No-Reference Stereoscopic Image Quality Assessment Based on Visual Attention and Perception

Abstract

In recent years, the methods of no-reference stereoscopic image quality assessment (NR-SIQA) have been well investigated, but there still remain challenges due to the inaccurate extraction of binocular perception information. In this paper, we propose an NR-SIQA method based on visual attention and perception. We combine saliency and just noticeable difference (JND) to model visual attention and perception, respectively, and weight the global and local features extracted from the left and right views. Meanwhile, in order to obtain the accurate binocular perception information, the global structural features reflecting spatial correlation are extracted from the cyclopean map that is synthesized by the left and right views. Then, a regression model is learned based on a support vector machine regression (SVR) to evaluate the quality of stereoscopic images. The experiments on popular SIQA datasets demonstrate that the proposed NR-SIQA method has better and more reliable performance than the state-of-the-art methods.

Keywords

• Stereoscopic image quality assessment
• saliency model
• JND model
• cyclopean map