Omnidirectional Image Quality Assessment With a Superpixel-Based Sparse Model

Abstract

Panoramic interaction represented by omnidirectional images and virtual reality will be an important format of information technology in the future. How to evaluate the quality of omnidirectional images accurately and quickly is essential for the user experience of panoramic interaction. In this paper, we propose a novel superpixel-based sparse model for full reference omnidirectional image quality assessment. First, we segment the omnidirectional images into superpixel regions. The Entropy of Primitives (EoP) is then calculated as image information based on the sparse model. Furthermore, the Kullback-Leibler divergence is exploited to represent the difference of visual information between original and distorted images. The quality score is predicted by an SVR model trained from the visual information features. Experimental results show that the proposed metric achieves high consistency with the subjective evaluation on the OIQA database.