講演抄録/キーワード |
講演名 |
2021-10-22 13:45
A Tiny Convolutional Neural Network for Image Super-Resolution ○Kazuya Urazoe・Nobutaka Kuroki・Yu Kato・Shinya Ohtani(Kobe Univ.)・Tetsuya Hirose(Osaka Univ.)・Masahiro Numa(Kobe Univ.) IMQ2021-7 |
抄録 |
(和) |
(まだ登録されていません) |
(英) |
This paper surveys three techniques for reducing computational costs of convolutional neural network (CNN) for image super-resolution. The main purpose of this paper is finding a tiny CNN for image super-resolution. After the three techniques are introduced, this paper shows a tiny CNN for high-speed image super-resolution that combines those techniques. Experimental results of 2$times$ image magnifications have shown that the processing speed of the tiny CNN is about 32.89, 7.99, and 3.10 times faster than that of conventional SRCNN, MCH, and FSRCNN, respectively, with no image quality degradation. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Super-resolution / Resolution enhancement / Convolutional neural network / Deep learning / Global residual learning / / / |
文献情報 |
信学技報, vol. 121, no. 217, IMQ2021-7, pp. 2-7, 2021年10月. |
資料番号 |
IMQ2021-7 |
発行日 |
2021-10-15 (IMQ) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
IMQ2021-7 |
|