Combined Light and Dark Priors over Variational Auto-encoder (CLDP-VAE) for single image dehazing

Haze lowers the quality of images by making them harder to see and warping colours and shapes that makes it challenging to use in many computer vision apps. This article suggests a new Combined Light and Dark Priors Variational Autoencoder (CLDP-VAE) for cleaning up a single picture. This approach is based on the fact that the separation among an image's highest and lowest colour channels is negatively related to its depth. This method slowly brings back features like frameworks, borders, angles, and colours through focussing on sub-pixels and blocks with high contrast. Some of the important contributions highlighted are the mechanism of simultaneously incorporating Light Contrast Prior (LCP) together with Dark Contrast Prior (DCP) in order to obtain the contrast details from both the illuminated and the shaded areas for effective dehazing. The model also uses the LAB color space to enhance the attributes of clarity and structure in restored pictures. The chosen blocks are merged together to make a better picture. This better picture is sub sequently used to teach a Variational Autoencoder (VAE) how to remove haze. We test how well this approach works through conducting a lot of studies on the REalistic Single Image DEhazing (RESIDE) dataset using metrics like peak-signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). These findings show that the CLDP-VAE does a better job than the latest techniques, such as FFA-Net, Maxim, UDN, AECR-Net, and C2PNet, getting clearer images and better structure resemblance. This suggested model provides a strong and effective way to clear up haze in an identical picture, making a big step forward inside this field.