MORPHOLOGICAL SNOWFLAKES FOR ROBUST NONLINEAR FILTERING
Résumé
The aim of this paper is to introduce nonlinear operators which are robust to noise, by combining the computation of the max/min for the dilation/erosion with an embedded robust filtering step. More precisely, the unitary robust nonlinear filters are computed using a new set of hexagonal symmetry-based structuring elements, called snowflakes. Each snowflake is composed of the union of a central pixel and six micro- neighbourhoods. In this framework, there are two different families of filters which can be defined: second order-operators and local selective operators. Besides the comparison of the practical behaviour of the various families of filters, some preliminary results on their algebraic properties and robustness against noise are given.