Unsupervised contour estimation
Mário A. T. Figueiredo, José M. N. Leitão · 2002
We introduce a fully adaptive active contour model in which no parameters have to be set a priori or tuned by the user. It is based on elliptic Fourier contour description and on the minimum description length (MDL) principle. The proposed technique estimates all the observation model parameters (e.g., noise variances), the order of the contour description (number of Fourier coefficients), and the contour itself.