Attention-Driven Low Light Image Enhancement Using Lightweight CNN

Low-light image enhancement is an important task in computer vision that aims to improve visibility and preserve critical details in images captured under poor lighting conditions. In the base paper, a CNN-based method using a simple encoder–decoder architecture is employed for image enhancement. Although this approach effectively increases overall brightness, it presents several limitations. The model treats all pixels equally without prioritizing important regions, lacks an attention mechanism to focus on semantically significant features such as faces or text, and struggles to generalize effectively across diverse real-world lighting conditions. To address these limitations, this work proposes an enhanced Low-Light Image Enhancement (LLIE) model that integrates the Convolutional Block Attention Module (CBAM) into the encoder–decoder network. CBAM introduces both Channel Attention, which prioritizes important feature channels, and Spatial Attention, which focuses enhancement on key regions of the image.By incorporating these attention mechanisms, the proposed model improves brightness and clarity while preserving important structural details and features. As a result, the enhanced system produces higher visual quality images and becomes more suitable for practical applications such as surveillance systems, digital photography, and mobile vision applications.