Section: New Results

Spatially-variant kernel for optical flow under low signal-to-noise ratios

Participant : Charles Kervrann.

Local and global approaches can be identified as the two main classes of optical flow estimation methods. This year, we have proposed a framework to combine the advantages of these two principles, namely robustness to noise of the local approach and discontinuity preservation of the global approach. The idea is to adapt spatially the local support of the local parametric constraint in the combined local-global model [42]. To this end, we jointly estimate the motion field and the parameters of the spatial support. We apply our approach to the case of Gaussian filtering, and we derive efficient minimization schemes for usual data terms. The estimation of a spatially varying standard deviation map prevents from the smoothing of motion discontinuities, while ensuring robustness to noise. We validated our method for a standard model and demonstrated how a baseline approach with pixel-wise data term can be improved when integrated in our framework. The method has been evaluated on the Middlebury benchmark with ground truth and on real fluorescence microscopy data for which noise is the main limitation for usual optical flow methods.

Collaborator: Denis Fortun (EPFL-BIG, Lausanne, Switzerland)

                         Noémie Debroux (Laboratory of Mathematics, INSA Rouen, Normandie University)

Figure 10. Visual results obtained with variants of CLG (“Combined Local-Global”) methods for several values of regularization parameters $\lambda$: CLG-A (our adaptive CLG), CLG${}_0$ (pointwise method) and CLG ( [42]) on a image pair from a CElegans sequence.