Abstract:
A number of computer vision systems and applications require high resolution, visually enhanced images with high contrast and preserved color and detail information, as inputs. However, in reality, due to the camera sensor limitations and challenging and adverse imaging conditions, such as poor lighting or bad weather, the captured images may suffer from low contrast, reduced visibility, haze, distorted colors or low resolution. Therefore the images need to be enhanced before they can be used for various computer vision systems. Image enhancement techniques aim to improve the visual appearance of images, and make them suitable for human/machine perception, so that they can be used in their required image processing and computer vision applications, such as surveillance and security systems, target identification, scene analysis, medical image processing, satellite imagery and remote sensing. This thesis presents various image enhancement techniques from the perspective of resolution enhancement using super resolution and visibility enhancement using image dehazing. In this regard, five different image enhancement techniques focusing on resolution and visibility enhancement are presented. The first technique focuses on image resolution enhancement, in which compressive sensing through sparse representation, based on self example dictionary learning and guided filtering is used for super resolution of images. The effectiveness of the proposed methodology is verified through quantitative and visual analysis. The last four techniques target visibility enhancement of different types of hazy images including outdoor, underwater, satellite/aerial and low light images through various dehazing methodologies. The second technique is based on filtering, detail enhancement and contrast improvement for the visibility enhancement of underwater images with poor visibility. Visibility enhancement and dehazing of images using local Laplacian filtering and l0 gradient decomposition is proposed as the third enhancement technique. The fourth image visibility enhancement technique uses image decomposition, detail enhancement and fusion for dehazing of images. The fifth technique for visibility enhancement makes use of edge preserving image decomposition and application of different enhancement strategies on the basis of whether the image is a dark low light image, or a hazy image. The technique works well for low light, as well as underwater and outdoor hazy images. The presented techniques generate effectively dehazed, visually plausible images, with enhanced visibility, improved contrast and preserved image details. Visual and quantitative comparison of the presented techniques with existing state of the art techniques demonstrates the effectiveness of the proposed image enhancement methodologies.