MULTI-EXPOSURE FUSION FOR LOW-LIGHT AND OVEREXPOSED IMAGE ENHANCEMENT

Abstract

Abstract: Image quality degradation due to improper exposure remains a critical challenge in digital photography and computer vision applications. Low-light conditions result in poor visibility, high noise levels, and loss of detail, while overexposure leads to clipped highlights and information loss in bright regions. This paper presents a comprehensive investigation of multi-exposure fusion (MEF) techniques for enhancing both underexposed and overexposed images. We evaluate various exposure fusion algorithms, including weight map-based approaches, deep learning methods, and hybrid techniques. Our experimental results demonstrate that adaptive weight assignment strategies combined with pyramidal decomposition achieve superior performance in preserving details, maintaining color fidelity, and reducing artifacts. The proposed framework processes multiple exposure brackets to generate high-quality images with balanced illumination across all regions. Quantitative analysis using structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and perceptual quality metrics reveals significant improvements over single-exposure enhancement methods. The findings indicate that MEF approaches are particularly effective for challenging scenarios involving extreme dynamic range variations, with average SSIM improvements of 23.7% and PSNR gains of 4.8 dB compared to conventional tone-mapping techniques. Keywords: Multi-exposure fusion, low-light enhancement, overexposure correction, image quality assessment, dynamic range compression, computational photography