| Summary: | The aim of this thesis is to develop new image and video coding algorithms which can outperform the existing block transform-based standards at low bit rates. Instead of deep research in a particular approach, several different techniques are considered in order to be able to forecast the future in image and video compression. First, various block transform methods are examined, and their strengths and weaknesses are assessed. Then, classical methods such as linear prediction and vector quantisation are considered. Different linear prediction models and memoryless vector quantisation types are compared. The analysis-by-synthesis approach which has been successfully applied in speech coding, is used to design a linear predictive gain-shape vector quantiser. Since the main weakness of the block transform methods in low bit rates is the blocking effect which is visible at block boundaries, subband approach is considered as an alternative. Several aspects of filter design, region-of-support extension and decomposition structure are examined. Wavelet approach is also considered, however its application in digital image compression is not very different than the subband decomposition. A novel subband image and video codec which exploits the edge orientations in the lowest band in the adaptive vector quantisation of the higher bands is described. Moreover, another novel video codec based on predictive entropy coding of the gain-shape vector quantisation parameters is proposed. More importance is given to the promising segmentation-based techniques. Segmentation-based image coding is explained and several new techniques for contour representation, contour smoothing, edge profile smoothing and jagged edge rectification which improve the coding performance are applied. Then, a novel video coding algorithm based on joint region and motion segmentation is described. A number of control points, the locations of which can be predicted in video signals, are used to represent the region contours and another novel segmentation-based video codec using contour and texture prediction and working at very low bit rates is presented. Finally, the future of image and video coding is discussed and several research directions are recommended.
|
|---|
