FUSION-BASED TECHNIQUES FOR IMAGE DEHAZING

Abstract

Image dehazing is a common term used to refer to the restoration, enhancement and de-blurring of images captured outdoors. It is a widely studied topic in literature and spans a large range of areas namely, outdoor surveillance cameras, satellite imaging, remote sensing, underwater imaging, object tracking, computer vision, etc. Haze or corruption is a general term for the lack of clarity in an image when captured in moving/ low-light/ foggy conditions. Clarity here refers to good contrast, good lighting and good reflectivity of the object or scene, to be captured in the image. Atmospheric haze usually occurs due to multiple factors, especially in outdoor shoots, like dust, smoke, hailstorm, clouds and fog. Fog is an aerosol of water vapour and tiny ice crystals suspended in the air. In contrast to the general aerosol-type nature of the contamination, the case of haze in underwater images is an interesting topic to consider. Several authors consider the underwater haze similar to dense fog on land, but light below water is spectrally deprived. This happens because water redirects different wavelengths of light by different angles of refraction causing Spectral Distortion. The Dissertation encompasses a review of the different means and techniques of image dehazing followed by a novel strategy to dehaze indoor and outdoor environmental images. The review on Image Dehazing Techniques first introduces haze in images, its types, causes and effects observed. The relevant literature is compiled and evaluated first. Following are the sections talking about the classifications of dehazing techniques in detail. Each sub-section lists the comparison of different papers for that particular class. A separate section just for the databases in common parlance helps the reader to select amongst available choices. Eventually, the review includes the most recent research areas and topics under focus and finally draws out a relevant conclusion and charts a suitable goal area for research works. Image dehazing by Image Enhancement focuses on improving the view of hidden objects in the image. On the other hand, Image Restoration in its fullest sense is a method that aims at re-doing or inverting the degradations performed on the image, that made it blurry in the first place. In a general sense, the Image Restoration method for Image Dehazing is a Restoration-Degradation model using a clear version of the image. Image Enhancement offers a significant advantage in the sense that it does not mandate the use of a template but, it is not efficient in case there is a vast array of images of the same area from different angles. Image Fusion then takes the lead because the method works by inculcating features from all available images of the scene/object under surveillance. Furthermore, in instances where images of a ‘moving’ object, like a human lung, growing cancer tumour, etc., are recorded, image fusion becomes necessary to combine all geometric transformations of the object to form a single master image. The Dissertation introduces a novel method that uses Multi-Scale Image Fusion. The same is followed by another idea which combines the Retinex technique with Multi Scale Image Dehazing so as to improve the results. The choice of the Retinex method is based on the fact that the output of the Retinex method mimics human vision and gives a good representation of the features of the hazy image. The star of the show is the simultaneous use of both the hazy and ground truth’s reflectance maps while performing fusion. The visual and quantitative comparisons of the present and previous works have also been performed and presented in this Dissertation.