Communications in computer and information science, 2017
The Multimedia Forensics community has developed a wide variety of tools for investigating the pr... more The Multimedia Forensics community has developed a wide variety of tools for investigating the processing history of digital videos. One of the main problems, however, is the lack of benchmark datasets allowing to evaluate tools performance on a common reference. In fact, contrarily to the case of image forensics, only a few datasets exist for video forensics, that are limited in size and outdated when compared to today’s real-world scenario (e.g., they contain videos at very low resolution, captured with outdated camcorders, compressed with legacy encoders, etc.). In this paper, we propose a novel dataset made of 622 native videos, most of which in FullHD resolution, captured with 35 different portable devices, belonging to 11 manufacturers and running iOS, Android and Windows Phone OS. Videos have been captured in three different scenarios (indoor, outdoor, flat-field), and with three different kinds of motion (move, still, panrot). Since videos are increasingly shared through social media platforms, we also provide the YouTube version of most videos. Finally, in order to avoid that the proposed dataset becomes outdated in a few moths, we propose a mobile application (MOSES) that allows the acquisition of video contents from recent iOS and Android devices along with their metadata. In this way, the dataset can grow in the future and remain up-to-date.
Multimedia data manipulation and forgery has never been easier than today, thanks to the power of... more Multimedia data manipulation and forgery has never been easier than today, thanks to the power of Artificial Intelligence (AI). AI-generated fake content, commonly called Deepfakes, have been raising new issues and concerns, but also new challenges for the research community. The Deepfake detection task has become widely addressed, but unfortunately, approaches in the literature suffer from generalization issues. In this paper, the Face Deepfake Detection and Reconstruction Challenge is described. Two different tasks were proposed to the participants: (i) creating a Deepfake detector capable of working in an “in the wild” scenario; (ii) creating a method capable of reconstructing original images from Deepfakes. Real images from CelebA and FFHQ and Deepfake images created by StarGAN, StarGAN-v2, StyleGAN, StyleGAN2, AttGAN and GDWCT were collected for the competition. The winning teams were chosen with respect to the highest classification accuracy value (Task I) and “minimum average...
The last decade witnessed a renaissance of machine learning for image processing. Super-resolutio... more The last decade witnessed a renaissance of machine learning for image processing. Super-resolution (SR) is one of the areas where deep learning techniques have achieved impressive results, with a specific focus on the SR of facial images. Examining and comparing facial images is one of the critical activities in forensic video analysis; a compelling question is thus whether recent SR techniques could help face recognition (FR) made by a human operator, especially in the challenging scenario where very low resolution images are available, which is typical of surveillance recordings. This paper addresses such a question through a simple yet insightful experiment: we used two state-of-the-art deep learning-based SR algorithms to enhance some very low-resolution faces of 30 worldwide celebrities. We then asked a heterogeneous group of more than 130 individuals to recognize them and compared the recognition accuracy against the one achieved by presenting a simple bicubic-interpolated ver...
With the increasing diffusion in our life of digital audio and visual contents, the investigation... more With the increasing diffusion in our life of digital audio and visual contents, the investigation on multimedia objects is acquiring more and more interest within the framework of digital investigations, that consider all the aspects including digital data and digital devices. Starting from the international standards and recommendations on the correct procedures to deal with digital evidences and investigations appropriately, and exploiting the Multimedia Forensics technologies, we propose in this paper a possible methodology for correctly investigating audio-visual contents. Going through the steps of the proposed methodology in an illustrative case study, we show the capability of Multimedia Forensics as a tool for understanding the history of multimedia contents presented to the court as potential digital evidence.
... Electronics and Telecommunications University of Florence Italy marco.fontani@gmail.com ... f... more ... Electronics and Telecommunications University of Florence Italy marco.fontani@gmail.com ... from margin is calculated differently for horizontally and vertically oriented images); however, DFB is defined in such a way that RONI blocks will ... Furthermore, it should be possible to de-...
Multimedia Forensics allows to determine whether videos or images have been captured with the sam... more Multimedia Forensics allows to determine whether videos or images have been captured with the same device, and thus, eventually, by the same person. Currently, the most promising technology to achieve this task, exploits the unique traces left by the camera sensor into the visual content. Anyway, image and video source identification are still treated separately from one another. This approach is limited and anachronistic if we consider that most of the visual media are today acquired using smartphones, that capture both images and videos. In this paper we overcome this limitation by exploring a new approach that allows to synergistically exploit images and videos to study the device from which they both come. Indeed, we prove it is possible to identify the source of a digital video by exploiting a reference sensor pattern noise generated from still images taken by the same device of the query video. The proposed method provides comparable or even better performance, when compared t...
Detection of multiple JPEG compression of digital images has been attracting more and more intere... more Detection of multiple JPEG compression of digital images has been attracting more and more interest in the field of multimedia forensics. On the other side, techniques to conceal the traces of multiple compression are being proposed as well. Motivated by a recent trend towards the adoption of universal approaches, we propose a counter-forensic technique that makes multiple compression undetectable for any forensic detector based on the analysis of the histograms of quantized DCT coefficients. Experimental results show the effectiveness of our approach in removing the artifacts of double and also triple compression, while maintaining a good quality of the image.
