Adversarial Images Against Super-Resolution Convolutional Neural Networks for Free

Super-Resolution Convolutional Neural Networks (SRCNNs) with their ability to generate highresolution images from low-resolution counterparts, exacerbate the privacy concerns emerging automated (CNNs)-based image classifiers. In this work, we hypothesize and empirically show that adversarial examples learned over CNN classifiers can survive processing by SRCNNs lead them poor quality are hard classify correctly. We demonstrate a user small is able learn noise without requiring any customization for thwart threat posed pipeline of SRCNN (95.8% fooling rate Fast Gradient Sign ε = 0.03). evaluate survivability generated in both black-box white-box settings learning (when classifier unknown) at least as effective only known). also assess our hypothesis on robust CNNs observe supper-resolved fool these more than 71.5% time.