Minimum preserving subsampling-based fast image de-fogging

Abstract

Dark channel prior based techniques have been widely used in image and video de-fogging which produce real and impressive results. Their major limitation is the large computational cost of dark channel estimation. For the image of size M × N, to find n × n dark channel, operations are required, which increase its computational complexity. In this work, a novel approach of image subsampling is proposed, which preserves the value of local minimum in a patch. This subsampled image is used to construct the dark channel to improve the computational efficiency. Transmission map is refined using fast guided filter to remove blocking artifacts. Atmospheric light is calculated by ignoring pixels of bright light sources. To make the results uniformly bright, adaptive post processing is performed on de-fogging results. The image de-fogging technique is further extended for videos. It is demonstrated that proposed technique produces better results than existing state-of-the-art techniques while achieving real-time processing speed.