In this dissertation, a new block-based motion estimation (ME) method is proposed which uses the Kalman filtering (KF) with adaptive block partitioning (ABP) to improve the motion estimates resulting from the conventional block-matching algorithms (BMAs). In our work, a first-order autoregressive (AR) model is applied to the motion vectors (MVs) obtained by BMAs. A new approach is developed for adaptively adjusting the state parameters of the Kalman filter and the motion correlations between neighboring blocks are referred to predict motion information. According to the statistics of the frame MVs, 16x16 macro-blocks (MBs) are split into 8x8 blocks or 4x4 sub-blocks adaptively for fine grain operation of the Kalman filtering. To further improve the performance of MV prediction, we adopt a zigzag scanning of blocks or sub-blocks and the state parameters of the Kalman filter are updated successively during each iteration in accordance with the outcome of the zigzag based block or sub-block scanning. The experimental results indicate that the proposed method can effectively improve the ME performance in terms of the peak-signal-to-noise-ratio (PSNR) of the motion compensated images with smoother motion vector fields as compared to the existing approaches in the literature. The scheme described herein is also tested on high resolution video samples yielding least motion artifacts in the reconstructed image frames. Such robust performance makes it an ideal temporal redundancy extraction engine for a wide variety of video transmissions and new digital TV applications.
From micro to nano-scale medium development point of view, however, the block-based KF motion prediction is not the most cost effective and fastest approach for mobile video communications and computing devices including the third (G3) and fourth (G4) generation technology standards. To this end, as a second part of the research, we focused on developing a fast binary partition tree based variable size video coding system. New adaptive algorithms proposed herein are applied to a video encoder with binary partition trees. First, to reduce the computation for block-matching, an adaptive search area method is described which adjusts the searching region according to the size of each block. Second, an early termination method is introduced which terminates the binary partitioning process adaptively according to the statistics of the peak-signal-to-noise-ratio values during each step of block splitting. Third, we put forward a new model for fast rate-distortion (R-D) estimation to decrease the computation of matching pursuit (MP) coding for residual images. Simulation results show that the proposed techniques provide significant gain in computation speed with little or no sacrifice of R-D performance, when compared with non-adaptive binary partitioning scheme.