Improved Low-Light Image Enhancement using DecNet with Attention Mechanisms

Low-light image enhancement is an issue of immediate concern in computer vision and has deep implications in many applications like surveillance, autonomous vehicles, and medical imaging. Conventional enhancement techniques tend to fail in the tasks of denoising, preserving details, and preserving color integrity. Advanced deep learning-based methods based on Retinex theory have recently emerged with spectacular outcomes by decomposing images into illumination and reflectance parts for selective enhancement. A better framework for low-light image enhancement is introduced by adding a Convolutional Block Attention Module (CBAM) to the Decomposition Network (DecNet). CBAM facilitates more accurate feature extraction with the combination of channel and spatial attention mechanisms and thus adaptive refinement of features for improved brightness adjustment, noise suppression, and detail preservation. Our approach refines the illumination and reflectance components and dynamically adapts the enhancement process according to image content. Benchmark large-scale experiments show that our DecNet improved with CBAM performs better than standard DecNet and other deep learning models. Quantitative measurements with Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) illustrate better performance with PSNR being 24.58 dB and SSIM being 0.887. Qualitative assessment shows better contrast and subjective quality, and the method is better suited for real-world low-light image improvement tasks.