We implemented the inverse compositional image alignment framework, initially proposed by Baker & Matthews with an application to facial AAM fitting in mind.
The main variants of the fitting algorithm (simultaneous, project-out, weighted simultaneous) have been implemented. We have used GNU Octave (a libre Matlab clone) as a framework for test and development. Some parts of the algorithm have been implemented in C++ for efficiency purpose.
Update : a new pure C++ version could also be found on the github repository.
Please see the Github project