Practical first-difference edge detection with subpixel accuracy

摘要

Edge detection using conventional zero-crossing analysis based on second differences appears to have certain limitations, while attempts at first-difference edge detection are often troubled by such things as spurious or multiple responses and poor localization. Based on a study of human visual threshold behaviour, and experience gained with both zero-crossing analysis and related topics using hexagonal sampling arrays, a novel form of first-difference edge detection is proposed which is free from these problems and other limitations of zero-crossing analysis. It is robust against random additive noise and has the property of sensing unresolved line features (as opposed to edges) as lines rather than first differences of lines. The basic properties of the method are discussed and the practical capabilities illustrated briefly.