Dual Thinking and Perceptual Analysis of Deep Learning Models using Human Adversarial Examples