| dc.description.abstract | In digital communication, compression is necessary for transmitting huge amounts of information over very narrow channels or for efficient storage of digital information. Visual information plays a key role in modern digital communication services. The amount of information associated with a video sequence is enormous, and so compressing it into a very low bit?rate stream poses a challenging problem. Since in most video applications we are interested only in the subjective quality of the decoded image sequence, lossy compression techniques may be applied to achieve higher compression.
Video compression techniques rely on two principles:
the reduction of inherent spatial and temporal redundancies, and
the exploitation of the properties of the human visual system.
The challenge of video coding lies in efficiently reducing temporal redundancy, and many algorithms have been proposed for this purpose. For real?time transmission of video, the efficiency of these algorithms is measured mainly on the basis of:
(a) compression ratio,
(b) quality of the decoded video, and
(c) encoding/decoding time.
One of the most popular interframe coding techniques is Motion Estimation/Compensation using the Block Matching Algorithm (BMA). To reduce the computational burden associated with the straightforward 2?D full?search algorithm, many fast search algorithms have been developed that are computationally efficient but have associated problems such as incapability to catch small motions, and susceptibility to getting trapped in local minima. Once the motion vector is estimated, the residual error is then coded and transmitted for compensation, and Vector Quantization (VQ) may be employed for this purpose. The performance of VQ depends on the size and quality of the codebook used. For fast encoding, codebook size can be kept small but at the cost of quality. Hence, Adaptive Vector Quantization (AVQ) techniques may be adopted to obtain a small but near?optimal codebook. However, all known methods for codebook design have drawbacks, such as heavy computational burden and inaccuracy in assigning membership values.
In this thesis, a critical review of these motion?compensation and VQ algorithms is presented, highlighting their limitations. An attempt has been made to provide solutions to these problems and develop a time?efficient algorithm for interframe coding of image sequences.
A new interframe image?sequence coding scheme based on motion compensation is developed with the aim of minimizing the coding time while maintaining the decoded image quality and compression ratio. The proposed IDHS algorithm reduces the number of search locations. Compensation for the residual error is done by VQ using a small codebook. To achieve high quality even with a small codebook, two techniques are employed:
A new AVQ scheme for continuous updating of the codebook, and
PSCVQ for high?quality codebook generation.
Our proposed fast?search algorithm for VQ speeds up both the codebook?generation process and the encoding process. Simulation results confirm that this technique for motion?compensated interframe coding can be used efficiently for reduction of temporal redundancy in real?time low?bit?rate image?sequence coding. | |
