Abstract

I develop a method for efficient, multiscale template matching using orthogonal wavelet representations. I also develop an efficient approximate mechanism for initially finding the \correct” values for the basis coefficients ajk at the finest scale (as opposed to using the sample values as coefficients). I then present applications of these techniques to two important image coding/computer vision tasks: motion estimation and multiple model matching (model recognition).