Sub-Block Based Global Motion Estimation for Affine Motion Model

Nowadays, Digital video stabilization techniques acquire a very important role for allowing high quality video footage even in non-optimal conditions. Several studies on video stabilization have been done in digital, mechanical and optical platform. This paper proposes a computationally efficient video stabilization algorithm incorporating an affine model to obtain the stabilized video frames. Both rotational and translational motion between frames are estimated to fit into the affine motion model and the parameters are estimated using phase correlation algorithm. The major computational advantage for the proposed work is obtained by deriving the affine parameters from a subblock of a video frame rather than from the entire frame. A criteria is devised for the selection of this sub-block which reduces the estimation errors in addition to the computational saving. The proposed algorithm tested on sample videos shows better performance in terms of PSNR with reduced computations.