face alignment for face recognition

Question

I have a problem to align faces using opencv. I detect the face then I use flandmarks and calculate the angle of rotation. I use this function to rotate the image :

cv::Mat rotate(cv::Mat src, double angle)
{
    cv::Mat dst;
    cv::Point2f pt(src.cols/2., src.rows/2.);
    cv::Mat r = getRotationMatrix2D(pt, angle, 1.0);
    cv::warpAffine(src, dst, r, cv::Size(src.cols, src.rows));
    return dst;
}

I want to find the new position of eyes after rotation to crop the face based on the eyes center.

Are you performing eye detection BEFORE alignment and going to find that corresponding region after rotation, or are you going to just perform eye detection AFTER rotation? — , Jul 23 '15 at 11:50
I detect the eyes then rotate based on the eyes position ... after that I want to find the new position of eyes to crop the face — Tyranitar, Jul 23 '15 at 12:03

score 0 · Accepted Answer · edited Apr 13 '17 at 12:19

For rotating(counterclockwise by an angle θ) a point (x,y) around another point (p,q) you need to use:

x′ = (x−p)cos(θ)−(y−q)sin(θ)+p,
y′ = (x−p)sin(θ)+(y−q)cos(θ)+q.

where x',y' are coordinates after rotation. In your case (p,q) is the center of the image if you rotated around center. Detailed Explanation could be found here: https://math.stackexchange.com/questions/270194/how-to-find-the-vertices-angle-after-rotation

So if you have eyes segmented as some areas you need to perform that operation on each pixel from area detected as eyes.

face alignment for face recognition

1 Answers1