Binocular JND model based stereo image coding

Chuan Chang, Jui-Chiu Chiang, Wen-Nung Lie, Tzu-Chan Wu

Abstract: 3D image/video applications attract lots of attention these days. Due to increased data amount compared to 2D image/video, many researches try to well compress 3D data by removing the inter-view redundancy between different views. In general, only interpixel redundancy is exploited, not psychovisual redundancy. Currently, binocular just noticeable difference (BJND) model is proposed to measure the perceptible distortion of binocular vision for stereoscopic image. In this paper, in addition to the inter-view redundancy, the psychovisual redundancy is also removed by taking the binocular visual characteristics into consideration during the encoding of stereo images. Experimental results show that the proposed coding scheme achieves up to 20% bitrate reduction, compared to the conventional stereo image coding scheme, while providing comparable stereoscopic visual perception.