Implementing Delaunay Triangulation and Thin Plate Spine algorithms to blend faces or: How I learned to stop worrying and love the glossy, fabulous, and incomprehensible Internet machine.
Delaunay Triangulation involves generating an optimal triangular mesh over an image where the corners of each triangle are at control points (eyes, nose, mouth, etc.) To blend faces, meshes are generated for two face images and averaged into one mesh. The pixels in each triangle of one image are then warped towards the corresponding triangle in the average mesh. Doing this for both images and overlaying them results in a blended face. Using the Thin Plate Spline method involves the same concept, but uses one single transformation for all pixels determined by a spline function. The final result is the ability to blend two faces together into a final face that is easily tweak-able to look more like one face or the other.
CIS 581 Computer Vision & Computational Photography project