A non-parametric method for fast joint rate-distortion optimization of motion estimation and DFD coding

Exact joint rate-distortion (RD) optimization of motion estimation and displaced-frame-difference (DFD) signal coding is highly computationally complex. We propose a low-complexity but effective search scheme. Major complexity reduction is achieved by: (1) eliminating suboptimal displacement vector (DV) candidates from joint optimization, and (2) using a dynamic programming algorithm to optimize DFD coding. DV candidate elimination is based on predicting the order of the RD-Lagrange costs. Order prediction is found to be much more reliable than using parametric functions to predict the exact cost values. In comparison with an H.263 test model, with advanced prediction mode enabled, the proposed scheme furnishes up to about 1 dB of PSNR improvement