Tetrolet Transform and Dual Dictionary Learning-Based Single Image Fog Removal

The removal of fog or haze from video frames and images has been a major focus in the area of computer vision since fog has a detrimental effect on monitoring and surveillance systems, as well as on the recognition of scene objects and other applications. Numerous defogging strategies have been presented thus far, including those based on the "colour-line model", polarization, "anisotropic diffusion", and the "dark channel prior" (DCP). Nevertheless, when the scene counters a thick fog and sky regions, these approaches fail to provide high-quality output. The authors suggest a novel haze/fog removal approach that uses tetrolet transformation to decompose a foggy image into low- and high-frequency components based on their structural information and dual dictionary learning-based residual frequency extractor to extract additional residual image information. DCP operation is performed on the low-frequency component to recover more fog-free information while sharpening the tetrolet coefficients extracts finer details. The inverse transformed image is then added to the residual high-frequency image component and post-processed using contrast limited adaptive histogram equalization to balance the contrast. Lastly, S and V channel gain regulator optimizes the contrast-enhanced image's colour and intensity. Compared to current methodologies, the suggested method significantly improves the overall picture quality. Quantitative and qualitative data support the statements.