Understanding cv2.perspectiveTransform implementation

Question

When performing cv2.perspectiveTransform(points, Matrix),

where points is 4 pixels locations, Matrix is homography transform ma