A rapid algorithm for curve denoising
Zhifang Lu, Baojiang Zhong · 2015
Contour curves of shapes are important features for image analysis and understanding. A contour curve might be affected by various kinds of noise. In this paper a rapid curve denoising algorithm is proposed, which is equivalent to the existing Gaussian smoothing according to a convergence property of noise. A key problem of the rapid algorithm is to select a proper smoothing radius. We first sample a few of points on the noisy curve, with which the severity of noise is measured. Then a linear model is established to interpret the relationship between the noise severity and the smoothing radius. Finally, based on a series of numerical tests, the smoothing radius is evaluated. Experimental results show that the proposed algorithm has a comparable performance as the Gaussian; however, at much less computational cost.