Adaptable edge quality metric

摘要: A new quality metric for evaluating edges detected by digital image processing algorithms is presented. The a weighted sum of measures edge continuity, smoothness, thinness, localization, detection, nd noisiness. Through training process, we can design weights that optimize the different users and applications. We have used to compare results ten detectors when applied degraded varying degrees blur types noise. As expected, more optimum Laplacian-of-Gaussians (LoG) filter Haralick's second derivative method outperform simpler gradient detectors. At high SNR, best choice, although it exhibits sudden drop in performance at lower SNRs. LoG filter's degrades almost linearly with SNR maintains reasonably level same relative performances are observed as varied. For most tested, drops increasing noise correlation. Noise correlated direction destructive tested.