Develop image registration or matching algorithm using mutual information or similar technqiue

I want someone to develop a mutual information algorithm to take 'x' number of templates and find the region of interest on an image which best matches one of the templates. The algorithm must return a similarity measure which can be used to evaluate how good the match is. The algorithm should be translational and rotational invariant. i.e moving template size and static image size will not match, also the target ROI in the static image can be at any angle or orientation. Algorithm should be developed in C++ and should be fast. Templates along with sample static images will be provided to develop the algorithm. You must have experience in mutual information or other similar algorithm development to bid on the project.