Learning an adaptive model for extreme low-light raw image processing

• Published in: IET image processing Vol. 14; no. 14; pp. 3433 - 3443
• Main Authors: Fu, Qingxu, Di, Xiaoguang, Zhang, Yu
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.12.2020
• Subjects: adaptive model; brightness; brightness prediction; campus image data; current deep learning models; ES exposure‐shifting operator; global image brightness; guideline exposure time; high‐quality low‐light image enhancement; image denoising; image enhancement; image enhancement vector; image quality; image reconstruction; image resolution; image scene; learning (artificial intelligence); low illumination; lowest noise level estimation score; low‐light raw image enhancement network; low‐light raw image processing; noise‐free image; real‐world images; Research Article; severe noise; single raw image; structural similarity loss; real-world images; campus image data; single raw image; high-quality low-light image enhancement; structural similarity loss; low illumination; image scene; ES exposure-shifting operator; image enhancement; noise-free image; current deep learning models; guideline exposure time; learning (artificial intelligence); image resolution; global image brightness; adaptive model; low-light raw image enhancement network; image reconstruction; image denoising; image quality; brightness; brightness prediction; low-light raw image processing; severe noise; image enhancement vector; lowest noise level estimation score
• ISSN: 1751-9659; 1751-9667
• DOI: 10.1049/iet-ipr.2020.0100

Low-light images suffer from severe noise and low illumination. In this work, the authors propose an adaptive low-light raw image enhancement network to avoid parameter-handcrafting in current deep learning models and to improve image quality. The proposed method can be divided into two sub-models: brightness prediction and exposure shifting (ES). The former is designed to control the brightness of the resulting image by estimating a guideline exposure time $t_1$t1. The latter learns to approximate an exposure-shifting operator ES, converting a low-light image with real exposure time $t_0$t0 to a noise-free image with guideline exposure time $t_1$t1. Additionally, structural similarity loss and image enhancement vector are introduced to promote image quality, and a new campus image data set (CID) for training the proposed model is proposed to overcome the limitations of the existing data sets. In quantitative tests, it is shown that the proposed method has the lowest noise level estimation score compared with the state-of-the-art low-light algorithms, suggesting a superior denoising performance. Furthermore, those tests illustrate that the proposed method is able to adaptively control the global image brightness according to the content of the image scene. Lastly, the potential application in video processing is briefly discussed.