IEEE Transactions on Circuits and Systems for Video Technology · 2021 · 334 citations · 59 references
EngineeringMachine LearningLow-light ImagesUnified Deep FrameworkLow-level VisionImage AnalysisData ScienceImage-based ModelingLow ContrastComputational ImagingComputational PhotographyRadiologyHealth SciencesMachine VisionMedical ImagingRetinex DecompositionComputer ScienceDeep LearningImage EnhancementComputer VisionBiomedical ImagingImage DenoisingImage Restoration
Low‑light images suffer from low contrast and unclear details, limiting both human perception and computer vision applications, and current enhancement techniques focus on Retinex‑based and learning‑based methods. The study aims to bridge the gap between Retinex‑based and learning‑based low‑light image enhancement methods. The authors introduce a generative strategy for Retinex decomposition and a unified deep framework that estimates latent components and performs low‑light enhancement without external images, allowing adjustable illumination. The method weakens the coupling between components during decomposition and outperforms ten state‑of‑the‑art algorithms on seven public datasets. Code is available at https://github.com/zhaozunjin/RetinexDIP.
Low-light images suffer from low contrast and unclear details, which not only reduces the available information for humans but limits the application of computer vision algorithms. Among the existing enhancement techniques, Retinex-based and learning-based methods are under the spotlight today. In this paper, we bridge the gap between the two methods. First, we propose a novel “generative” strategy for Retinex decomposition, by which the decomposition is cast as a generative problem. Second, based on the strategy, a unified deep framework is proposed to estimate the latent components and perform low-light image enhancement. Third, our method can weaken the coupling relationship between the two components while performing Retinex decomposition. Finally, the RetinexDIP performs Retinex decomposition without any external images, and the estimated illumination can be easily adjusted and is used to perform enhancement. The proposed method is compared with ten state-of-the-art algorithms on seven public datasets, and the experimental results demonstrate the superiority of our method. Code is available at: <uri>https://github.com/zhaozunjin/RetinexDIP</uri>.
59