Color image processing with B-spline modeling

B-splines have been popular for several purposes. In this work, B-splines are used for modeling noisy multivariate data. A new algorithm is proposed to approximate image data with B-spline surfaces, where image data is assumed to be corrupted by Gaussian noise mixed with outliers. The resulting surface gives a restored continuous representation of discrete data while preserving edges. A gradient operator is applied on the resulting surface to detect edges.