Conditional Adversarial Camera Model Anonymization

Abstract

The model of camera that was used to capture a particular photographic image (model attribution) is typically inferred from high-frequency model-specific artifacts present within the image. Model anonymization is the process of transforming these artifacts such that the apparent capture model is changed. We propose a conditional adversarial approach for learning such transformations. In contrast to previous works, we cast model anonymization as the process of transforming both high and low spatial frequency information. We augment the objective with the loss from a pre-trained dual-stream model attribution classifier, which constrains the generative network to transform the full range of artifacts. Quantitative comparisons demonstrate the efficacy of our framework in a restrictive non-interactive black-box setting.

arXiv commentECCV 2020 - Advances in Image Manipulation workshop (AIM 2020)

References in corpus (19)

• Adam: A Method for Stochastic Optimization
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Conditional Generative Adversarial Nets
• Intriguing properties of neural networks
• Generative Adversarial Text to Image Synthesis
• Shortcut Learning in Deep Neural Networks
• Conditional Image Synthesis With Auxiliary Classifier GANs
• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Toward Multimodal Image-to-Image Translation
• The Perception-Distortion Tradeoff
• One weird trick for parallelizing convolutional neural networks
• The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
• Invertible Conditional GANs for image editing
• First Steps Toward Camera Model Identification with Convolutional Neural Networks
• Disentangling factors of variation in deep representations using adversarial training
• A Counter-Forensic Method for CNN-Based Camera Model Identification
• Object Discovery with a Copy-Pasting GAN
• Age Progression and Regression with Spatial Attention Modules