Abstract

Edge detection is a basic image processing operation usually used in the first stage of the complex image processing systems, such as restoration, and its quality has a direct effect on the performance of the systems. The extraction of correct edges from a noise-contaminated image or an image with severe deformation is a challenging task. The objective of the work of this thesis is to develop an edge detection method to extract effectively edge signals from the images with the edge information seriously damaged while being acquired in high dynamic range (HDR) scenes. To achieve the objective, an edge detection method based on a multiple-high-pass-filtering process scheme has been proposed. Each of the filtering processes is designed to suit one of the signal deformation conditions, and is applied to the entire input image, instead of the designated regions, in order to spare the computation of image segmentation. A fusion process is then performed to merge the gradient maps generated by the multiple filtering processes into one. A detection procedure has been designed for a typical case of HDR images acquired with three different kinds of deformations due to the non-ideal characteristics of acquisition device. Based on the study of the characteristics, three high-pass filtering processes are designed to generate gradient signals with different modulations. A simple selection algorithm is developed for an easy fusion process. The results of the simulation with different types of HDR images have shown that, compared to some of most commonly used detection processes, the proposed one leads to a better quality of edge signals from severely deformed HDR images.