Data from a target system (616) that exhibits a non-linear relationship between system properties and measurement data, is processed by applying a linear filter (F) to improve resolution and accuracy. The filter reduces inaccuracies that are introduced by an algorithm used to reconstruct the data. The filter is defined by assigning time varying functions, such as sinusoids (200), to elements (210) in a model (Y) of the system to perturbate the elements. The resulting data output from the model is reconstructed using the same algorithm used to subsequently reconstruct the data from the target system. The filter is defined as a matrix (F) that transforms the reconstructed data of the model (X) back toward the known properties of the model (Y). A library of filters can be pre-calculated for different applications. In an example implementation, the system is tissue that is imaged using optical tomography.