Lossless and nearly lossless digital video coding

In lossless coding, compresssion and decompression of source data result in the exact recovery of the individual elements of the original source data. Lossless image / video coding is necessary in applications where no loss of pixel values is tolerable. Examples are medical imaging, remote sensing, in image/video archives and studio applications where tandem- and trans-coding are used in editing, which can lead to accumulating errors. Nearly-lossless coding is used in applications where a small error, defined as a maximum error or as a root mean square (rms) error, is tolerable. In lossless embedded coding, a losslessly coded bit stream can be decoded at any bit rate lower than the lossless bit rate. In this thesis, research on embedded lossless video coding based on a motion compensated framework, similar to that of MPEG-2, is presented. Transforms that map integers into integers and embedded source coding, which are the main ingredients of lossless embedded coding are discussed in the context of intra frames, which are similar to still images and non-intra frames, which contain motion compensated prediction errors. The lifting concept, which forms the integer wavelet transforms, and the intrinsic properties of the block orthogonal transforms, such as the Discrete Cosine (DCT the Discrete Sine (DST) and the Walsh-Hadamard (WHT) are used to design the integer versions of the N-point DCT, DST and WHT, where N is any integer power of two. Furthermore, the design and the use of transforms with spatially adaptive numbers of vanishing / preserving moments, which are suitable for non-intra frames, and non-linear transforms are presented. The current and prospective embedded coding scannning methods are analysed and an adaptive quad tree splitting (AQS) based scanner is presented. The performance of the above transforms for both types of frames is analysed using the zeroth order entropy values and the coded bit rates, achieved by Embedded Lossless Image Coding (ELIC), which is based on AQS and ecient context modelling. In addition to the above experiments, the use of a transform in coding highly decorrelated non-intra frames is also investigated. Finally, the components discussed above are intregated together to analyse the importance of motion compensation in lossless video coding and the robustness of embedded decoding at quasi-lossless decoding in an assymetric codec, where a Group Of Picture (GOP) structure based motion prediction is involved.