Human face image colorization with dual-scale attention U-Net

Liangqi Chen; Ben Wang; Zhouxin Lu

doi:10.1117/12.2662622

16 December 2022 Human face image colorization with dual-scale attention U-Net

Liangqi Chen, Ben Wang, Zhouxin Lu

Proceedings Volume 12500, Fifth International Conference on Mechatronics and Computer Technology Engineering (MCTE 2022); 125003C (2022) https://doi.org/10.1117/12.2662622
Event: 5th International Conference on Mechatronics and Computer Technology Engineering (MCTE 2022), 2022, Chongqing, China

Abstract

In order to colorize grayscale human face image and solve the problems of boundary leakage and detail loss in current colorization algorithms, we propose the dual-scale attention U-Net. First, on the basis of U-Net, we improve the original convolution block and select two convolution kernels of different sizes to extract features, which improves the ability of deep feature extraction for different scales. Second, integrating CBAM into skip connections helps network to focus on salient regions and suppress unnecessary regions while sharing features and reducing information loss. In addition, inspired by the field of image super resolution, MS-SSIM-L1 is adopted as the loss function, and the edge and texture information of the image can be taken into account when coloring. Experiments show that compared with other colorization algorithms, the proposed method not only reduces boundary leakage and detail loss, but also has outstanding performance in colorizing old photos of historical figures.

Citation Download Citation

Liangqi Chen, Ben Wang, and Zhouxin Lu "Human face image colorization with dual-scale attention U-Net", Proc. SPIE 12500, Fifth International Conference on Mechatronics and Computer Technology Engineering (MCTE 2022), 125003C (16 December 2022); https://doi.org/10.1117/12.2662622

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available