Countering Adversarial Images using Input Transformations

TL;DR

This paper explores input transformation techniques like total variance minimization and image quilting to defend against adversarial attacks on image classifiers, demonstrating high effectiveness especially when combined with training on transformed images.

Contribution

It introduces and evaluates input transformation defenses such as total variance minimization and image quilting, highlighting their robustness due to non-differentiability and randomness.

Findings

Total variance minimization and image quilting are highly effective defenses.

Best defenses eliminate 60% of gray-box and 90% of black-box attacks.

Transforming images during training enhances defense effectiveness.

Abstract

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