Abstract: The visibility of images of outdoor scenes is degraded by bad weather conditions. Atmospheric phenomena like haze and fog reduce significantly the visibility of the captured image. Haze is the atmospheric phenomenon that dims the clarity of an observed scene due to the particles such as smoke, fog, and dust. Due to these atmospheric particles there is a significant degradation in the color and contrast of the captured image in the bad weather conditions. If two or more images of same scene are given, then the process of image matching requires find valid corresponding feature points in images. Image matching is a fundamental aspect of many problems in computer vision. Besides the geometric and photometric variation, outdoor and aerial images that are subjected to the process of matching are often degraded by the atmospheric phenomenon of haze. In this paper we presented an efficient physics based method for recovering a haze-free image when a single photograph is available as an input. This technique restores the hazy images based on the estimated transmission (depth) map. Our method benefits from three main contributions. The first is a new constraint on the scene transmission. Second contribution is contextual regularization that enables us to incorporate a filter bank into image dehazing. Our final contribution is an efficient optimization scheme, enables us to quickly dehaze images of large sizes. Our method requires some general assumption and can restore a high quality haze free image. At final stage we have performed matching of images by SURF operator to show efficiency of our method and also demonstrate that our technique is suitable for the challenging problem of image matching based on local feature points.

Keywords: Single image dehazing; Haze; Air light; Transmission map; Image matching; Local feature detectors and descriptors; Speeded up Robust Feature (SURF).