Lucas-Kanade algorithm with GNC

The similarity of two arbitrary real functions can be analysed calculating the squared Euclidean distance, the parameter space. The application of this method to digital images was proposed by Lucas and Kanade. In this paper the concept of graduated nonconvexity (GNC) is applied to the problem of evaluating the parameter function. It is shown that the application of GNC to the parameter function and the construction and evaluation of a pyramid of spatially smooth and subsampled images are equivalent operations in particular, but practically relevant cases.