Abstract | ||
---|---|---|
This paper shows that it is possible to train large and deep convolutional neural networks (CNN) for JPEG compressionartifacts reduction, and that such networks can provide significantly better reconstruction quality compared topreviously used smaller networks as well as to any other state-of-the-art methods. We were able to train networkswith 8 layers in a single step and in relatively short time by combining residual learning, skip architecture, andsymmetric weight initialization. We provide further insights into convolution networks for JPEG artifact reductionby evaluating three different objectives, generalization with respect to training dataset size, and generalization withrespect to JPEG quality level. |
Year | Venue | Field |
---|---|---|
2016 | arXiv: Computer Vision and Pattern Recognition | Residual,Lossless JPEG,Pattern recognition,Compression artifact,Convolution,Convolutional neural network,Computer science,JPEG,Artificial intelligence,Initialization,Quality level,Machine learning |
DocType | Volume | Issue |
Journal | abs/1605.00366 | 2 |
Citations | PageRank | References |
13 | 0.54 | 19 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Pavel Svoboda | 1 | 13 | 0.88 |
Michal Hradis | 2 | 132 | 14.19 |
David Barina | 3 | 41 | 8.24 |
Pavel Zemcík | 4 | 120 | 24.73 |