Distinguishing Computer-Generated Graphics from Natural Images Based on Sensor Pattern Noise and Deep Learning

Abstract

Computer-generated graphics (CGs) are images generated by computer software. The rapid development of computer graphics technologies has made it easier to generate photorealistic computer graphics, and these graphics are quite difficult to distinguish from natural images (NIs) with the naked eye. In this paper, we propose a method based on sensor pattern noise (SPN) and deep learning to distinguish CGs from NIs. Before being fed into our convolutional neural network (CNN)-based model, these images—CGs and NIs—are clipped into image patches. Furthermore, three high-pass filters (HPFs) are used to remove low-frequency signals, which represent the image content. These filters are also used to reveal the residual signal as well as SPN introduced by the digital camera device. Different from the traditional methods of distinguishing CGs from NIs, the proposed method utilizes a five-layer CNN to classify the input image patches. Based on the classification results of the image patches, we deploy a majority vote scheme to obtain the classification results for the full-size images. The experiments have demonstrated that (1) the proposed method with three HPFs can achieve better results than that with only one HPF or no HPF and that (2) the proposed method with three HPFs achieves 100% accuracy, although the NIs undergo a JPEG compression with a quality factor of 75.

Learn with diversity and from harder samples: Improving the generalization of CNN-Based detection of computer-generated images

“…For the CG image forensic problem, there mainly exist two types of methods: hand-crafted-feature-based methods [6,7,8,9,10,11,12] and CNN-based methods [13,14,15,16,17,18,19].…”

“…Inspired by the notable success of CNN in the field of computer vision and pattern recognition, some recent works also applied CNN to solve the CG image forensic problem [13,14,15,16,17,18,19]. Rahmouni et al [13] explicitly extracted low-order statistical information of convoluted image as discriminative features and trained a model to distinguish computer graphics from photographic images.…”

“…They also tried to understand what the deep model has learned about the differences between these two kinds of images by using a number of CNN visualization tools. Yao et al [15] proposed a method based on sensor pattern noise and CNN to solve this task. In their method, they used several high-pass filters to enhance the residual signal as well as sensor pattern noise introduced by the digital camera device.…”

“…In the past decades, researchers have proposed hand-crafted-feature-based methods [6,7,8,9,10,11,12] and CNN(convolutional neural network)-based methods [13,14,15,16,17,18,19] to solve this forensic problem. The former usually consists of multidimensional feature extraction and classifier training, and the latter directly formulates an optimization problem to find a good mapping function from a given image to its label and solves it efficiently in an end-to-end manner.…”

Distinguish computer generated and digital images: A CNN solution

“…Yao et al proposed a method based on sensor pattern noise (SPN) and deep learning to distinguish CGs from NIs. Rahmouni et al proposed a custom pooling layer in CNN to optimize current best‐performing algorithms feature extraction scheme, then local estimates of class probabilities are computed and aggregated to predict the label of the whole picture.…”