On the Detection of Image-Scaling Attacks in Machine Learning