TY - GEN
T1 - Single image reflection removal beyond linearity
AU - Wen, Qiang
AU - Tan, Yinjie
AU - Qin, Jing
AU - Liu, Wenxi
AU - Han, Guoqiang
AU - He, Shengfeng
N1 - Funding Information:
Acknowledgements. This project is supported by the National Natural Science Foundation of China (No. 61472145, No. 61702104, and No. 61702194), the Innovation and Technology Fund of Hong Kong (Project No. ITS/319/17), the Special Fund of Science and Technology Research and Development on Application From Guangdong Province (SF-STRDA-GD) (No. 2016B010127003), the Guangzhou Key Industrial Technology Research fund (No. 201802010036), and the Guangdong Natural Science Foundation (No. 2017A030312008).
Publisher Copyright:
© 2019 IEEE.
PY - 2019/6
Y1 - 2019/6
N2 - Due to the lack of paired data, the training of image reflection removal relies heavily on synthesizing reflection images. However, existing methods model reflection as a linear combination model, which cannot fully simulate the real-world scenarios. In this paper, we inject non-linearity into reflection removal from two aspects. First, instead of synthesizing reflection with a fixed combination factor or kernel, we propose to synthesize reflection images by predicting a non-linear alpha blending mask. This enables a free combination of different blurry kernels, leading to a controllable and diverse reflection synthesis. Second, we design a cascaded network for reflection removal with three tasks: Predicting the transmission layer, reflection layer, and the non-linear alpha blending mask. The former two tasks are the fundamental outputs, while the latter one being the side output of the network. This side output, on the other hand, making the training a closed loop, so that the separated transmission and reflection layers can be recombined together for training with a reconstruction loss. Extensive quantitative and qualitative experiments demonstrate the proposed synthesis and removal approaches outperforms state-of-the-art methods on two standard benchmarks, as well as in real-world scenarios.
AB - Due to the lack of paired data, the training of image reflection removal relies heavily on synthesizing reflection images. However, existing methods model reflection as a linear combination model, which cannot fully simulate the real-world scenarios. In this paper, we inject non-linearity into reflection removal from two aspects. First, instead of synthesizing reflection with a fixed combination factor or kernel, we propose to synthesize reflection images by predicting a non-linear alpha blending mask. This enables a free combination of different blurry kernels, leading to a controllable and diverse reflection synthesis. Second, we design a cascaded network for reflection removal with three tasks: Predicting the transmission layer, reflection layer, and the non-linear alpha blending mask. The former two tasks are the fundamental outputs, while the latter one being the side output of the network. This side output, on the other hand, making the training a closed loop, so that the separated transmission and reflection layers can be recombined together for training with a reconstruction loss. Extensive quantitative and qualitative experiments demonstrate the proposed synthesis and removal approaches outperforms state-of-the-art methods on two standard benchmarks, as well as in real-world scenarios.
KW - Low-level Vision
KW - Vision + Graphics
UR - http://www.scopus.com/inward/record.url?scp=85078720468&partnerID=8YFLogxK
U2 - 10.1109/CVPR.2019.00389
DO - 10.1109/CVPR.2019.00389
M3 - Conference article published in proceeding or book
AN - SCOPUS:85078720468
T3 - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
SP - 3766
EP - 3774
BT - Proceedings - 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2019
PB - IEEE Computer Society
T2 - 32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2019
Y2 - 16 June 2019 through 20 June 2019
ER -