Communications in computer and information science, 2017
The Multimedia Forensics community has developed a wide variety of tools for investigating the pr... more The Multimedia Forensics community has developed a wide variety of tools for investigating the processing history of digital videos. One of the main problems, however, is the lack of benchmark datasets allowing to evaluate tools performance on a common reference. In fact, contrarily to the case of image forensics, only a few datasets exist for video forensics, that are limited in size and outdated when compared to today’s real-world scenario (e.g., they contain videos at very low resolution, captured with outdated camcorders, compressed with legacy encoders, etc.). In this paper, we propose a novel dataset made of 622 native videos, most of which in FullHD resolution, captured with 35 different portable devices, belonging to 11 manufacturers and running iOS, Android and Windows Phone OS. Videos have been captured in three different scenarios (indoor, outdoor, flat-field), and with three different kinds of motion (move, still, panrot). Since videos are increasingly shared through social media platforms, we also provide the YouTube version of most videos. Finally, in order to avoid that the proposed dataset becomes outdated in a few moths, we propose a mobile application (MOSES) that allows the acquisition of video contents from recent iOS and Android devices along with their metadata. In this way, the dataset can grow in the future and remain up-to-date.
Multimedia data manipulation and forgery has never been easier than today, thanks to the power of... more Multimedia data manipulation and forgery has never been easier than today, thanks to the power of Artificial Intelligence (AI). AI-generated fake content, commonly called Deepfakes, have been raising new issues and concerns, but also new challenges for the research community. The Deepfake detection task has become widely addressed, but unfortunately, approaches in the literature suffer from generalization issues. In this paper, the Face Deepfake Detection and Reconstruction Challenge is described. Two different tasks were proposed to the participants: (i) creating a Deepfake detector capable of working in an “in the wild” scenario; (ii) creating a method capable of reconstructing original images from Deepfakes. Real images from CelebA and FFHQ and Deepfake images created by StarGAN, StarGAN-v2, StyleGAN, StyleGAN2, AttGAN and GDWCT were collected for the competition. The winning teams were chosen with respect to the highest classification accuracy value (Task I) and “minimum average...
The last decade witnessed a renaissance of machine learning for image processing. Super-resolutio... more The last decade witnessed a renaissance of machine learning for image processing. Super-resolution (SR) is one of the areas where deep learning techniques have achieved impressive results, with a specific focus on the SR of facial images. Examining and comparing facial images is one of the critical activities in forensic video analysis; a compelling question is thus whether recent SR techniques could help face recognition (FR) made by a human operator, especially in the challenging scenario where very low resolution images are available, which is typical of surveillance recordings. This paper addresses such a question through a simple yet insightful experiment: we used two state-of-the-art deep learning-based SR algorithms to enhance some very low-resolution faces of 30 worldwide celebrities. We then asked a heterogeneous group of more than 130 individuals to recognize them and compared the recognition accuracy against the one achieved by presenting a simple bicubic-interpolated ver...
With the increasing diffusion in our life of digital audio and visual contents, the investigation... more With the increasing diffusion in our life of digital audio and visual contents, the investigation on multimedia objects is acquiring more and more interest within the framework of digital investigations, that consider all the aspects including digital data and digital devices. Starting from the international standards and recommendations on the correct procedures to deal with digital evidences and investigations appropriately, and exploiting the Multimedia Forensics technologies, we propose in this paper a possible methodology for correctly investigating audio-visual contents. Going through the steps of the proposed methodology in an illustrative case study, we show the capability of Multimedia Forensics as a tool for understanding the history of multimedia contents presented to the court as potential digital evidence.
... Electronics and Telecommunications University of Florence Italy marco.fontani@gmail.com ... f... more ... Electronics and Telecommunications University of Florence Italy marco.fontani@gmail.com ... from margin is calculated differently for horizontally and vertically oriented images); however, DFB is defined in such a way that RONI blocks will ... Furthermore, it should be possible to de-...
Multimedia Forensics allows to determine whether videos or images have been captured with the sam... more Multimedia Forensics allows to determine whether videos or images have been captured with the same device, and thus, eventually, by the same person. Currently, the most promising technology to achieve this task, exploits the unique traces left by the camera sensor into the visual content. Anyway, image and video source identification are still treated separately from one another. This approach is limited and anachronistic if we consider that most of the visual media are today acquired using smartphones, that capture both images and videos. In this paper we overcome this limitation by exploring a new approach that allows to synergistically exploit images and videos to study the device from which they both come. Indeed, we prove it is possible to identify the source of a digital video by exploiting a reference sensor pattern noise generated from still images taken by the same device of the query video. The proposed method provides comparable or even better performance, when compared t...
Detection of multiple JPEG compression of digital images has been attracting more and more intere... more Detection of multiple JPEG compression of digital images has been attracting more and more interest in the field of multimedia forensics. On the other side, techniques to conceal the traces of multiple compression are being proposed as well. Motivated by a recent trend towards the adoption of universal approaches, we propose a counter-forensic technique that makes multiple compression undetectable for any forensic detector based on the analysis of the histograms of quantized DCT coefficients. Experimental results show the effectiveness of our approach in removing the artifacts of double and also triple compression, while maintaining a good quality of the image.
Uploads
Papers by Marco Fontani