Optimal Filters for Extended Optical Flow

Estimation of optical flow and physically motivated brightness changes can be formulated as parameter estimation in linear models. Accuracy of this estimation heavily depends on the filter families used to implement the models. In this paper we focus on models whose terms are all data dependent and therefore are best estimated via total-least-squares (TLS) or similar estimators. Using three different linear models we derive model dependent optimality criteria based on transfer functions of filter families with given fixed size. Using a simple optimization procedure, we demonstrate typical properties of optimal filter sets for optical flow, simultaneous estimation of optical flow and diffusion, as well as optical flow and exponential decay. Exemplarily we show their performance and state some useful choices.