During the past decade the living style of many people has changed quite a lot due to the rapid progresses in multimedia presentation and communications and the blooming of the Internet. As indispensable components of multimedia, image and video are capable of carrying huge amount of information. This makes it absolutely necessary to compress the image and video signals. In this thesis, we will present contributions in three different sub-areas of video and image compression namely, Vector Quantization (VQ), arbitrarily shaped Discrete Cosine Transform (DCT) and transcoder with arbitrarily sampling rate alternate....[ Read more ]

During the past decade the living style of many people has changed quite a lot due to the rapid progresses in multimedia presentation and communications and the blooming of the Internet. As indispensable components of multimedia, image and video are capable of carrying huge amount of information. This makes it absolutely necessary to compress the image and video signals. In this thesis, we will present contributions in three different sub-areas of video and image compression namely, Vector Quantization (VQ), arbitrarily shaped Discrete Cosine Transform (DCT) and transcoder with arbitrarily sampling rate alternate.

Vector quantization is theoretically attractive due to the results from rate-distortion theory. It does yield superior performance over scalar quantization schemes. Despite this, however, a big gap exists between the performance predicted by the theory and that which can actually be achieved. This is due to the failure to observe the two basic assumptions of rate-distortion theory, namely, that the block length can be infinitely large, and the source is stationary. In this thesis, two methods that can amend this big gap are presented. The two methods are dedicated to combating the invalidity of the two assumptions, respectively. A window-based fast search algorithm is proposed to speed up the VQ process.

The MPEG-4 standard provides some key technologies that enable interactivity between the user and video objects. One of the prerequisite conditions is the ability to encode arbitrarily shaped video objects. Computationally complex shape-adaptive DCT algorithms have been proposed in the literature. An invention that can achieve shape-adaptivity via padding and regular base-N DCT (N is a power of 2) is presented in this thesis work. The biggest benefit is that many available DCT chipsets and fast algorithms can be used directly to handle arbitrary shapes. A rate-optimization technique is proposed in order to further improve the coding performance.

More and more portable devices are capable of accessing the Internet. These devices are featured small and irregular size, and low communication bandwidth. A transcoder architecture that can achieve bit rate reduction and arbitrarily resizing at the same time is proposed. The two key steps, namely the resizing and motion estimation techniques that are specially tailored for resizing are discussed.