landmark detection vs face recognition

Question

Is face landmark detection does the same thing has face recognition? Or are they 2 diff things? If yes, how they differ from each other?

Any clarification would be appreciate it. Thank you in advance & take care everyone

Answer 1

They are totally different but they are connected.

Facial landmark detection finds several landmark points on face. For example, dlib finds 46 landmarks and retinaface finds 106 facial points. We can apply 2d alignment to the face with those landmarks. Besides, you can model a face with 3D if you have landmarks.

Face recognition is based on verifying a face pair. It represents facial images as multidimensional vectors with a cnn architecture (e.g. vgg), then compare two vectors. It is expected that face pair of a same person should have a low distance, face pair of different persons should have a high distance. Distance could be found with euclideand or cosine formulas.

As I mentioned, face recognition models are convolutional neural networks. And you feed the facial images as inputs to those models. If you apply alignment before feeding images to cnn, then it will increase the accuracy. You can think landmark detection is an early step of a face recognition system.

Answer 2

IMHO face recognition is a higher level process that might include several phases (sub-processes) such as face detection (is there a face on a image and where if so), finding face landmarks (!), computing face encodings, face comparing, based on face encodings.