Studies on Content-Adaptive Image Coding by Universal Codes
| Project/Area Number |
62550235
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
電子通信系統工学
|
|---|
| Research Institution | University of Tokyo |
|---|
Principal Investigator |
KATO Shigeo Research Associate Institute of Industrial Science University of Tokyo, 生産技術研究所, 助手 (00143529)
|
|---|
| Project Period (FY) |
1987 – 1988
|
| Project Status |
Completed (Fiscal Year 1988)
|
| Budget Amount *help |
¥1,900,000 (Direct Cost: ¥1,900,000)
Fiscal Year 1988: ¥200,000 (Direct Cost: ¥200,000)
Fiscal Year 1987: ¥1,700,000 (Direct Cost: ¥1,700,000)
|
|---|
| Keywords | Image Coding / Source Coding / Universal Code / Arithmetic Coding / データ圧縮 |
|---|
| Research Abstract |
Recently, document-image communication systems and videotex systems have been developed. Images in these systems are quite different in the statistical characteristics from each other. It is hoped to develop a universal coding algorithm commonly applying to these systems. In this study, we proposed new content-adaptive image coding schemes using universal codes. Typical universal codes are Arithemtic code and Ziv-Lempel code. At first, we proposed a dynamic Markov model coding scheme based on a markov model using an arithmetic code, in which coding parameters can be dynamically changed in accordance with local statistical characteristics of the image. This characteristics is effective for coding of images, in which the statistical properties locally change. But in the mixed images with chracters and graphics, reference pel positions which give the minimum conditional entropy also change according to the local statistical structure.
In our coding scheme, reference pels, besides coding parameters, can be dynamically selected by measuring the conditional level distribution under various reference pels, while executing the coding process. Simulation results show that the coding entropy in this scheme for the combined CCITT test document and dithered image is about 30 percents better than in the conventional Markov model entropy.
Next, we investigated the way to apply the Ziv-Lempel code to the multi-level images. Ziv-Lempel code has learning effects in itself so it is possible to automatically follow to the statistical characteristics. In our method, the predictive coding scheme is combined to the original Ziv-Lempel code, so that the coding efficiencies are more increased.
|
Report
(3 results)
Research Products
(22 results)
