Distortion prediction for motion-compensated lifted haar wavelet transform and its application to rate allocation

In this paper we present a solution for low complexity rate allocation for video compression using motion-compensated lifted wavelet transform. We assume that after temporal decomposition a standard still image coding scheme like JPEG2000 is used to encode the resulting temporal bandpass frames. We show that the reconstruction distortion of a group of pictures can be accurately predicted from the distortion introduced into the temporally decomposed frames despite the non-orthonormality of the transform. We model the operational rate-distortion function of the still image coding using a simple parametric model and solve the rate allocation problem by minimizing a Lagrangian cost Junction. We compare our rate-allocation approach with a heuristic scheme and observe significant improvements in terms of rate-distortion performance at dramatically reduced computational complexity. The gains are particularly significant if the motion field is represented with sub-pel accuracy.