A motion-compensated prediction is considered as a key step in the coding of moving image sequence. Yet it is mainly to reduce the temporal redundancy and further efforts are followed to reduce remained redundancy, mostly spatial redundancy, through various intraframe prediction or transform algorithms. However for pel recursive motion compensation, the spatial correlation, i.e. the redundancy of the displaced frame difference, would not be large enough to give any performance gain for intraframe prediction.
Since it is noticed that the spatial gradient frames(2-D. spatial prediction error frames) nearly keep the temporal redundancy, i.e. the motion information, this thesis proposes a motion compensated(PRA based) temporal prediction applied to the gradient frames.
This temporal prediction after spatial is shown to have superior performance to the spatial prediction after temporal. The performance improvement is more than 4dB in SNR, ratio of moving area prediction error power to input signal peak power, and also more than 0.5bit/pel in entropy of moving area prediction error.
