Blind Quality Assessment for High Dynamic Range Video Systems

Abstract

Compared with traditional imaging methods, high dynamic range (HDR) imaging techniques can provide the visual contents with wider dynamic range, higher contrast, and richer detail information. In HDR video system, distortion is inevitably produced during capturing, processing and coding HDR video, leading to the degradation of visual quality. Therefore, how to objectively assess the quality of HDR video is an important problem to be solved. This paper proposes a new blind HDR video quality assessment method based on luminance partition and motion perception, considering the perceptual impacts of texture and motion information in HDR video. Firstly, the distorted HDR video is divided into groups of frames (GoFs), and tensor decomposition is performed on each GoF to obtain its static information map containing the main spatial information and the corresponding motion information maps containing the main temporal information of HDR video. Considering the different visual perception to texture details of the bright, middle and dark regions in the distorted HDR video, the luminance partition is performed on the GoF's static information map to obtain the brightness, middle and dark regions, and the texture features are extracted from the different luminance partitioned regions to form the global and local texture perception feature sets. Then, considering the temporal perception, the motion information map is used to extract motion perception features and structure perception features. The extracted features are fused in the temporal to form the final feature set used to predict the quality of distorted HDR videos. Moreover, a subjective database of HDR videos is established to verify the effectiveness of the proposed method. Experimental results show that the proposed method can evaluate the HDR video quality accurately, and has good consistency with human visual perception.