Effectiveness of random deep feature selection for securing image manipulation detectors against adversarial examples.