Transform coding and multiple description image coding

With the expansion of world wide web and increasing remote applications, reliable and fast transmission of signals and images continue to be major areas of research. Coding is an effective tool that can improve reliability as well as the speed of transmission of signals and images. This dissertation presented two techniques to improve the performance of popular wavelet transform coding methods and a new scheme of multiple description coding that improves coding efficiency.;First, an adaptive wavelet coding of images was proposed where different wavelets for the various sub-bands were chosen using an adaptive strategy. It was known that no one wavelet gives the best performance for all images. In this dissertation, it was shown that even for a single image no one wavelet provides optimal performance for all sub-bands. Hence an adaptive selection of wavelets for each sub-band was proposed for wavelet coding of images. Experiments showed that the performance of the adaptive scheme was better than the single transform coding.;Second, new transforms were generated by applying singular value decomposition to the matrices of a bi-orthogonal wavelet transform and grouping the resulting matrices in a way which was different from the way they were decomposed. It was shown that the resulting transform had a better energy compacting property than the original wavelet transform.;In the third and final contribution, this dissertation presented a new multiple description coding technique in which multiple descriptions were generated using different transforms and a linear fusion was employed to reconstruct the central description. A technique was also presented to further reduce the distortion of the central description by introducing forced errors to side descriptions. These two techniques were applied to various data such as memoryless gaussian signals and natural images. The example multiple description image coders designed using these techniques provided much higher peak signal to noise ratios than the current coders.