An Improved Deep Learning Based Low Light Image Enhancement Model

Abstract

Low-light image enhancement (LLIE) is important for various practical uses, aiming to improve their visual quality. Deep learning advances have been the driving force behind recent development in this field. Modern techniques utilize sophisticated neural networks to boost image brightness, contrast, and reduce noise, representing the cutting edge of image processing and computer vision. A new approach An Improved Deep Learning Model (AIDLM) is introduced, designed to tackle the challenges of LLIE. AIDLM consists of three main modules: a retinex decomposition network, an enhancement attention network, and a denoising network. The retinex decomposition network uses an explicit parametric regularized Retinex model tailored to individual pixels, while the enhancement attention network enhances the reflectance and illuminance of the V-channel in the HSV color space through spatial and channel attention mechanisms added to the UNet-like architecture , thus preventing color distortion. The denoising module further cleans the enhanced RGB image by eliminating noise from the H and S channels. Comprehensive experiments showed that AIDLM , greatly surpasses current baselines in LLIE, providing higher image quality and robustness.