Between moving least-squares and moving least-\(\ell_1\) (Q747639): Difference between revisions

The author addresses the problem of function approximation based on scattered data affected by noise. One approximation method considered is the moving least-squares method consisting in computing local polynomials which approximate the data locally, using the classical weighted least squares. Even if in the case of a smooth function the approximation is smooth, the disadvantage of the method occurs when one has to deal with outliers. On the other hand, another kind of approximation based on the moving least 1-norm is not sensitive to outliers, but has a lack in the terms of smoothness of the approximation result. Based on these remarks, the author proposes a moving approximation method based on a functional which is a tradeoff between the 1-norm and the 2-norm, leading to a smooth approximation, not sensitive to outliers. An effective iterative algorithm for computing the combined norm is presented and analyzed, showing also that for scattered data with outliers the new error measure is very effective.