Graphs are a rich data structure allowing to encode efficiently both sub parts of objects and the relationships between these sub parts. At a first glance, graphs seems to be a promising model for the classification of complex objects. Unfortunately, there is no free lunch and graphs charge for this flexibility in several ways: Firstly most of (useful) graph algorithms are either NP-complete or NP-hard. Secondly, metrics defined on graph spaces induce specific properties on these spaces which should be manipulated with care. In this talk we will review three families of graph metrics together with the properties of the associated spaces. We will also adopt a slightly different point of view and describe recent advances on Graph neural networks.

Biometrics and video surveillance are ubiquitous. Fingerprint and face verification on smartphones has brought biometrics to the masses. Automated video surveillance is deployed in public spaces for anomalous event and behaviour detection. Biometrics, and to a lesser extent video surveillance, is also integral to border security. Automated Border Control (ABC) eGates enable automated clearance for low-risk travellers. While such systems have been transformative and have many benefits over manual controls there exist many challenges. The talk will begin by discussing these issues and determining an overall vision of border security of the future. Next, the outcomes of an EU FastPass project, which aimed to develop a harmonized approach to eGates incorporating both innovative development and application of biometrics and video surveillance, will be presented. The talk will then advance the vision by proposing contactless biometric-based, free-flowing on-the-move border control systems taking into account privacy and security issues which such innovation raises. In this context, the recently completed EC PROTECT project will be presented including the novel biometric and surveillance concepts and technical solutions that were developed, including results of two demonstrations addressing travellers in vehicles as well as on foot on-the-move. Finally, the talk will extend the vision to consider how the PROTECT work fits within the overall concept of so called “no-gate crossing solutions” and why the whole identity lifecycle of a traveller needs to be addressed.

More than 15 years ago, the research community claimed that fingerprints «are very difficult to reproduce and steal». We lost our “innocence” when, between 2001 and 2002, some scholars fabricated artificial replicas of the fingers, named “fake fingers” or “gummy fingers”, that, if put on the fingerprint sensor surface, provided images impossible to distinguish from those of the live fingers, even by visual inspection of experts. This led to a plethora of countermeasures reported in journals, conference proceedings and international projects: generally speaking, pattern recognition and machine learning-based approaches, where the main problem was to find the best feature set able to describe the differences between live and fake fingerprints. Since the basic problem was the lack of data, the organization of the International Fingerprint Liveness Detection Competition, known as “LivDet”, helped our research group to acquire a strong know-how on the difficulties in fabricating fake fingers and the effectiveness of the state-of-the-art algorithms. In the 2019 edition, we tested algorithms integrating presentation attacks detectors and verification systems. It was the first time that liveness detection was handle as a part of a more complex problem, that is, the design of an “intrinsically secure” fingerprint verification system. Therefore, what is the state-of-the-art? Where are we? In this lecture, we summarize our experience on the problem and trace the path from the first fingerprint liveness detection algorithms to the most recent approaches. In particular, the “battle” between the handcrafted features and the deep learning- based paradigms which characterized the last four years of research, is pointed out. On the basis of what reported by the scientific state-of-the-art and the LivDet results over years, we show that the word “end” cannot be still said on this challenging topic.

Download the presentation here.

In order to be effective, traditional pattern recognition methods typically require a careful manual design of features, involving considerable domain knowledge and effort by experts. The recent popularity of deep learning is largely due to the automatic configuration of effective early and intermediate representations of the data presented. The downside of deep learning is that it requires a huge number of training examples.

Trainable COSFIRE filters are an alternative to deep networks for the extraction of effective representations of data. COSFIRE stands for Combinations of Shifted Filter Responses. Their design was inspired by the function of certain shape selective neurons in areas V4 and TEO of visual cortex. A COSFIE filter is configured by the automatic analysis of a single attern. The highly non-linear filter response is computed as a combination of the responses of simpler filters, such as Difference of (color) Gaussians or Gabor filters, taken at different positions of the concerned pattern. The identification of the parameters of the simpler filters that are needed and the positions at which their responses are taken is done automatically. An advantage of this approach is its ease of use as it requires no programming effort and little computation – the parameters of a filter are derived automatically from a single training pattern. Hence, a large number of such filters can be configured effortlessly and selected responses can be arranged in feature vectors that are fed into a traditional classifier.

This approach is illustrated by the automatic configuration of COSFIRE filters that respond to randomly selected parts of many handwritten digits. We configure automatically up to 5000 such filters and use their maximum responses to a given image of a handwritten digit to form a feature vector that is fed to a classifier. The COSFIRE approach is further illustrated by the detection and identification of traffic signs and of sounds of interest in audio signals.

The COSFIRE approach to representation learning and classification yields performance results that are comparable to the best results obtained with deep networks but at a much smaller computational effort. Notably, COSFIRE representations can be obtained using numbers of training examples that are many orders of magnitude smaller than those used by deep networks.