Works fine but not all the time. Best practice is to make a Deep Learning model to train with the images.