CCMMF: color constancy method based on mobileViT and multi-scale fusion

Xiaoyu Zheng; Xuming Lu

doi:10.1117/12.2689475

31 July 2023 CCMMF: color constancy method based on mobileViT and multi-scale fusion

Xiaoyu Zheng, Xuming Lu

Proceedings Volume 12747, Third International Conference on Optics and Image Processing (ICOIP 2023); 127470A (2023) https://doi.org/10.1117/12.2689475
Event: Third International Conference on Optics and Image Processing (ICOIP 2023), 2023, Hangzhou, China

Abstract

In recent years, color constancy has made great progress due to the application of convolutional neural networks (CNNs). However, CNNs cannot extract global features efficiently, and the use of only high-level features for estimation results in a large degree of information loss. In addition, the model based on CNN used for color constancy is not lightweight enough for application. To overcome those problems, a color constancy method based on MobileViT and multi-scale fusion is proposed. The method adopts the improved network structure based on MobileViT as the feature extraction module. It can make the extraction of local features and global representations simultaneously, with a complexity-reduced model. In further, a novel bidirectional multi-scale fusion network with two paths is presented to facilitate feature fusion between different levels. It makes full use of the multi-scale representation with strong semantics. As a result, more accurate illumination estimation is obtained. Experimental results show that compared with other methods, the proposed method has better comprehensive performance. The illumination estimation error of this method reaches the state-of-art level. With the size of parameters 1.24 MB, it is much lighter than mainstream networks.

Citation Download Citation

Xiaoyu Zheng and Xuming Lu "CCMMF: color constancy method based on mobileViT and multi-scale fusion", Proc. SPIE 12747, Third International Conference on Optics and Image Processing (ICOIP 2023), 127470A (31 July 2023); https://doi.org/10.1117/12.2689475

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