Image Enhancement
Overview
Image enhancement studies how to improve images affected by low light, poor exposure, haze, rain, noise, compression, or limited dynamic range, while retaining useful detail for people and vision systems. Common approaches include Retinex decomposition and tone curves, convolutional and transformer networks, state-space models, lookup tables, physical priors, and diffusion-based restoration; training may use paired data, unpaired data, or test-time adaptation. Methods are evaluated on task-specific benchmarks such as LOL for low-light images, UIEB for underwater enhancement, and MIT-Adobe FiveK for retouching, alongside real-world tests and downstream tasks like detection or segmentation. A central tension is improving brightness and perceptual appeal without distorting color, erasing structure, or inventing details, while also generalizing across cameras and scenes at practical speeds.
Latest papers
156 papers9 papers in the last four weeks, down 25% on the four weeks before. 0.1% of all new papers.