Adaptive predictive block matching with weighted histogram filter postprocessing

In this paper we propose to increase the precision of block matching motion estimation by introducing adaptive weights known from bilateral filtering into the SAD cost function. Estimation is carried out pixel-wise, such that a dense vector field is estimated of same resolution as the input images. A combination of hierarchical and spatio-temporally predictive estimation framework is used for processing. With the proposed method highly precise motion vectors even for fine image structures can be computed. We further propose a histogram based adaptive filter method for vector field postprocessing to reduce the amount of remaining outliers while further increasing the vector field precision to boundaries of moving objects.