2019 Fiscal Year Final Research Report
A study on person re-identification of low-resolution images using degradation model
| Project/Area Number |
17K00238
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Perceptual information processing
|
|---|
| Research Institution | Tottori University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2017-04-01 – 2020-03-31
|
| Keywords | 低解像度 / ダウンサンプリング / 人物照合 / カメラ映像 |
|---|
| Outline of Final Research Achievements |
A problem arises when a camera with a wide field of view has an extremely low resolution of a person, which dramatically reduces the performance of person re-identification. To solve the problem, I have developed a method to identify low-resolution person images using the appropriate resampling technique. I collected a low-resolution image dataset for varying the camera lens's focal length and the distance from the person to the camera. Experimental results show that a downsampling technique for generating low-resolution images is more effective than an upsampling technique for generating high-resolution images. I demonstrated that the low-frequency components obtained by a downsampling technique contained many discriminative features.
|
| Free Research Field |
画像認識
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究の学術的な特色として，低解像度でも人物画像を正しく照合する方法の実現に向けて，画像劣化モデルに基づくリサンプリングを設計したことが挙げられる．低解像度の人物画像を単に高解像度へ復元すれば照合性能が高まると一般的には考えられているが，本研究の成果が示すように必ずしもこれは成り立たない．解像度に応じて逆に解像度を落とす復元方法の開発が重要であることを明らかにした．本研究で開発された方法により，低解像度の人物画像を照合できるようになれば，空港や商業施設など人々が集まる場所に設置されたカメラで見守ることができる範囲が広がるため，本研究の成果は安心・安全な社会の実現に貢献できる意義がある．
|
