In this section, the core of the proposed algorithm, the implementation of the proposed texton framework is described. The proposed approach requires a dictionary of textons, such that the histograms can be determined. The histograms are then used as features for the \(k\)-Nearest Neighbors (\(k\)NN) algorithm. The outputs of this regression technique are possible \(x,y\)-coordinates for a given image.

Importantly, for generating the training dataset, no subsampling was used. Therefore, the training dataset will not show any variance, if the same images were used.

In the following pseudo code, \(M\) represents the number of particles, \(z_{t}^{x}\) the output of the texton framework at time \(t\), and \(f_{t}^{x}\) the estimated flow at time \(t\).