I'm trying to do face retouching for a video frame by frame. Firstly, I detect some facial landmarks for the face in the video as control points. I adjust the control points using moving least squares to make the face look good. The moving percent is calculated like dviding moving offset from origin control points to distance from origin control points to control point on the nose. The parameters are saved and used by every frame in the video.
The problems are as follows:
- How to keep image border constant? Some black pixels are on the border after deformation. (Like "pin edges" in Liquify tool in photoshop ?)
- Any other high-performance image deformation? MLS seems slow, I'm trying to get it real time in python.
