Quality assessment of stereoscopic (3D) multimedia

Abstract

Reliable quality assessment of stereoscopic 3D (S3D) images is essential for immersive media applications such as virtual reality, 3D video, and depth-based visualization systems. This thesis presents a no-reference image quality assessment (NR-IQA) framework that operates without access to ground truth reference images and captures human-like perception of stereoscopic content. The work begins with the generation of cyclopean images by simulating binocular fusion of stereo pairs. A novel tensor-based approach is used to compute pixel-level disparity maps, where gradients and chrominance-depth differences across the HSV channels are processed to extract dominant structural cues using eigen decomposition. These disparity maps are used to align stereo views and synthesize perceptually unified cyclopean images for quality analysis.