Countering Adversarial Images using Input Transformations
This addresses security vulnerabilities in image classifiers for applications like autonomous vehicles or surveillance, though it is incremental as it builds on existing transformation-based defense methods.
The paper tackles defending image-classification systems against adversarial attacks by applying input transformations like total variance minimization and image quilting, achieving elimination of 60% of strong gray-box and 90% of strong black-box attacks on ImageNet.
This paper investigates strategies that defend against adversarial-example attacks on image-classification systems by transforming the inputs before feeding them to the system. Specifically, we study applying image transformations such as bit-depth reduction, JPEG compression, total variance minimization, and image quilting before feeding the image to a convolutional network classifier. Our experiments on ImageNet show that total variance minimization and image quilting are very effective defenses in practice, in particular, when the network is trained on transformed images. The strength of those defenses lies in their non-differentiable nature and their inherent randomness, which makes it difficult for an adversary to circumvent the defenses. Our best defense eliminates 60% of strong gray-box and 90% of strong black-box attacks by a variety of major attack methods