Quantifying Perceptual Distortion of Adversarial Examples