Center-Emphasized Gradient-Feature Based Quality Assessment Method for Screen Content Images

Screen content images (SCI) are composed of texts, graphics, images, emails, web pages, and other computergenerated components which make it different from the pure natural images. As a result, image quality assessment (IQA) methods designed for natural images are not always suitable for screen content images. However, as time goes, many screen contents are being generated which share several properties of natural images. As a result, IQA methods for SCI need to be more generic while giving priority to the prominent SCI features. The image gradient is a very important feature for all kinds of images, especially for SCI. In this paper, we proposed a novel image gradient based similarity index (CGSI) for screen content images where the gradient is used both as a quality map and extractor for another feature map. Our previous studies show that the center part of an image is visually the most important part and HVS is more sensitive to any distortion in the middle area. To address this issue, the center area of both gradient and feature similarity maps are raised by element-wise squaring. Eventually, the final quality score is calculated using a summation of standard deviation based pooling strategy. We evaluated our proposed method on three large scale SCI databases and compared with 8 other state-of-the-art IQA methods designed for both natural and screen content images. Results show that the proposed approach provides competitive performance and defeats all other methods with a large gap in overall performance. Also, the relatively faster running time makes it suitable for most of the real-time applications.