Application of the EM technique to estimation of affine modeled image motion

The objective of this investigation is to utilize the iterative estimation-maximization (EM) technique to find the maximum likelihood estimate of the parameters of the affine motion model. The authors present the theoretical development of affine modeled image motion estimation in the presence of noise. Affine modeled image motion represents a very important class of motion such as reflection, rotaton, skew, scaling, and translation. The affine motion model is a parametric function; hence, when images undergo this type of motion, the estimation of the motion parameters proves to be efficient and computationally less exhaustive than the pixel by pixel motion estimation algorithms. The EM technique used converts the maximization of a log-likelihood function from a coupled parameter maximization problem into simpler and decoupled parameter maximization problems. The simulation results prove the validity of the development of the EM technique in dealing with estimation of affine image motion