A Novel Low-Light Image Enhancement Algorithm Based on Retinex Channel Attention Network

The main challenges of low-light images include increased noise, loss of detail, and lack of contrast. Due to poor lighting, details in images are often difficult to discern, and noise in low-light environments can easily affect image quality. Although existing low-light image enhancement technologies improve brightness and detail recovery, there are still some problems such as insufficient noise suppression, slow processing speed, and poor adaptability to specific scenes, which limit their widespread popularity and effect stability in practical applications. To address these issues, we propose a novel Retinex channel attention network (RCA-Net) to enhance low-light images. Firstly, the decomposition network is used to decompose the low-illumination image into reflectance and illumination components. Then the illumination image is enhanced, while the denoising method is carried out on the reflected image. Finally, the adjusted illuminance and reflectance are reconstructed to produce the enhanced image. Compared with the conventional Retinex-Net, our approach incorporates a channel attention mechanism into the enhancement network and integrates perceptual loss into the model’s loss function. Experimental results show that the proposed network achieves better results than the existing methods.