Frequency component vectorisation for image dehazing

Image captured in bad weather conditions confines scene prominence, appears grey and diminishes image contrast. This usually happens due to atmospheric dispersing phenomenon that affects the quality of outdoor computer vision frameworks. This deprivation relies on the gap between the object point and the camera and mostly differs for every pixel present in an image. Transmission coefficients, which state the aforementioned dependence are used to manage the haze level in each pixel. Our algorithm is subject to the presumption that the haze-free image forms clusters given in RGB space. The pixels over the image plane are often found at different locations and their distance from camera also differs. These fluctuating distances give rise to different transmission coefficients. Consequently, these colour clusters form the certain lines of colours in RGB space known as haze lines. The first step is to assure accurate estimation of the atmospheric light, for this an additional wavelet channel is recommended, based on frequency subdivision. The next step is to separate the average gradients present in the foggy regions of an image according to the frequency criteria. Lastly, the haze-free image information is retrieved by utilising the atmospheric scattering model on low and high frequencies according to the edges of unpredicted change in the field depth. Using the non-local and frequency information retrieval, proposed algorithm recovers the haze-free image, efficiently.