Integrating Contextual Mechanisms into Unsupervised Semantic Segmentation for Improving Low-Light Image Enhancement

Low-light image enhancement becomes increasingly important in computer vision applications, especially in scenarios such as nighttime surveillance, autonomous driving, and medical imaging. Contextual semantic information is crucial for opti-mizing feature representation and improving perceptual quality, as it helps the model understand the relationships between objects in the image and their spatial distribution, thereby enabling more effective detail recovery and enhancement of image quality. Although the Semantic-Guided Zero-shot (SG Z) model utilizes semantic information, it still faces the issue of insufficient integration of contextual semantic information, which leads to an incomplete understanding of object relationships and negatively impacts detail recovery and image quality. To address this issue, we propose a method that integrates contextual mechanisms into unsupervised semantic segmentation to improve low-light image enhancement (SGZSA). This novel approach embeds self-attention into the feature extraction process of the SG Z model's Unsupervised Semantic Segmentation (USS), enhancing its capa-bility to capture contextual relationships. Our experiments on benchmark datasets demonstrate that SG ZSA surpasses existing methods in both qualitative and quantitative metrics, achieving notable improvements in brightness and contrast while retaining fine details. These results underscore the effectiveness of self-attention mechanisms in low-light image enhancement and pave the way for future advancements in the field.