Detecting GAN generated Fake Images using Co-occurrence Matrices

View on arXiv ← Back to list

Authors: Lakshmanan Nataraj, Tajuddin Manhar Mohammed, Shivkumar Chandrasekaran, Arjuna Flenner, Jawadul H. Bappy, Amit K. Roy-Chowdhury, B. S. Manjunath

Published: 2019-03-15 23:24:08+00:00

AI Summary

This paper proposes a novel approach for detecting GAN-generated fake images using co-occurrence matrices and a deep convolutional neural network (CNN). The method extracts co-occurrence matrices from image color channels and trains a CNN to classify real and fake images, achieving over 99% accuracy on two diverse datasets.

Abstract

The advent of Generative Adversarial Networks (GANs) has brought about completely novel ways of transforming and manipulating pixels in digital images. GAN based techniques such as Image-to-Image translations, DeepFakes, and other automated methods have become increasingly popular in creating fake images. In this paper, we propose a novel approach to detect GAN generated fake images using a combination of co-occurrence matrices and deep learning. We extract co-occurrence matrices on three color channels in the pixel domain and train a model using a deep convolutional neural network (CNN) framework. Experimental results on two diverse and challenging GAN datasets comprising more than 56,000 images based on unpaired image-to-image translations (cycleGAN [1]) and facial attributes/expressions (StarGAN [2]) show that our approach is promising and achieves more than 99% classification accuracy in both datasets. Further, our approach also generalizes well and achieves good results when trained on one dataset and tested on the other.

Key findings

The proposed method achieves over 99% accuracy in detecting GAN-generated images on both datasets. The approach also demonstrates good generalization ability, maintaining high accuracy when trained on one dataset and tested on the other. Performance degrades with JPEG compression but improves when training on compressed data.

Approach

The approach uses co-occurrence matrices computed directly from the RGB channels of an image, bypassing the need for image residuals or filtering. These matrices are then fed into a deep CNN for training and classification of real versus fake images.

Datasets

CycleGAN dataset (36,302 images: 18,151 GAN-generated and 18,151 real images) and StarGAN dataset (19,990 images: 17,991 GAN-generated and 1999 real images).

Model(s)

A deep convolutional neural network (CNN) with multiple convolutional and max pooling layers, followed by fully connected layers and a sigmoid layer for binary classification.

Author countries

USA, India, Bangladesh

← Previous