A5043644234- Digital Media Forensic Detection
- Advanced Steganography and Watermarking Techniques
- Generative Adversarial Networks and Image Synthesis
- Law in Society and Culture
- Seismic Imaging and Inversion Techniques
- Music and Audio Processing
- Speech and Audio Processing
- Face recognition and analysis
- Anomaly Detection Techniques and Applications
- Hydraulic Fracturing and Reservoir Analysis
- Advanced Image Processing Techniques
- Reservoir Engineering and Simulation Methods
- Air Quality Monitoring and Forecasting
- Music Technology and Sound Studies
- Advanced machining processes and optimization
- Adversarial Robustness in Machine Learning
- Cell Image Analysis Techniques
- Advanced Surface Polishing Techniques
- Image Processing Techniques and Applications
- Robot Manipulation and Learning
- Seismic Waves and Analysis
- Speech Recognition and Synthesis
- Atmospheric and Environmental Gas Dynamics
- Advanced Image and Video Retrieval Techniques
- Biometric Identification and Security
Politecnico di Milano
2014-2025
Microelectronica (Romania)
2022
University of Siena
2022
In the last few years, several techniques for facial manipulation in videos have been successfully developed and made available to masses (i.e., FaceSwap, deepfake, etc.). These methods enable anyone easily edit faces video sequences with incredibly realistic results a very little effort. Despite usefulness of these tools many fields, if used maliciously, they can significantly bad impact on society (e.g., fake news spreading, cyber bullying through revenge porn). The ability objectively...
A common issue of seismic data analysis consists in the lack regular and densely sampled traces. This problem is commonly tackled by rank optimization or statistical features learning algorithms, which allow interpolation denoising corrupted data. In this paper, we propose a completely novel approach for reconstructing missing traces pre-stack data, taking inspiration from computer vision image processing latest developments. More specifically, exploit specific kind convolutional neural...
Seismic data processing algorithms greatly benefit from regularly sampled and reliable data. Therefore, interpolation denoising play a fundamental role as one of the starting steps most seismic workflows. We exploit convolutional neural networks for joint tasks random noise attenuation 2D common shot gathers. Inspired by great contributions achieved in image computer vision, we investigate particular architecture network referred to U-net, which implements autoencoder able describe complex...
The widespread use of deep learning techniques for creating realistic synthetic media, commonly known as deepfakes, poses a significant threat to individuals, organizations, and society. As the malicious these data could lead unpleasant situations, it is becoming crucial distinguish between authentic fake media. Nonetheless, though deepfake generation systems can create convincing images audio, they may struggle maintain consistency across different modalities, such producing video sequence...
A problem deeply investigated by multimedia forensics researchers is that of detecting which device has been used to capture a video. This enables us trace down the owner video sequence, proves extremely helpful solve copyright infringement cases as well fight distribution illicit material (e.g., child exploitation clips and terroristic threats). Currently, most promising methods tackle this task exploit unique noise traces left camera sensors on acquired images. However, given recent...
Source identification is an important topic in image forensics, since it allows to trace back the origin of image. This represents a precious information claim intellectual property but also reveal authors illicit materials. In this paper we address problem device based on sensor noise and propose fast accurate solution using convolutional neural networks (CNNs). Specifically, 2-channel-based CNN that learns way comparing camera fingerprint at patch level. The proposed turns out be much...
In the last few years, we have witnessed rise of a series deep learning methods to generate synthetic images that look extremely realistic. These techniques prove useful in movie industry and for artistic purposes. However, they also dangerous if used spread fake news or online accounts. For this reason, detecting an image is actual photograph has been synthetically generated becoming urgent necessity. This paper proposes detector based on ensemble Convolutional Neural Networks (CNNs). We...
Abstract Source device identification is an important topic in image forensics since it allows to trace back the origin of image. Its counterpart source anonymization, that is, mask any on can be useful for identifying device. A typical exploited photo response non-uniformity (PRNU), a noise pattern left by acquired images. In this paper, we devise methodology suppressing such from natural images without significant impact quality. Expressly, turn PRNU anonymization into combination global...
The widespread diffusion of synthetically generated content is a serious threat that needs urgent countermeasures. As matter fact, the generation synthetic not restricted to multimedia data like videos, photographs or audio sequences, but covers significantly vast area can include biological images as well, such western blot and microscopic images. In this paper, we focus on detection These are largely explored in biomedical literature it has been already shown they be easily counterfeited...
The rise of AI-generated synthetic media, or deepfakes, has introduced unprecedented opportunities and challenges across various fields, including entertainment, cybersecurity, digital communication. Using advanced frameworks such as Generative Adversarial Networks (GANs) Diffusion Models (DMs), deepfakes are capable producing highly realistic yet fabricated content, while these advancements enable creative innovative applications, they also pose severe ethical, social, security risks due to...
