Modeling the displaced frame difference as an alpha-stable distribution

We propose a probabilistic model of the displaced frame difference (DFD) within a region based on symmetric alpha-stable distributions. The motivation for this model lies an the robust estimation of motion. When the usual assumption of a single motion within a region is violated the observed DFD distribution shows heavier tails than those of the commonly assumed Gaussian distribution. This behavior can be modeled by alpha-stable distributions that comprise a broad class of probability density functions, including the Gaussian case. Some examples applying this model to the DFD between regions of natural sequences are provided