Perceptual Adaptive Quantization of Stereoscopic Video Coding Using Mpeg-2's Temporal Scalability Structure

Signi cant perceptual improvements can be gained in the coding of stereoscopic videos, as is illustrated here by modifying the adaptive quantization of an MPEG-2 codec. Applying human visual properties to MPEG-2's temporal scalability framework, a perceptual adaptive quantization approach to stereoscopic video coding is presented. To optimize the perceived picture quality of the reconstructed stereo images, binocular visibility of stereo coding artifacts is investigated, including image fusion and visual masking. Four normalized indicators are introduced to account for 3D visual artifacts, and are incorporated to determine the quantization parameters. They are (1) prediction accuracy, (2) prediction correlation, (3) fusion indicator, and (4) texture intensity. Simulations indicate the importance of perceptual stereo coding, with improvements in overall stereo quality and reductions in binocular artifacts.