Yixiang Sun, Mengyao Ni, Ming Zhao, Zhenyu Yang, Yuanlong Peng, Danhua Cao
{"title":"采用双分支特征融合和可学习正则化注意力的弱光增强方法","authors":"Yixiang Sun, Mengyao Ni, Ming Zhao, Zhenyu Yang, Yuanlong Peng, Danhua Cao","doi":"10.1007/s12200-024-00129-z","DOIUrl":null,"url":null,"abstract":"<p><p>Restricted by the lighting conditions, the images captured at night tend to suffer from color aberration, noise, and other unfavorable factors, making it difficult for subsequent vision-based applications. To solve this problem, we propose a two-stage size-controllable low-light enhancement method, named Dual Fusion Enhancement Net (DFEN). The whole algorithm is built on a double U-Net structure, implementing brightness adjustment and detail revision respectively. A dual branch feature fusion module is adopted to enhance its ability of feature extraction and aggregation. We also design a learnable regularized attention module to balance the enhancement effect on different regions. Besides, we introduce a cosine training strategy to smooth the transition of the training target from the brightness adjustment stage to the detail revision stage during the training process. The proposed DFEN is tested on several low-light datasets, and the experimental results demonstrate that the algorithm achieves superior enhancement results with the similar parameters. It is worth noting that the lightest DFEN model reaches 11 FPS for image size of 1224×1024 in an RTX 3090 GPU.</p>","PeriodicalId":12685,"journal":{"name":"Frontiers of Optoelectronics","volume":null,"pages":null},"PeriodicalIF":4.1000,"publicationDate":"2024-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11324645/pdf/","citationCount":"0","resultStr":"{\"title\":\"Low-light enhancement method with dual branch feature fusion and learnable regularized attention.\",\"authors\":\"Yixiang Sun, Mengyao Ni, Ming Zhao, Zhenyu Yang, Yuanlong Peng, Danhua Cao\",\"doi\":\"10.1007/s12200-024-00129-z\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>Restricted by the lighting conditions, the images captured at night tend to suffer from color aberration, noise, and other unfavorable factors, making it difficult for subsequent vision-based applications. To solve this problem, we propose a two-stage size-controllable low-light enhancement method, named Dual Fusion Enhancement Net (DFEN). The whole algorithm is built on a double U-Net structure, implementing brightness adjustment and detail revision respectively. A dual branch feature fusion module is adopted to enhance its ability of feature extraction and aggregation. We also design a learnable regularized attention module to balance the enhancement effect on different regions. Besides, we introduce a cosine training strategy to smooth the transition of the training target from the brightness adjustment stage to the detail revision stage during the training process. The proposed DFEN is tested on several low-light datasets, and the experimental results demonstrate that the algorithm achieves superior enhancement results with the similar parameters. It is worth noting that the lightest DFEN model reaches 11 FPS for image size of 1224×1024 in an RTX 3090 GPU.</p>\",\"PeriodicalId\":12685,\"journal\":{\"name\":\"Frontiers of Optoelectronics\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":4.1000,\"publicationDate\":\"2024-08-14\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11324645/pdf/\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Frontiers of Optoelectronics\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://doi.org/10.1007/s12200-024-00129-z\",\"RegionNum\":3,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ENGINEERING, ELECTRICAL & ELECTRONIC\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Frontiers of Optoelectronics","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.1007/s12200-024-00129-z","RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
Low-light enhancement method with dual branch feature fusion and learnable regularized attention.
Restricted by the lighting conditions, the images captured at night tend to suffer from color aberration, noise, and other unfavorable factors, making it difficult for subsequent vision-based applications. To solve this problem, we propose a two-stage size-controllable low-light enhancement method, named Dual Fusion Enhancement Net (DFEN). The whole algorithm is built on a double U-Net structure, implementing brightness adjustment and detail revision respectively. A dual branch feature fusion module is adopted to enhance its ability of feature extraction and aggregation. We also design a learnable regularized attention module to balance the enhancement effect on different regions. Besides, we introduce a cosine training strategy to smooth the transition of the training target from the brightness adjustment stage to the detail revision stage during the training process. The proposed DFEN is tested on several low-light datasets, and the experimental results demonstrate that the algorithm achieves superior enhancement results with the similar parameters. It is worth noting that the lightest DFEN model reaches 11 FPS for image size of 1224×1024 in an RTX 3090 GPU.
期刊介绍:
Frontiers of Optoelectronics seeks to provide a multidisciplinary forum for a broad mix of peer-reviewed academic papers in order to promote rapid communication and exchange between researchers in China and abroad. It introduces and reflects significant achievements being made in the field of photonics or optoelectronics. The topics include, but are not limited to, semiconductor optoelectronics, nano-photonics, information photonics, energy photonics, ultrafast photonics, biomedical photonics, nonlinear photonics, fiber optics, laser and terahertz technology and intelligent photonics. The journal publishes reviews, research articles, letters, comments, special issues and so on.
Frontiers of Optoelectronics especially encourages papers from new emerging and multidisciplinary areas, papers reflecting the international trends of research and development, and on special topics reporting progress made in the field of optoelectronics. All published papers will reflect the original thoughts of researchers and practitioners on basic theories, design and new technology in optoelectronics.
Frontiers of Optoelectronics is strictly peer-reviewed and only accepts original submissions in English. It is a fully OA journal and the APCs are covered by Higher Education Press and Huazhong University of Science and Technology.
● Presents the latest developments in optoelectronics and optics
● Emphasizes the latest developments of new optoelectronic materials, devices, systems and applications
● Covers industrial photonics, information photonics, biomedical photonics, energy photonics, laser and terahertz technology, and more