A method of operating a video processor is provided. The method includes receiving a first image data representing a first frame of a scene, dividing at least a central portion of the first frame into a plurality of tiles, identifying feature point candidates in the first image data, selecting a set of feature points within each specific tile based on the luminance variance σ2 of the specific tile, receiving a second image data representing a second frame of the scene, deriving feature point motion vectors between the first and second frames corresponding to the identified feature, grouping the motion vectors into motion vector groups based on vector magnitude ratios and angular differences, and calculating the affine transform of each group of motion vectors.