On the vulnerability of deep learning to adversarial attacks for camera model identification