Simple fast and memory conservative methods for dehazing high resolution images

Abstract

Images recorded under inclement weather often exhibit problems such as being too newlinelight, too dark or not having enough contrast. Image enhancement methods process newlinethese types of images so that the resultant images are more suitable than the original newlineimage for specific applications. Visibility in digital images of outdoor scenes is newlinereduced by atmospheric degradation such as fog, haze, snow and rain. newlineOne of the most common atmospheric effects which reduces visibility is haze, newlinewhich is formed due to dust, smoke, moisture and suspended particles in the air. newlineImages acquired under hazy environment need processing for improving their contrast newlineand color fidelity. Removing haze is an important pre processing technique as it newlinehelps in improving accuracy of many computer vision algorithms. The prime focus newlineof this work is to remove haze effects from images acquired under hazy environment newlineand to retrieve the clear image. Though several techniques have been proposed in newlinethe literature for removing haze from an acquired image, the techniques which give newlineanything close to satisfactory results require a large number of advanced and critical newlinesteps for computation. This motivates us to develop haze removal algorithms which newlineare computationally simple, and at the same time, produce excellent results. newlineAs the first part of the work, a simple and fast dehazing algorithm is developed. newlineTo this purpose, a center surround filter is employed to improve speed and memory newlinerequirements of the transmission estimation in image dehazing. The proposed newlinetechnique relies on deriving an alternative transmission estimate by filtering the newlineinput image in three different color spaces, namely RGB, Lab and HSV. Results newlineare evaluated on a collection of images ranging from low to very high resolution. newlineThough applying a simple filter resulted in a fast method, it did not yield desirable newlineresults.