Screen content image quality measurement based on multiple features

With the increasing use of multimedia devices, the demand for Screen Content Images (SCIs) has surged. However, the transmission process inevitably leads to visual degradation of image quality. Effective measurement of the quality of SCIs is therefore an urgent task. In this paper, we propose an image quality measurement method based on multiple features. According to the content characteristics of SCIs, we extract multiple features in terms of both structure and color. As SCIs contain a large amount of text and graphics, we calculate different gradient-weighted local ternary pattern histograms on the gradient domain to capture the structural degradation of the image from various aspects. Then, considering color as another crucial visual factor, we extract contrast energy and saturation from the opponent color space, and design parameter models that can accurately characterize the color information of SCIs. Finally, we use the Adaboost-BP neural network to train the quality measurement model. Experimental comparisons on three public SCIs databases (SIQAD, SCID, QACS) demonstrate that the proposed method is more in line with human perception compared to other state-of-the-art quality metrics. In addition, we demonstrate in the experiment that the proposed method can be a better alternative to the Peak Signal-to-Noise Ratio (PSNR) to assess the visual quality of watermarked SCIs in practice applications.