No reference quality assessment for screen content images

Abstract

In this paper, we propose a novel no reference visual quality assessment (VQA) metric for screen content images (SCIs) by luminance features and texture features based on the properties of human vision system (HVS). First, we calculate the luminance map through the local normalization and the statistical luminance features is extracted from the luminance map. Furthermore, we extract the texture features from high order derivatives which can capture the details of image texture, in which we use the sobel filters to compute the gradient map based on the luminance map which is considered the first order derivatives, the texture features are extracted from the second order derivatives calculated based on the gradient map by the local binary patterns (LBPs). Both luminance and texture features are represented in the form of histogram. The support vector regression (SVR) is adopted as the mapping function from the quality-aware features to subjective ratings. Experimental results on the large-scale SCI database show that the proposed method can obtain better performance of visual quality prediction of SCIs than other existing methods and even including some full reference visual quality assessment methods.