On performance gains in MPEG-2 video coding via a rate-distortion optimal route

The freedom offered by the MPEG-2 video compression standard is exploited to improve its coding performance by means of a rate-distortion optimal framework using a Lagrange formalism. Optimization is performed on a picture-by-picture basis, with either a bit-rate or a distortion target, and concerns a combination of quantization, prediction modes and motion vectors. Results from the individual optimization of each of these features have been reported in the literature. However, the combined optimization of all features together is new. The primary goal is the investigation of the achievable coding gains, so that bounds on the performance of MPEG-2 can be established. The gain in bit-rate may exceed 20%. The improvement in average distortion ranges between 0.5 and 2 dB, with peaks of 3.5 dB.