| Presentation | 2007-05-21 Image restoration with the truncated Gaussian model Hiroyuki TANAKA, Keiji MIURA, Masato OKADA, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | On the theory of image restoration, one tries to reconstruct the original image from the image to which a noise is added because of a channel or surrounding environment. A general strategy of it is to use the framework of Bayesian estimation. Under that framework, information of original imageis represented by probability distribution. When considering more realistic model of original image, one should treat the boundary of the pixel value (gray scale, color, brightness, etc) correctly. The reason is that the pixel value is defined by finite region, e.g, white-to-black. However, when using the Gaussian model that is a conventional analog image model, probability density is positive outside the boundary. To solve this problem, we introduce a truncated Gaussian model (TG model). In this model, distribution is cut at upper and lower bound, and added a weight at the boundary as δ peak. In order to analyze such a nonlinear model, we introduce a generalized prior probability. This prior includes the Gaussian, Ising, Q-spin, and TG model as a special cases. Moreover, we can analytically obtain average performance of the generalized model by using the replica method. Thus, we can choose an appropriate model depending on statistical property of the image. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | statistical mechanics / replica method / image restoration / bayesian estimation / TG model |
| Paper # | NC2007-7 |
| Date of Issue |
| Conference Information | |
| Committee | NC |
|---|---|
| Conference Date | 2007/5/14(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Neurocomputing (NC) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Image restoration with the truncated Gaussian model |
| Sub Title (in English) | |
| Keyword(1) | statistical mechanics |
| Keyword(2) | replica method |
| Keyword(3) | image restoration |
| Keyword(4) | bayesian estimation |
| Keyword(5) | TG model |
| 1st Author's Name | Hiroyuki TANAKA |
| 1st Author's Affiliation | Graduate School of Frontier Sciences, University of Tokyo() |
| 2nd Author's Name | Keiji MIURA |
| 2nd Author's Affiliation | Graduate School of Frontier Sciences, University of Tokyo |
| 3rd Author's Name | Masato OKADA |
| 3rd Author's Affiliation | Graduate School of Frontier Sciences, University of Tokyo:Brain Science Institute, RIKEN |
| Date | 2007-05-21 |
| Paper # | NC2007-7 |
| Volume (vol) | vol.107 |
| Number (no) | 50 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |