Interesting! Are you able to share anything about how your algorithm works? I've been playing around with feature detection (using HoG descriptors) plus constrained local models to identify locations in an image and then manipulating the mesh to better match these locations and extracting a corresponding texture.