Two-dimensional frequency domain motion compensation for spatiotemporal signals

Motion compensation has been widely used in video signal compression to reduce the redundancy among successive frames. A two-dimensional (2-D), frequency domain (F-D), motion compensation technique is developed which employs adaptive system modeling and exploits both spatial and temporal redundancies. Here, motion competition is accomplished by using two optimal predictors designed in the 2-D F-D. The predictors are characterized by their coefficient which are adaptively updated for every frame. Two implementation schemes, forward and backward, are proposed and successfully applied to both video signal and the 2-D wavelet transform coefficients of the video signals. The proposed 2-D F-D linear prediction approach is computationally comparable, but has several important advantages over existing spatial block matching methods such as improving the compensation efficiency by controlling the predictor order. Also, the technique presented is applicable for motion compensation under more general conditions, namely, spatial shift, rotation and scaling. For illustration, simulation results are given to show the performance of the proposed technique.