Two-way exposure fusion network model and method for low-illumination image enhancement

Abstract

The invention discloses a double-channel exposure fusion network model and method for low-illumination image enhancement. The network model comprises a first enhancement branch, a second enhancement branch and a fusion module. Wherein the first enhancement branch and the second enhancement branch are used for processing a to-be-enhanced low-illumination image by using different preset enhancementstrategies respectively; and the fusion module comprises an attention unit and a fine adjustment unit and is used for carrying out weighted fusion on the enhancement results obtained by the first enhancement branch and the second enhancement branch to obtain an enhanced image. In addition, the invention also provides a denoising strategy divided into two steps, so as to carry out efficient adaptive suppression on the noise in the enhancement process. The model provided by the invention has clear physical explanation and is light and effective. The network model can be used for low-illuminationenhancement tasks in various low-light environments, especially in consideration of changes of imaging environments and high requirements for processing real-time performance.

Description

technical field [0001] The invention relates to the technical field of low-illuminance image enhancement, in particular to a two-way exposure fusion network model and method for low-illuminance image enhancement. Background technique [0002] Compared with imaging conditions with sufficient light, images captured in low light are often accompanied by darker areas, unpredictable noise and blurred details, with lower contrast and dynamic range. Low-light shots typically occur when the environment is relatively dark (e.g., at night or when brightness is limited), or it may be due to the photographer failing to properly adjust the imaging device (e.g., inappropriate ISO, aperture size, or exposure time setting). Since the enhancement of low-light images is of great significance in low-level or high-level semantic image processing tasks such as night imaging, monitoring and security equipment, and unmanned driving technology, how to effectively enhance low-light images has becom...

AI Technical Summary

Problems solved by technology

For example, a method trained on a particular dataset is likely to favor a certain range of luminances and scenes, and thus be limited in its ability to generalize

Although GANs can be trained with unmatched datasets, they can sometimes generate inappropriate colors, making the augmentation results unnatural or oversaturated

Furthermore, direct image-to-image translation is relatively more difficult to implement than augmentation strategies incorporating physical priors, especially when noise in low-light conditions is also taken into account.

Images