Interframe video coding using adaptive sampling and classified vector quantization

The authors propose an adaptive sampling and vector quantization scheme to compress motion compensated differential images (MCDI). MCDI are adaptively sampled based on either their local spectral or geometrical characteristics. Various image decomposition schemes are proposed to design the nonuniform sampling patterns that are used in the sample selection process. The sampling structure can serve as a classifier to distinguish different features of the subimage regions, i.e., uniform, edge, etc. The set of vectors formed by the chosen samples associated with each distinct sampling pattern is used to generate a codebook for vector quantization (VQ), resulting in a classified multicodebook approach to VQ. This adaptive sampling and vector quantization scheme demonstrated superior performance over direct VQ in terms of both coding fidelity and computational efficiency. Simulation results have shown good coding fidelity at low bit rates: below 0.1 bits/sample with signal-to-noise ratio above 30 dB