Quasi-interpolation operators based on the trivariate seven-direction \(C^2\) quartic box spline (Q639957)

The paper lies on the background of some applications as scientific visualization and medical imaging based on the problem of the construction of non-discrete models from given discrete data on volumetric grids. Standard approaches are based on trivariate tensor-product splines which have, in some cases, the disadvantage related to the problem of nonsmooth models and to the high degrees of the polynomials piecewise. A smooth alternative for this is to take sums of univariate and bivariate \(C^1\) quadratic spline quasi-interpolants. Another known method is to construct approximating schemes based on quadratic or cubic \(C^1\) spline on type-6 tetrahedral partitions (uniform partitions of \(\mathbb R^3\) obtained from a given cube partition of the space by subdividing each cube into 24 tetrahedra by cutting with six planes). The advantage of these methods is based on the possibility to construct smooth spline of lower degree. More recently methods related to the reconstruction of sampled data in signal processing, based on trivariate box splines are used as alternatives to tensor-product schemes. In this framework, the author proposes new quasi-interpolation schemes with the approximation order 4, based on the trivariate \(C^2\) quartic box spline, defined on a type-6 tetrahedral partition. More precisely, a box spline is defined by a set of direction vectors which describe the shape of the support of the box spline and by continuity properties. In the paper, a trivariate \(C^2\) quartic box spline \(B\) defined by a set \(X\) of seven directions that forms a regular partition of the space into tetrahedra is considered. The author investigates the space spanned by the integer translates of this box spline and defines local spline quasi-interpolants by linear combinations of integer translates mentioned above and local linear functionals. In the first step, a differential quasi-interpolant is proposed, having coefficient functionals as linear combinations of values of \(f\) with its partial derivatives at the center of the support of integer translates. The problem arisen is the computation of the derivatives. After this, three discrete quasi-interpolants are obtained by convenient discretisation of the differential coefficients mentioned before, whose coefficient functionals are linear combinations of the values of the function at specific points in the support of integer translates. Also, upper bounds for the norm of the discrete quasi-interpolant operators and for the approximation error are constructed. The performance of the proposed interpolants is illustrated by some numerical examples.