Optimal second order derivative filter families for transparent motion estimation

Optimal filter families used to discretize a transparent motion model are constructed in this paper. This model is formulated as an optical-flow-like linear constraint equation. Accuracy of the estimation of its parameters heavily depends on the filter families used to implement it. To optimize filter families we derive a model dependent optimality criterion based on transfer functions of separable filter families with given fixed size. Using a simple optimization procedure, we demonstrate typical properties of optimal filter sets and state some useful choices. Exemplarily we show their performance on synthetic and real data from a botanical application.