Low-Light Image Enhancement and Color Correction Using a Contrast-Driven Neural Network

This paper presents a novel approach to low-light image enhancement and color correction, utilizing a contrast-driven neural network. The proposed method employs a Contrast Encoder to identify and learn the low-light regions within images, followed by an Enhancement Decoder Network that enhances image quality specifically within these areas. To prevent issues like color distortion and over-correction, training is conducted using the Low-Light (LOL) Dataset and an Underwater Image Dataset, which provide a variety of challenging low-light conditions. Experimental results indicate that the proposed method effectively improves both illuminance and contrast when compared to existing approaches. Moreover, the generated images exhibit high quality, demonstrated by low scores on the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE), which indicate better-perceived image quality.