I am assuming that you are using the STEP method of VISION.CASCADEOBJECTDETECTOR that returns a bounding box.
You could just use the bounding box data to crop out the face from the image.
For example, if I is the RGB image and bounding box is [x,y,w,h], you could do:
croppedFace = I(x:x+w, y-h:y,:);
