Fitting smooth curves

A novel method for fitting smooth curves to noisy data is presented. A curve is represented by two nonlinear functions of arc length. The smoothing procedure is essentially iterative and includes three major steps: coefficient estimation, computation of fitted values, and reparameterization. The coefficients are first locally estimated based on the weighted-least-squares principle. Fitted values are then found through data projection. This leads to a self-consistent curve. Finally, the functions are reparameterized to unit speed, which gives the curve a smooth appearance. The problem of shrinkage with curver fitting does not occur with the proposed method, and no shrinkage compensation is required. Preliminary experimental results using synthetic contours and real computer tomography image data are given