Development of Edge Preserving Image Enhancement Techniques

Abstract

Due to some constraints in imaging techniques and computer graphics, images that are captured have many problems like distorted edges, presence of noise, artifacts, and oversaturated colours. Image enhancement plays a significant role in the preservation of edges, minimization of noise, artifacts, thus have broad applications in domains of image smoothing, filtering, contrast correction, de-hazing, de-blurring, rain-removal, and super-resolution. Specifically, edge preservation techniques are considered for image enhancement to improve the visual quality, which ultimately minimizes noise, artifacts and enhances the quality of the image. Existing image enhancement techniques mostly are inaccurate, result in noise, artifacts, blurred edges, so they do not perform well for different applications. This leads to less visually pleasing results having low quality. In this thesis, edge preservation techniques for image enhancement are proposed for different domains, which include image smoothing, filtering, de-blurring, de-hazing, rain removal, super-resolution, illumination normalization, low light image enhancement, vessel segmentation, re-colouring, and underwater image enhancement. Simple to implement techniques are proposed incorporating existing filters like guided filter, L0 minimization filter etc and machine learning algorithms like PCA, k-means clustering, etc. in different colour spaces like RGB and YCbCr to preserve edges, minimize noise, artifacts, contrast correction and produce visually pleasing results. Image is segmented for smoothing and deblurring, quad-tree decomposition for dehazing, specular band decomposition for illumination normalization, undergoes DFT for recolouring, and Laplace decomposition for underwater image enhancement. Operations such as histogram processing, sharpening, de-noising, morphological operations, arithmetic operations, clustering, color balancing, and white balancing are also performed to preserve edges and minimize noise, artifacts. The proposed techniques produce results with minimum noise, artifacts, and blurred edges. Visual and quantitative comparison (with state of the art existing techniques) is performed to verify the significance of the proposed methods. Simulation results reveal that the proposed techniques are more accurate in edge preservation, minimization of noise, and artifacts as compared to the state of the art techniques.