TY - JOUR
T1 - Discriminative Transfer Learning for General Image Restoration
AU - Xiao, Lei
AU - Heide, Felix
AU - Heidrich, Wolfgang
AU - Scholkopf, Bernhard
AU - Hirsch, Michael
N1 - Funding Information:
Manuscript received June 28, 2017; revised March 10, 2018; accepted April 23, 2018. Date of publication April 30, 2018; date of current version May 24, 2018. This work was in part supported by King Abdullah University of Science and Technology under individual baseline funding. The associate editor coordinating the review of this manuscript and approving it for publication was Prof. Weisi Lin. (Corresponding author: Lei Xiao.) L. Xiao was with the Department of Computer Science, The University of British Columbia, Vancouver, BC V6T 1Z4, Canada. He is now with Facebook Reality Labs, Redmond, WA 98052 USA (e-mail: leixiao@cs.ubc.ca).
Publisher Copyright:
© 1992-2012 IEEE.
PY - 2018/8
Y1 - 2018/8
N2 - Recently, several discriminative learning approaches have been proposed for effective image restoration, achieving convincing tradeoff between image quality and computational efficiency. However, these methods require separate training for each restoration task (e.g., denoising, deblurring, and demosaicing) and problem condition (e.g., noise level of input images). This makes it time-consuming and difficult to encompass all tasks and conditions during training. In this paper, we propose a discriminative transfer learning method that incorporates formal proximal optimization and discriminative learning for general image restoration. The method requires a single-pass discriminative training and allows for reuse across various problems and conditions while achieving an efficiency comparable to previous discriminative approaches. Furthermore, after being trained, our model can be easily transferred to new likelihood terms to solve untrained tasks, or be combined with existing priors to further improve image restoration quality.
AB - Recently, several discriminative learning approaches have been proposed for effective image restoration, achieving convincing tradeoff between image quality and computational efficiency. However, these methods require separate training for each restoration task (e.g., denoising, deblurring, and demosaicing) and problem condition (e.g., noise level of input images). This makes it time-consuming and difficult to encompass all tasks and conditions during training. In this paper, we propose a discriminative transfer learning method that incorporates formal proximal optimization and discriminative learning for general image restoration. The method requires a single-pass discriminative training and allows for reuse across various problems and conditions while achieving an efficiency comparable to previous discriminative approaches. Furthermore, after being trained, our model can be easily transferred to new likelihood terms to solve untrained tasks, or be combined with existing priors to further improve image restoration quality.
KW - Image restoration
KW - discriminative learning
KW - proximal optimization
UR - http://www.scopus.com/inward/record.url?scp=85046376142&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85046376142&partnerID=8YFLogxK
U2 - 10.1109/TIP.2018.2831925
DO - 10.1109/TIP.2018.2831925
M3 - Article
C2 - 29993740
AN - SCOPUS:85046376142
SN - 1057-7149
VL - 27
SP - 4091
EP - 4104
JO - IEEE Transactions on Image Processing
JF - IEEE Transactions on Image Processing
IS - 8
ER -