I work with the Biometrics lab in the CIC building and am involved in the automatic Facial Landmarking project. Given a face picture, the program can highlight the major facial features such as the eyes, eyebrows, nose, nose bridge, lips, and outline of the face. Approximately 79 points will be automatically fitted.
The program initially has a bunch of data from a training stage. In this phase, the points of interested for the facial outline are manually chosen so that a mean can be calculated as to where these points could be located once face detection occurs. Using this training data, the face is detected and then the image normalized. The testing part of the program begins by placing the mean locations on the face and then finding the optimal location according to a certain algorithm. I work extensively on transferring the R&D from Matlab code to C++ code. Using the OpenCv library, I try to optimize the speed and efficiency thus making the program much faster once in VC++ executable and generating a general framework to allow further fast R&D modules.