Eigenface can be a good algorithm to start with if you want to build a system for educational purposes, as it is relatively simple and serves as a starting point for many other algorithms in this area. Basically, you make a bunch of face images (training data), switch them to shades of gray if they are RGB, change their size so that each image has the same dimensions, turns images into vectors, stacking columns of images (which are now two-dimensional matrices), calculate the average value of each pixel in all images and subtract this value from each entry in the matrix, so that the component vectors will not be affine. After that, you calculate the covariance matrix of the result, solve its eigenvalues and eigenvectors, and find the main components.These components will serve as the basis for the vector space and together describe the most important ways in which facial images differ from each other.
, , , .
, , , . , , . , . ( - , , II, . , , .)
, , , , . . , , , , . , / - .
FR, . , ; , , . , , , , , , , .
EDIT: - . ( , : , RGB- .) , . - , , , - , , .