Image sequence coding using Motion Embedded 3D Vector Quantization

This paper presents a new strategy to code image sequences using Motion Embedded 3D Vector Quantization (ME3DVQ). The proposed algorithm first generates a 3D uniform grid over a set of consecutive frames for the formation of non-overlapping approximated cubes using a 3D spatiotemporal interpolation. The corresponding residual (original - approximated) cubes are then coded using VQ. The VQ coder is designed to operate in two modes, ZDVQ and 3DVQ, according to the existence of temporal activities (TA) (motions) within the residual cubes. It operates in the ZDVQ mode with simple block duplications if the residual cube contains no TA, otherwise it switches to the 3DVQ mode. This 3DVQ mode is capable of coding the residual cubes with embedded motion information without the need to perform motion compensation. Simulation results show that the reconstructed sequence has an average PSNR of 30.7 dB with a good fidelity, at an average bit rate of 0.126 bits per pixel (bpp).