This paper deals with the approximation of a given large scattered univariate or bivariate data set that possesses certain shape properties, such as convexity, monotonicity, and/or range restrictions. The data are approximated for instance by tensor-product B-splines preserving the shape characteristics present in the data. Shape preservation of the spline approximant is obtained by additional linear constraints. Constraints are constructed which are local linear sufficient conditions in the unknowns for convexity or monotonicity. In addition, it is attractive if the objective function of the resulting minimisation problem is also linear, as the problem can then be written as a linear programming problem. A special linear approach based on constrained least squares is presented, which in the case of large data reduces the complexity of the problem sets in contrast with that obtained for the usual ℓ2-norm as well as the ℓ∞-norm. An algorithm based on iterative knot insertion which generates a sequence of shape preserving approximants is given. It is investigated which linear objective functions are suited to obtain an efficient knot insertion method.
- linearisation of shape constraints
- Spline approximation
- linear constraints