Log sigmoid function based patch independent image haze removal method

This paper contains a novel haze-removal algorithm, where the evaluation of the desired channel containing pixels with minimum intensity values is carried out using a patch-independent method in contrast to most of the existing methods, where this was done by dividing the images into several patches of fixed size. This helps to overcome the halo-artifacts, which are mainly present along the edges, where there is a sharp change of intensities due to non-uniform transmission within the patches. Here the atmospheric light evaluated by using a binary Support Vector Machine classifier and transmission estimation is performed by introducing a constant K1, whose value is dependent on the pixel intensity values in respective color channels. The contrast enhancement of the haze-free images obtained after scene radiance recovery and removal of artifacts present mostly in the sky region is done by using log-sigmoid function and a constant K2, whose values are dependent on standard deviation values and mean values of histograms of each color channel respectively. Moreover satisfactory results are obtained by performing comparative study of qualitative and quantitative analyses of output images obtained by applying this proposed method on hazy images with respect to various, noteworthy existing methods.