Learning based image transformation using convolutional neural networks

Hou, Xianxu; Gong, Yuanhao; Liu, Bozhi; Sun, Ke; Liu, Jingxin; Xu, Bolei; Duan, Jiang; Qiu, Guoping

Abstract

We have developed a learning-based image transformation framework and successfully applied it to three common image transformation operations: downscaling, decolorization, and high dynamic range image tone mapping. We use a convolutional neural network (CNN) as a non-linear mapping function to transform an input image to a desired output. A separate CNN network trained for a very large image classification task is used as a feature extractor to construct the training loss function of the image transformation CNN. Unlike similar applications in the related literature such as image super-resolution, none of the problems addressed in this paper have a known ground truth or target. For each problem, we reason about
a suitable learning objective function and develop an effective solution. This is the first work that uses deep learning to solve and unify these three common image processing tasks. We present experimental results to demonstrate the effectiveness of the new technique and its state-of-the-art performances.

Citation

Hou, X., Gong, Y., Liu, B., Sun, K., Liu, J., Xu, B., …Qiu, G. (2018). Learning based image transformation using convolutional neural networks. IEEE Access, 6, 49779-49792. https://doi.org/10.1109/access.2018.2868733

• Journal Article Type: Article
• Acceptance Date: Aug 28, 2018
• Online Publication Date: Sep 6, 2018
• Publication Date: Sep 28, 2018
• Deposit Date: Feb 1, 2019
• Publicly Available Date: Feb 1, 2019
• Journal: IEEE Access
• Electronic ISSN: 2169-3536
• Publisher: Institute of Electrical and Electronics Engineers
• Peer Reviewed: Peer Reviewed
• Volume: 6
• Pages: 49779-49792
• DOI: https://doi.org/10.1109/access.2018.2868733
• Keywords: Deep learning; image downscaling; image decolorization; HDR image tone mapping
• Public URL: https://nottingham-repository.worktribe.com/output/1513622
• Publisher URL: https://ieeexplore.ieee.org/document/8456517