| Edge detection is fundamental task for computer vision systems. It has been used extensively as a preprocessing step for a myriad of image processing algorithms, including image enhancement, object detection/recognition, compression, and digital watermarking algorithms. Such algorithms, in turn, have been used for medical, military, security, and consumer applications. As many systems rely on edge detection, the development of accurate edge detection in both clean and noisy environments is a must. Currently, no single edge detection method has been produced whose performance is superior for all applications. This is largely due to the subjective nature of the edge detection problem. Secondly, a reliable, unbiased measure to objectively assess the performance of edge detector outputs has not been developed which can be used universally.;Two new edge detection algorithms and a new objective edge map evaluation measure are introduced upon establishing new generalizations for edge detection and edge map evaluation measures. One edge detection algorithm is based on mean-separate decomposition which directly makes use of the presented measure to determine the best edge detector output of the system. A second method is based on a new generalized set of kernels for edge detection. The proposed methods are compared to edge detection standards, and experimental results show that the proposed algorithms are capable of outperforming standard edge detection techniques in both clean and noisy environments. Throughout the thesis, the presented measure is used as a quantitative, objective means of assessing edge detector performance in addition to human evaluation. |
