Gradient-based motion vector estimation using variable block shape

Motion vector estimation of image sequence is one of the important process in image coding, robot vision etc. A useful method of motion vector estimation in the gradient-based method which uses the relationship between the spatial intensity gradient of image and the intensity difference between frames. We propose two block gradient-based methods using variable block shape and size in order to realize accurate estimation of motion vector. The first one uses variable block shape, and estimation is performed at every pixels. The second method uses square block with variable size, and estimation is done for each block. Simulation results show that both algorithms give more accurate estimation than conventional block gradient-based method.