Direct Estimation of Displacement Histograms

To overcome the shortcomings of conventional displacement estimation techniques we propose to estimate the displacement probability density function (pdf) rather than a single displacement vector. The displacement pdf is related to the cross spectrum between successive frames and the signal power spectrum. We derive techniques to calculate the displacement pdf using either a forward image model, a backward image model, or a combination of both models. In an application example the displacement pdf is estimated for a real image pair, and the result is used for motion-based image segmentation.