Novel motion estimators for video compression.

Abstract

In this thesis, the problem of motion estimation is addressed from two perspectives, namely, hardware architecture and reduced complexity algorithms in the spatial and transform domains. First, a VLSI architecture which implements the full search block matching algorithm in real time is presented. The interblock dependency is exploited and hence the architecture can meet the real time requirement in various applications. Most importantly, the architecture is simple, modular and cascadable. Hence the proposed architecture is easily implementable in VLSI as a codec. The spatial domain algorithm consists of a layered structure and alleviates the local optimum problem. Most importantly, it employs a simple matching criterion, namely, a modified pixel difference classification (MPDC) and hence results in a reduced computational complexity. In addition, the algorithm is compatible with the recently proposed MPEG-1 video compression standard. Simulation results indicate that the proposed algorithm provides a comparable performance (compared to the algorithms reported in the literature) at a significantly reduced computational complexity. In addition, the hardware implementation of the proposed algorithm is very simple because of the binary operations used in the matching criteria. Finally, we present a wavelet transform based fast multiresolution motion estimation (FMRME) scheme. Here, the wavelet transform is used to exploit both the spatial and temporal redundancies resulting in an efficient coder. In FMRME, the correlations among the orientation subimages of the wavelet pyramid structure are exploited resulting in an efficient motion estimation process. In addition, this significantly reduces side information for motion vectors which corresponds to significant improvements in coding performance of the FMRME based wavelet coder for video compression. Simulation results demonstrate the superior coding performance of the FMRME based wavelet transform coder. (Abstract shortened by UMI.)