In this thesis, we investigate two topics for video coding: (1) motion estimation and motion compensation in wavelet domain, and (2) fast global motion estimation for video coding....[ Read more ]

In this thesis, we investigate two topics for video coding: (1) motion estimation and motion compensation in wavelet domain, and (2) fast global motion estimation for video coding.

Wavelet transform is a new tool in signal processing. It can localize a signal in both time and frequency domain and provides a multi-resolution representation of a signal. Wavelet transform has successfully been implemented in the new still image compression standard, JPEG200 [25], as the core technology. Since video is an extension of still image, many researchers try to extend the usage of wavelet transform to video coding. In this thesis, we propose a symmetric padding motion estimation and an overlapping motion compensation in wavelet domain to improve the quality of motion estimation and compensation in wavelet domain. It can be as good as traditional motion estimation and motion compensation in spatial domain.

Global motion estimation (GME) is to estimate the motion of background in the video sequence, which is generally induced by camera's motion. Applying global motion estimation and global motion compensation (GMC) in video coding can create a better quality predicted image in motion compensation, and reduce the amount of bits needed to encode the motion field. However, the main problem of applying GME in video coding is that GME requires a huge amount of computation. In the Verification Model of MPEG-4, the encoding speed with GME enabled is nearly 10 times slower than with GME disabled. Even with the fast GME in the Optimization Model 3 of MPEG-4. the actual encoding speed with GME enabled is still slow. In this thesis, we propose a fast GME called Adaptive PMVFAST Segmentation Global Motion Estimation (APSGME) which performs faster than the existing fast GME in the Optimization Model 3 of MPEG4.