Low light imaging conditions significantly degrade visibility, reduce contrast, and amplify noise, which adversely affects computer vision applications. This paper presents a hybrid physics guided deep learning framework for low light image enhancement by integrating the Absorption Light Scattering Model with a Convolutional Neural Network. In the first stage, atmospheric light and transmission are estimated using a physically interpretable illumination correction strategy to generate a coarse enhanced image. In the second stage, a lightweight convolutional neural network refines the enhanced output by learning residual corrections from paired training data. The proposed framework combines physical modelling with data driven optimization to improve brightness restoration, structural consistency, and noise suppression. Experiments conducted on the LOL dataset demonstrate that the proposed method achieves an average peak signal to noise ratio of 22.84 decibels and a structural similarity index of 0.8670. The results confirm that the hybrid approach effectively preserves structural details while enhancing perceptual quality across diverse low light scenes.
Srinivasu et al. (Tue,) studied this question.