Multiple Scale Comparative Analysis of Classical, Dynamic and Intelligent Edge Detection Schemes.

Edge detection acts as a fundamental segmentation technique in the fields of remote sensing, computer vision and pattern recognition. It locates significant discontinuities and variations of digital images so as to identify intrinsic edge information involved. Various edge detection schemes have been implemented in numerous cases of science and engineering successfully, such as the classical edge detection (e.g. Canny, Sobel, Laplacian), dynamic edge detection (e.g. Gabor, Curvelets), and intelligent edge detection (e.g. Ant Colony Optimization, Particle Swarm Optimization, Genetic Algorithm) based on computational intelligence. However, there is still a lack of a systematic approach to analyze merits and drawbacks of the existing edge detection schemes from both qualitative and quantitative points of view. In fact, features of detected edges or contours can be represented at multiple scales, such as the 24-bit RGB scale, 8-bit gray scale and single bit binary scale. In this article, some typical edge detection techniques of Canny edge detection, Gabor edge detection and ACO edge detection are used to illustrate classical, dynamic and intelligent edge detection schemes, respectively. Several complex skyline digital images are selected in case studies. Qualitative analysis is conducted to examine visual appeals of detection outcomes based on three schemes at the RGB scale; while quantitative analysis will be conducted to compare edge detection outcomes based on three schemes at the gray and binary scales instead, in the frequency domain and spatial domain, respectively. It provides a comprehensive approach to thoroughly evaluate the overall quality of edge detection schemes.