Target: To capture spatial and temporal joint distribution of motion vectors.
Summary: MCF is a simple motion feature that characterizes motion content of a video. Essential idea is to represent joint distribution of motion vectors in a spatio-temporal manner. A finite alphabet is utilized to quantize motion vectors. Then, co-occurrence statistics of motion vectors are computed within a spatio-temporal search window. The resulting 2D co-occurrence histogram describes the motion content in compact manner, which in turn can be utilized for classification or matching purposes.
|Regular Motion||Irregular Motion|