In this thesis, we investigate two topics: (1) fast global motion estimation and (2) color interpolation for single-CCD digital cameras....[ Read more ]

In this thesis, we investigate two topics: (1) fast global motion estimation and (2) color interpolation for single-CCD digital cameras.

In video compression, global motion compensation improves the coding efficiency by reducing the bits of encoding the motion vectors and by constructing a more accurate prediction image for texture coding. However, its computational complexity is very high. Fast algorithms are thus needed. In this thesis, we propose two novel fast global motion estimation algorithms. The first one is based on motion vector prediction and diamond search. The other one applies prediction ideas to modify the current global motion estimation algorithm in the MPEG-4 VM coder. Simulation results suggest that the proposed method can reduce computation considerably without perceptual degradation when compared with the MPEG-4 VM coder.

Our second topic is color interpolation. Color images captured by a single-CCD digital camera are spatially sampled by a color filter array (CFA). Consequently, each pixel only contains a single color value such as Red (R), Green (G) or Blue (B). Color interpolation is a technique used to generate full resolution color details from the sub-sampled images. In this thesis, we propose a new color interpolation framework for single-CCD cameras. We achieve better quality color images by interpolating in an alternative color domain with smoother edge transitions. Simulation results suggest that the proposed framework produces better quality color images compared to existing algorithms.