PROJECT TITLE :

AIPNet Image-to-Image Single Image Dehazing With Atmospheric Illumination Prior

ABSTRACT:

The natural phenomena of haze is caused by the dispersion and absorption of the atmosphere, which greatly reduces the visibility of scenery. As a result, the image captured by the camera is susceptible to overexposure and confusion. It is impossible to dehaze a single image without using a prior-atmospheric illumination prior that is simple yet extraordinary. Statistics and theoretical evaluations across a wide range of colorspaces show that ambient light in cloudy weather mostly affects brightness (luminosity) and has little effect on chrominance (chrominance). We strive to preserve the foggy scene's original hue and contrast in accordance with this presumption. In order to accomplish this, we use multiscale convolutional networks, which can automatically detect foggy spots and restore texture information that has been lost. Comparatively to earlier methods, the deep CNNs not only achieve complete model trainability, but also a simple system architecture from one image to another. For a number of dehazing effects, the suggested methodology outperforms existing methods in detailed comparisons and evaluations.