Stereoscopic Image Coding Using a Global Disparity Estimation Algorithm Optimizing the Compensation Scheme Impact

This paper focuses on the disparity-compensated stereoscopic image coding. This coding scheme is the most implemented technique taking advantage of the typical stereoscopic image redundancy. Namely it first predicts one view using the other view and a disparity map. A compensated view is then computed by coding the remaining difference between the view and its prediction. This paper concerns the computation of the disparity map combining two existing techniques. The first one, more efficient at low bitrate, is an iterative search reducing the bit cost of losslessly storing the disparity map at the expense of a small increase in distortion. The second one, more efficient at high bitrate, selects disparities minimizing the distortions of the compensated view, assuming that JPEG is used in the compensation. This combination appears to be fruitful, achieving increased performance at low- and mid-range bitrate, when tested on a few stereoscopic images and compared to these two techniques.