Introduction to Professor Alessandro Piva and His Lab at University of Florence
- Could you briefly introduce yourself (and your University/Lab)?
I received the M.S. degree in Electronics Engineering and the Ph.D. degree in Computer Science and Telecommunications Engineering from the University of Florence, in 1995 and 1999, respectively. From 2002 until 2004 I was Research Scientist at the National Inter-university Consortium for Telecommunications (CNIT). Since 2005 until 2014, I was Assistant Professor at the University of Florence, and I’m presently Associate Professor at the Department of Information Engineering of the University of Florence. I’m also Head of FORLAB, the Multimedia Forensics Laboratory, located in Prato, Italy
I’m IEEE Fellow, effective 1 January 2019, for contributions to multimedia security, and IEEE Information Forensics and Security Technical Committee Member.
I’m currently a Senior Area Editor of the Journal of Visual Communication and Image Representation, and Associate Editor of IEEE Transactions on Information Forensics and Security, and General Co-Chair of 13th IEEE International Workshop on Information Forensics and Security (WIFS 2021), December 7 – 10, 2021, Montpellier, France
My research activity has been developed in the framework of both Italian national projects, as well as Projects funded by the European Community and by DARPA. Currently, I’m the Principal investigator of the Project UNCHAINED: UNcovering media manipulation CHains through contAINer and contEnt Detectable traces, funded by DARPA under the Reverse Engineering of Deceptions (RED) programme (2020-2022), and the Person responsible for the University of Florence of the Italian National Project “PREMIER: PREserving Media trustworthiness in the artificial Intelligence ERa” (2019-2022).
I’m one of the members of the Image Analysis Processing & Protection (IAPP) Group of the Department of Information Engineering (DINFO). The research activity carried out by this group involves the analysis, the processing and the protection of multimedia signals, focusing on images and video sequences, in the framework of both Italian regional and national projects, as well as of European projects. IAPP has a long experience in the study, design, and application of multimedia processing and protection tools, namely watermarking, cryptography, signal processing in the encrypted domain, and multimedia forensics.
The staff of IAPP is also involved in the activities of FORLAB, the Multimedia Forensics Laboratory at PIN s.c.r.l Educational and Scientific Services for the University of Florence (PIN is a non-profit organization with majority public share, located in Prato, Italy). FORLAB studies and exploits methodologies and techniques for the analysis and processing of audio-visual data for forensic purposes, focusing on the specific requirements coming from the court of law and investigative scenarios.
2. What have been your most significant research contributions up to now ?
In my research activity, I contributed to the advancement of the fields of watermarking, multimedia security, and information forensics.
I formed and guided one of the most active and worldwide recognized research groups in the field, that, for more than 15 years, contributed to building a solid theoretical framework, and developed a wide set of practical methods bridging the gap between theory and practice.
One of the main contributions lies in the definition of a hypothesis-testing framework going beyond the classical AWGN set-up. His papers on optimum detection of multiplicative watermarks embedded in non-Gaussian features led to significant performance improvement and provided important insights on the achievable performance of this class of watermarking algorithms. The mathematical framework he developed inspired a new line of research in the field and was adapted to several different scenarios as witnessed by the very large number of citations his works have received. This research line led to the release of three patents, one of them extended at the international level. In order to shed light on the perceptual aspects associated with image watermarking, I also contributed an innovative model describing the visibility of a watermark embedded in the wavelet domain, greatly improving the tradeoff between watermark invisibility and robustness.
A more recent contribution consists of a decision fusion framework based on Dempster-Shafer Theory of Evidence, exploiting the information provided by several image forensic tools to reach a global decision about the authenticity of an image. I also contributed to developing practical multimedia forensics techniques, with a particular focus on the modeling of DCT coefficients under double JPEG compression, in case of both aligned and non-aligned recompression. We then extended the model to detect the recompression on MP3 audio files and on MPEG videos.
Several tools developed by my group, namely the methods for forgery localization based on aligned double JPEG, not aligned double JPEG, and forgery localization based on spatial fusion, have been integrated into Amped Authenticate, a software suite by Amped Software, an Italian company providing solutions for analysis and enhancement of images and videos for forensic, security, and investigative applications, which are routinely adopted by the top forensic labs, law enforcement, military, security, and government agencies worldwide.
More recently, we have developed video integrity verification algorithms that exploit the information contained in the video container structure. Such tools, developed by UNIFI, achieved the highest AUC scores in the video manipulations task of MediFor evaluations while being among the least computationally expensive algorithms.
3. What problems in your research field deserve more attention (or what problems will you like to solve) in the next few years, and why ?
Previous works in image and video forensics have studied the detectability of different processing chains, providing interesting results in laboratory condition and well-defined toolchains. Nevertheless, the large diversity of manipulations and parameters that characterize typical real-world scenarios hinder most content-driven approaches and often require the availability of large training datasets, which are rarely available.
Recently, it has been demonstrated that useful insights can be inferred from information related to file formats, as shown by the studies on video container structures carried out by my team within the MediFor programme. The major limitation of pure container-based approaches lies in the fact that although they can identify whether the video has been uploaded onto a social media platform, they cannot reliably reconstruct the previous processing history, for the transcoding operation operated by the uploader which wipes out most of the forensic traces. Similar conclusions can be drawn when metadata-based only approaches are applied to images.
To overcome these limitations, we are trying to exploit the potential of fusing container-based and content-based cues in realistic scenarios where images and videos might undergo mixed toolchains of operations.
4. What advice would you like to give to the young generation of researchers/engineers?
I invite young researchers to work with an open mind. Do not concentrate your work on just one topic or one technology, try as much as possible to be multidisciplinary. Do not trust experimental results if they are not explainable and reproducible.