Convolutional Neural Networks (CNNs) have proved very accurate in multiple computer vision image classification tasks that required visual inspection the past (e.g., object recognition, face detection, etc.). Motivated by these astonishing results, researchers also started using CNNs to cope with forensic problems camera model identification, tampering However, vision, methods typically rely on cues easily detectable human eyes. Conversely, solutions almost invisible traces are often subtle...
Identifying the source camera of images and videos has gained significant importance in multimedia forensics. It allows tracing back data to their creator, thus enabling solve copyright infringement cases expose authors hideous crimes. In this paper, we focus on problem model identification for video sequences, that is, given a under analysis, detecting used its acquisition. To purpose, develop two different CNN-based methods, working novel multi-modal scenario. Differently from mono-modal...
In the last few years, forensic researchers have developed a wide set of techniques to blindly attribute an image device used shoot it. Among these techniques, those based on photo response non uniformity (PRNU) shown incredibly accurate results, thus they are often considered as reference baseline solution. The rationale behind is that each camera sensor leaves acquired images characteristic noise pattern. This pattern can be estimated and uniquely mapped specific acquisition through...
In recent years, the possibility of easily editing video sequences led to diffusion user generated compilations obtained by splicing together in time different shots. order perform forensic analysis on this kind videos, it can be useful split whole sequence into set originating As shots are seldom with a single device, possible way identify each shot is exploit sensor-based traces. State-of-the-art solutions for sensor attribution rely Photo Response Non Uniformity (PRNU). Despite approach...
Due to the increasingly unbridled practice of sharing visual content on web, tracing back past history uploaded images is getting far from being an easy task. Nonetheless, forensic analysts might be interested in probing digital published web assess its authenticity. In this vein, a possible indicator image integrity number JPEG compressions picture underwent. As matter fact, compression typically operated first at inception time directly acquisition device. Then, it customary re-applied...
This paper presents an extensive evaluation of the Deep Image Prior (DIP) technique for image inpainting on Synthetic Aperture Radar (SAR) images. SAR images are gaining popularity in various applications, but there may be a need to conceal certain regions them. provides solution this. However, not all techniques designed work Some intended use photographs, while others have specifically trained top huge set In this work, we evaluate performance DIP that is capable addressing these...
AI-generated synthetic media, also called Deepfakes, have significantly influenced so many domains, from entertainment to cybersecurity. Generative Adversarial Networks (GANs) and Diffusion Models (DMs) are the main frameworks used create producing highly realistic yet fabricated content. While these technologies open up new creative possibilities, they bring substantial ethical security risks due their potential misuse. The rise of such advanced media has led development a cognitive bias...
Over the years, forensic community has developed a series of very accurate camera attribution algorithms enabling to detect which device been used acquire an image with outstanding results. Many these methods are based on photo response non uniformity (PRNU) that allows tracing back picture shoot it. However, when privacy is required, it would be desirable anonymize photos, unlinking them from their specific device. This paper investigates new and alternative approach anonymization task. The...
To decide whether a digital video has been captured by given device, multimedia forensic tools usually exploit characteristic noise traces left the camera sensor on acquired frames. This analysis requires that pattern characterizing and extracted from frames under are geometrically aligned. However, in many practical scenarios this does not occur, thus re-alignment or synchronization to be performed. Current solutions often require time consuming search of realignment transformation...
Synthetic Aperture Radar (SAR) images are a valuable asset for wide variety of tasks. In the last few years, many websites have been offering them free in form easy to manage products, favoring their widespread diffusion and research work SAR field. The drawback these opportunities is that such might be exposed forgeries manipulations by malicious users, raising new concerns about integrity trustworthiness. Up now, multimedia forensics literature has proposed various techniques localize...
Our scientific community faces a sort of paradox. A large bulk work has been done on data-oriented techniques devised to improve peer reputation and knowledge extraction from data, so as trustworthiness digital services involving coordination cooperation among heterogeneous peers. But, perhaps surprisingly, the best our knowledge, such have rarely applied (for own community, crucial) process reducing noise in reviewing papers. Goal this is provide initial insights applicability methodologies...
The widespread diffusion of user friendly editing software for audio signals has made tampering extremely accessible to anyone. Therefore, it is increasingly necessary develop forensic methodologies aiming at verifying if a given content been digitally manipulated or not. Among the multiple available techniques, very common one time scaling, i.e., altering temporal evolution an signal without affecting any pitch component. For instance, this can be used slow-down speed-up speech recordings,...