Blind light field image quality assessment by analyzing angular-spatial characteristics

Abstract

Light Field Image (LFI) can simultaneously record the intensity and direction information of light rays, and provide users with strong immersion experience. However, heterogeneous artifacts may be introduced during LFI processing, which results in degradation of the perceptual quality of LFI. To evaluate the LFI quality effectively, a novel blind light field image quality assessment method by analyzing angular-spatial characteristics is proposed. The main strategy is to quantify the LFI quality degradation by evaluating the angular consistency and spatial quality simultaneously. Firstly, the multi-directional intra-block and inter-block differential operations are employed on macro-pixels to generate Macro-Pixel Residual Blocks (MPRBs) on hue, saturation and luminance of LFI. Secondly, effective perceptual feature extraction schemes based on MPRBs entropy distribution and natural scene statistics of discrete cosine transform coefficients for MPRBs are developed to measure the angular consistency on each color descriptor. Thirdly, Macro-Pixel Variance (MPV) map is defined, and the quality-aware features are extracted from MPV map to measure the occlusion areas of LFI. Fourthly, the perceptual features are extracted from sub-aperture images to comprehensively measure the spatial quality of LFI. Finally, all the extracted features are pooled to predict the objective quality of LFI. Extensive experimental results on four LFI datasets show that the proposed method significantly outperforms the representative 2D, 3D, multi-view image quality assessment methods as well as the state-of-the-art LFI quality assessment methods, and is more in line with the human visual perception.