Backward Drift Estimation with Application to Quality Layer Assignment in H.264/AVC Based Scalable Video Coding

We present an approach for accurate estimation of the reconstruction distortion in SNR scalable video coding with drift. Based on a linear model of predictive video coding, we derive an algorithm to quantify spatio-temporal drift properties subject to prediction structure and motion information. This allows for low-complex estimation of the reconstruction distortion on a per-block basis. The accuracy of the distortion estimation is experimentally verified. We then utilize the method for quality layer assignment within the framework of H.264/AVC scalable video coding (SVC), which is currently under standardization. The quality layers allow for bit stream truncation in a rate-distortion optimized sense. Compared to the quality layer assignment as implemented in the SVC test model, use of backward drift estimation allows for achieving equivalent coding efficiency with reduced complexity.