skip to main content


Title: AI vs. AI: Can AI Detect AI-Generated Images?

The proliferation of Artificial Intelligence (AI) models such as Generative Adversarial Networks (GANs) has shown impressive success in image synthesis. Artificial GAN-based synthesized images have been widely spread over the Internet with the advancement in generating naturalistic and photo-realistic images. This might have the ability to improve content and media; however, it also constitutes a threat with regard to legitimacy, authenticity, and security. Moreover, implementing an automated system that is able to detect and recognize GAN-generated images is significant for image synthesis models as an evaluation tool, regardless of the input modality. To this end, we propose a framework for reliably detecting AI-generated images from real ones through Convolutional Neural Networks (CNNs). First, GAN-generated images were collected based on different tasks and different architectures to help with the generalization. Then, transfer learning was applied. Finally, several Class Activation Maps (CAM) were integrated to determine the discriminative regions that guided the classification model in its decision. Our approach achieved 100% on our dataset, i.e., Real or Synthetic Images (RSI), and a superior performance on other datasets and configurations in terms of its accuracy. Hence, it can be used as an evaluation tool in image generation. Our best detector was a pre-trained EfficientNetB4 fine-tuned on our dataset with a batch size of 64 and an initial learning rate of 0.001 for 20 epochs. Adam was used as an optimizer, and learning rate reduction along with data augmentation were incorporated.

 
more » « less
Award ID(s):
2025234
NSF-PAR ID:
10521804
Author(s) / Creator(s):
;
Publisher / Repository:
Multidisciplinary Digital Publishing Institute
Date Published:
Journal Name:
Journal of Imaging
Volume:
9
Issue:
10
ISSN:
2313-433X
Page Range / eLocation ID:
199
Subject(s) / Keyword(s):
fake and real detection convolutional neural networks
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Building an annotated damage image database is the first step to support AI-assisted hurricane impact analysis. Up to now, annotated datasets for model training are insufficient at a local level despite abundant raw data that have been collected for decades. This paper provides a systematic approach for establishing an annotated hurricane-damaged building image database to support AI-assisted damage assessment and analysis. Optimal rectilinear images were generated from panoramic images collected from Hurricane Harvey, Texas 2017. Then, deep learning models, including Amazon Web Service (AWS) Rekognition and Mask R-CNN (Region Based Convolutional Neural Networks), were retrained on the data to develop a pipeline for building detection and structural component extraction. A web-based dashboard was developed for building data management and processed image visualization along with detected structural components and their damage ratings. The proposed AI-assisted labeling tool and trained models can intelligently and rapidly assist potential users such as hazard researchers, practitioners, and government agencies on natural disaster damage management. 
    more » « less
  2. While active efforts are advancing medical artificial intelligence (AI) model development and clinical translation, safety issues of the AI models emerge, but little research has been done. We perform a study to investigate the behaviors of an AI diagnosis model under adversarial images generated by Generative Adversarial Network (GAN) models and to evaluate the effects on human experts when visually identifying potential adversarial images. Our GAN model makes intentional modifications to the diagnosis-sensitive contents of mammogram images in deep learning-based computer-aided diagnosis (CAD) of breast cancer. In our experiments the adversarial samples fool the AI-CAD model to output a wrong diagnosis on 69.1% of the cases that are initially correctly classified by the AI-CAD model. Five breast imaging radiologists visually identify 29%-71% of the adversarial samples. Our study suggests an imperative need for continuing research on medical AI model’s safety issues and for developing potential defensive solutions against adversarial attacks. 
    more » « less
  3. null (Ed.)
    Artificial Intelligence (AI) techniques such as Generative Neural Networks (GNNs) have resulted in remarkable breakthroughs such as the generation of hyper-realistic images, 3D geometries, and textual data. This work investigates the ability of STEM learners and educators to decipher AI generated video in order to safeguard the public-availability of high-quality online STEM learning content. The COVID-19 pandemic has increased STEM learners’ reliance on online learning content. Consequently, safeguarding the veracity of STEM learning content is critical to ensuring the safety and trust that both STEM educators and learners have in publicly-available STEM learning content. In this study, state of the art AI algorithms are trained on a specific STEM context (e.g., climate change) using publicly-available data. STEM learners are then presented with AI-generated STEM learning content and asked to determine whether the AI-generated output is visually convincing (i.e., “looks real”) and whether the context being presented is plausible. Knowledge gained from this study will help enhance society’s understanding of AI algorithms, their ability to generate convincing video output, and the threat that those generated output have in potentially deceiving STEM learners who may be exposed to them during online learning activities. 
    more » « less
  4. Significant resources have been spent in collecting and storing large and heterogeneous radar datasets during expensive Arctic and Antarctic fieldwork. The vast majority of data available is unlabeled, and the labeling process is both time-consuming and expensive. One possible alternative to the labeling process is the use of synthetically generated data with artificial intelligence. Instead of labeling real images, we can generate synthetic data based on arbitrary labels. In this way, training data can be quickly augmented with additional images. In this research, we evaluated the performance of synthetically generated radar images based on modified cycle-consistent adversarial networks. We conducted several experiments to test the quality of the generated radar imagery. We also tested the quality of a state-of-the-art contour detection algorithm on synthetic data and different combinations of real and synthetic data. Our experiments show that synthetic radar images generated by generative adversarial network (GAN) can be used in combination with real images for data augmentation and training of deep neural networks. However, the synthetic images generated by GANs cannot be used solely for training a neural network (training on synthetic and testing on real) as they cannot simulate all of the radar characteristics such as noise or Doppler effects. To the best of our knowledge, this is the first work in creating radar sounder imagery based on generative adversarial network. 
    more » « less
  5. Purpose: To determine if saliency maps in radiology artificial intelligence (AI) are vulnerable to subtle perturbations of the input, which could potentially lead to misleading interpretations, using Prediction-Saliency Correlation (PSC) for evaluating the sensitivity and robustness of saliency methods. Materials and Methods: In this retrospective study, locally trained deep learning models and a research prototype provided by a commercial vender were systematically evaluated on 191,229 chest radiographs from the CheXpert dataset(1,2) and 7,022 MRI images of human brain tumor classification dataset(3). Two radiologists performed a reader study on 270 chest radiographs pairs. A model-agnostic approach for computing the PSC coefficient was used to evaluate the sensitivity and robustness of seven commonly used saliency methods. Results: Leveraging locally trained model parameters, we revealed the saliency methods’ low sensitivity (maximum PSC = 0.25, 95% CI: 0.12, 0.38) and weak robustness (maximum PSC = 0.12, 95% CI: 0.0, 0.25) on the CheXpert dataset. Without model specifics, we also showed that the saliency maps from a commercial prototype could be irrelevant to the model output (area under the receiver operating characteristic curve dropped by 8.6% without affecting the saliency map). The human observer studies confirmed that is difficult for experts to identify the perturbed images, who had less than 44.8% correctness. Conclusion: Popular saliency methods scored low PSC values on the two datasets of perturbed chest radiographs, indicating weak sensitivity and robustness. The proposed PSC metric provides a valuable quantification tool for validating the trustworthiness of medical AI explainability. Abbreviations: AI = artificial intelligence, PSC = prediction-saliency correlation, AUC = area under the receiver operating characteristic curve, SSIM = structural similarity index measure. Summary: Systematic evaluation of saliency methods through subtle perturbations in chest radiographs and brain MRI images demonstrated low sensitivity and robustness of those methods, warranting caution when using saliency methods that may misrepresent changes in AI model prediction. 
    more » « less