Screen content image quality assessment using distortion-based directional edge and gradient similarity maps

With the prevalent use of applications like Facebook, Twitter, and remote control applications, assessing the quality of the screen content images (SCIs) has become one of the critical fields of researches in image processing. In this paper, we develop a novel full-reference image quality assessment (IQA) method, called distortion-based directional edge and gradient similarity maps (DDEGSM) method, to evaluate the quality of SCIs by efficiently incorporating the effect of two challenging distortion types: contrast change (CC) and color saturation change (CSC). The proposed DDEGSM method has four main steps. First, we form an edge similarity map by extracting two edge and gradient features using the image’s gradient in 12 directions and the Laplacian of Gaussian filter. Next, we use the difference of Gaussian filter to weigh the strength of each pixel in the edge similarity map. Then, we examine the distorted image using two efficient approaches to find out whether the image contains the CC or CSC distortion. Finally, we calculate the quality score of the input SCI from the distortion-based feature maps using a pooling strategy. Our extensive experiments on three commonly-used SCI datasets indicate the proposed method is superior to the state-of-the-art full-reference IQA methods and is more consistent with the subjective assessments.