An Investigative Study on Deep Learning-Based Image Dehazing Techniques
G. L. Narasamba Vanguri, Sangram Keshari Swain, M. Vamsi Krishna · Advances in computer science research · 2024
Image dehazing is a complex challenge within the field of computer vision, particularly when dealing with hazy or foggy scenes.Photographs captured in unfavourable weather conditions (such as haze, fog, smog, and mist) often suffer from significant degradation.These deteriorated images pose difficulties for various computer vision applications, including video surveillance, smart transportation, weather forecasting, and remote sensing.The task of mitigating these adverse effects is commonly referred to as image dehazing.In recent years, deep learning (DL) techniques have garnered substantial attention for addressing challenging image dehazing problems.Notably, architectures like Generative Adversarial Networks (GANs) and Convolutional Neural Networks (CNNs) have revolutionized the field.CNNs excel at capturing spatial hierarchies and extracting meaningful features, while GANs leverage adversarial training to enhance the fidelity of dehazed images.By combining feature extraction and reconstruction, these DL models enable the restoration of clarity in hazy scenes.This article provides an extensive analysis of DL-based dehazing methods proposed by various researchers.It covers their performance, datasets used, evaluation metrics, and recent advancements.The goal is to improve the effectiveness and precision of dehazing algorithms, ultimately benefiting a wide range of applications.