No-reference light field image quality assessment based on depth, structural and angular information

Abstract

Light field imaging can capture more abundant scene information, including angular and spatial information, compared with traditional imaging technologies. However, light field image (LFI) processing will inevitably introduce distortion which is different from traditional imaging process. Therefore, LFI quality assessment has become one of important research issues on LFI processing. In this paper, we propose a depth, structural and angular information based no-reference LFI quality assessment method. Firstly, the horizontal and vertical mean difference images (MDIs) are defined to integrate the difference images of LFI's sub-aperture images (SAIs) and reflect the horizontal and vertical depth and structural information of LFI. Considering the multi-channel characteristics of human visual system, an effective feature extraction scheme with Curvelet decomposition is designed for MDIs and SAIs of the distorted LFI. As one of the representations of LFI, epipolar plane images (EPIs) contain LFI's angular and depth information, which reflect the angular consistency of LFI more intuitively. Therefore, an algorithm, namely local maximum similarity index statistics, is designed to extract the directional features from horizontal and vertical EPIs. Finally, the quality of distorted LFI is predicted by pooling these extracted features. Experimental results on three LFI datasets show that the proposed method can achieve better performance, compared with the representative traditional image quality assessment methods as well as the state-of-the-art LFI quality assessment methods.