Bad weather (such as fog, haze, mist) brings trouble to many computer vision applications like outdoor surveillance systems, object detection, pedestrian detection and intelligent transport systems. Dark channel method with fixed rectangular patches is used in many algorithms for the removal of fog. The paper aims at an improved dark channel prior where fixed rectangular patches are replaced by homogeneous regions, called superpixels. In most cases, superpixels are obtained when quick shift segmentation with fixed parameters has been employed. The proposed quick shift is automatic in nature and determines optimal parameters on the fly to obtain dark channel image. To judge improvement in visibility, a reference k-means based image is used. Improved algorithm based on k-means and quick shift segmentation has been proposed which is qualitatively more effective than other state of art methods in terms of edge visibility, halo effects and restoration of natural scene radiance.

automatic quick shift, dark channel, fog removal, superpixel, segmentation.