I am currently working in area of computer vision for my master's thesis and am trying to use distances between landmark points on the face for recognition.
My question is this: What landmark points should I use for measuring distances between landmarks? And, since distance is dependent on scale, what ratios of distances should I take?
I propose reading this answer in stackoverflow for "how to apply face keypoint to face recognition?"
But specifically for this question for using landmark point you should first normalize some criteria in all face. for example you can resize all faces into the same eye distance, so all of them will have the same eye distance and then choose some landmark points and calculate their distance as a feature. But I highly recommend that landmark distances are NOT a good feature due to their variable position when face pose changes. Another approach can be using them for face normalization for other types of feature extraction.
