A Deep Learning-Based Dehazing Approach for Enhanced Image Visibility
Goilkar Suhasini Shashikant, Shashikant S. Goilkar · 2026
Visual degradation due to atmospheric particles such as haze and dust significantly impairs image clarity, posing substantial challenges for computer vision applications. In response to the increasing demand for high-fidelity imagery in autonomous systems, it is required to develop effective dehazing algorithms. This research focuses on enhancing image quality and mitigating haze effects through the deployment of advanced deep learning frameworks. The proposed methodology incorporates a multi-stage pipeline, comprising image fusion techniques and pre-processing operations including air light estimation, contrast stretching, Contrast Limited Adaptive Histogram Equalization and Histogram Equalization. These stages aim to improve feature extraction and visibility in degraded images by restoring structural and contrast information. Various deep convolutional neural network architectures are implemented to identify the optimal model capable of generating perceptually enhanced, dehazed outputs. The usefulness of each model is quantitatively assessed using image quality assessment metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), benchmarked against ground-truth haze-free images. This study contributes to the evolution of computational image enhancement techniques and demonstrates significant potential for integration into real-time autonomous systems, remote sensing, and intelligent surveillance applications.