Detecting and Dehazing Image In Real-Time

Abstract

Haze formation results from the combined effects of airlight and attenuation. While attenuation reduces scene contrast, airlight contributes to increased whiteness in the captured image. Atmospheric conditions caused by suspended particles such as fog and haze significantly deteriorate image quality. Eliminating haze from a single image of a weather-degraded scene is a challenging problem, primarily because haze distribution depends on unknown depth information. Therefore, effective haze removal techniques are highly valuable for various computer vision applications including surveillance systems, object detection, tracking, and image segmentation. This study concentrates on reducing block artifacts and other degradation factors present in natural scene images affected by fog and haze, thereby simplifying the process of image enhancement. In this work, a Derived Guided Filter-based visibility restoration method is employed to address issues such as inaccurate transmission map estimation and color distortion. The Dark Channel Prior combined with the derived Guided Filter is utilized with the objective of integrating processes like denoising, color correction, and additional enhancement techniques within a unified singleimage dehazing framework.