Research Progress on Image Enhancement Technology Based on Deep Learning

Abstract

Image enhancement techniques can improve image quality, making it more suitable for human visual perception or machine analysis needs. In recent years, with the advancement of deep learning, image enhancement methods have seen significant improvements in both effectiveness and applicability. This paper reviews the major advancements in deep learning-based image enhancement techniques, covering typical tasks such as low-light enhancement, super-resolution, and underwater image restoration. It analyzes the mechanisms and performance of different models, and examines their capabilities in noise suppression, detail recovery, and color authenticity. However, current deep learningbased image enhancement methods still face limitations in computational efficiency and practical applicability. To improve deployability, this paper proposes a strategy combining lightweight networks with fully-supervised or semi-supervised learning, achieving effective image enhancement without significantly compromising visual authenticity. Through fully-supervised or semi-supervised approaches, the model maintains stability in scenarios with limited data or significant domain shifts. Future work will focus on inference speed, computational cost, model size, and practical performance on resource-constrained devices as core evaluation metrics, aiming to enhance real-time capability while preserving visual realism.