Research on Low-Light Image Enhancement Algorithm Based on Attention Mechanism

Abstract

Low-light image enhancement remains a critical challenge in computer vision applications, particularly in autonomous driving, surveillance systems, and mobile photography. Traditional enhancement methods suffer from noise amplification and detail loss, while recent deep learning approaches lack efficient feature selection mechanisms. This paper presents a novel attention-based neural network architecture specifically designed for low-light image enhancement. The proposed method integrates channel attention and spatial attention mechanisms within an encoder-decoder framework to selectively enhance important visual features while suppressing noise artifacts. The network employs multi-scale feature extraction modules combined with perceptual loss functions to preserve structural details and natural color reproduction. Extensive experiments on benchmark datasets demonstrate significant improvements in both quantitative metrics and visual quality compared to state-of-the-art methods. The proposed attention mechanism achieves superior performance with PSNR improvements of 2.8dB and SSIM gains of 0.12 over baseline approaches. Computational efficiency analysis reveals real-time processing capabilities suitable for practical applications.