristics, we propose a method of image quality assessment without reference information. Firstly, the image was segmented into a series of homogenous regions. Secondly, we analyzed statistical characteristics for each region. Lastly, the assessment model was constructed according to the statistical differences among regions. The results experimented on the LIVE database II show that this method is competitive to SSIM and PSNR.