2016 Fiscal Year Final Research Report
Learning-Based Design and Implementation of Non-separable Oversampled Lapped Transforms for Multidimensional Signal Restoration
| Project/Area Number |
26420347
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Communication/Network engineering
|
|---|
| Research Institution | Niigata University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2014-04-01 – 2017-03-31
|
| Keywords | 冗長変換 / スパース表現 / ボリュームデータ復元 / 辞書学習 / GPGPU実装 / 確率的勾配降下法 / タイトフレーム / 画像処理 |
|---|
| Outline of Final Research Achievements |
In this project, we proposed a multidimensional transform with the redundant, non-separable, overlapped, symmetric, compact-supported and tight property. We also conducted the theoretical analysis, design, implementation, and application development. First, we extended the existing non-separable lapped orthogonal transform to redundant configuration and clarify its properties. As well, we proposed an example-based learning design method and showed its effectiveness. In addition, we showed the possibility of real-time processing through GPGPU/FPGA implementation. Besides, it was applied to image/volumetric data restoration and validity was confirmed. We also extended the proposed transform to complex coefficient type as preparation for application development to complex image restoration processing.
|
| Free Research Field |
信号処理
|
