Image registration using the length code algorithm

Traditional image registration algorithms are often computationally intensive and therefore costly and time consuming. For instance, methods based on the cross-correlation function or on the average magnitude difference function (AMDF) involve a considerable amount of computation time. The paper investigates a technique which is referred to as the "length code algorithm". The length code of an object in an image is the distance between the centroid of the object and a point on the object boundary. An object can therefore be represented as a sequence of length codes. Since the cyclic auto-correlation of a length code sequence is both translational and rotational invariant, the translational and rotational displacements between two images can be determined. A set of experiments devised to compare the cost/performance of length code based image registration to the AMDF based registration shows that the length code algorithm is more efficient than the AMDF. That is, the length code algorithm produces about the same registration result with less computational effort.