Vector quantization for image compression: Algorithms and performance
Guardado en:
| Publicado en: | ProQuest Dissertations and Theses (1992) |
|---|---|
| Autor principal: | |
| Publicado: |
ProQuest Dissertations & Theses
|
| Materias: | |
| Acceso en línea: | Citation/Abstract Full Text - PDF |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Resumen: | Quantization is a process of converting a variable of continuous value into one of a set of discrete values. This is necessary in the conversion of analog representation into digital representation, which permits the use of computer technology in applications involving speech, images, and video. The problem is the tremendous amount of data associated with such applications. Vector quantization, in which several parameters are quantized together, can be used to reduce the data rate. However, computational difficulty arises with the use of large vectors. This dissertation is focused on complexity analysis, algorithms design, and performance evaluation of vector quantization for image compression. Both theoretical analysis and experimental results are presented. More efficient and effective algorithms have been developed for vector quantization. Although it has been widely believed that achieving optimal vector quantization is difficult, the exact complexity of the problem is not known. Many existing algorithms are based on heuristic search which does not give any performance guarantee. Based on a graph formulation of the problem, the complexity of optimal vector quantization design is proved to be NP-hard, which gives justification for the heuristic approaches. The performance of existing heuristics is analyzed in terms of approximation bounds. Unfortunately, finding approximate solutions which are provably within a constant factor of the optimal is also shown to be difficult. One practical alternative to solving the problem is tree-structured vector quantization, which has added advantages of fast quantization and successive approximations. This approach can be formulated as an optimization problem under various cost constraints. The complexity of the problem varies from different constraints. Efficient optimization as well as heuristic algorithms have been developed. The performance analysis of these algorithms indicates that most of the performance loss of tree-structured vector quantizers obtained is due to the heuristic tree search. Strategies have been proposed to significantly improve the performance of tree search. The computation difficulty of vector quantization also hinders its application for on-line compression. Massively parallel algorithms have been proposed to speed up the design of vector quantizers as well as the quantization process. |
|---|---|
| ISBN: | 9798208428627 |
| Fuente: | ProQuest Dissertations & Theses Global |