Statistics for A Watermarking-Based Framework for Protecting Deep Image Classifiers Against Adversarial Attacks