| Presentation | 2007-09-04 A Super-Resolution Method based on Learning of High Frequency Components which Minimize Errors Yasunori Taguchi, Toshiyuki Ono, Takeshi Mita, Takashi Ida, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In this paper, a novel learning method for example based super-resolution is proposed. Example based super-resolution methods provide sharp high resolution images, using correlation between low frequency component and high frequency component in a lot of examples. Most related studies take an approach to register as many examples as possible. On the other hand, a method for constructing an efficient example database by selecting or creating good examples for super-resolution, is not taken into consideration. This paper proposes a method to learn an optimal correlation of examples in the sense of minimizing the sum of squared errors of obtained images to training images. In this method, representative examples, which contribute to making images faithful to the original images, are created by the K-means method and the closed loop training method, and then registered. Using the proposed method, high quality super-resolution images will be provided with a small amount of memory. PSNR was improved about 1dB in comparison with an interpolation method and the conventional method registering all examples. A subjective evaluation experiment using a paired comparison method showed effectiveness of the proposed method. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | |
| Paper # | PRMU2007-88,HIP2007-97 |
| Date of Issue |
| Conference Information | |
| Committee | HIP |
|---|---|
| Conference Date | 2007/8/27(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 | Human Information Processing (HIP) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A Super-Resolution Method based on Learning of High Frequency Components which Minimize Errors |
| Sub Title (in English) | |
| Keyword(1) | |
| 1st Author's Name | Yasunori Taguchi |
| 1st Author's Affiliation | Corporate Research and Development Center, Toshiba Corporation() |
| 2nd Author's Name | Toshiyuki Ono |
| 2nd Author's Affiliation | Corporate Research and Development Center, Toshiba Corporation |
| 3rd Author's Name | Takeshi Mita |
| 3rd Author's Affiliation | Corporate Research and Development Center, Toshiba Corporation |
| 4th Author's Name | Takashi Ida |
| 4th Author's Affiliation | Corporate Research and Development Center, Toshiba Corporation |
| Date | 2007-09-04 |
| Paper # | PRMU2007-88,HIP2007-97 |
| Volume (vol) | vol.107 |
| Number (no) | 207 |
| Page | pp.pp.- |
| #Pages | 8 |
| Date of Issue | |