|
Ruggero
|
Links
 |
 |
 |
Research
The research activity achieved innovative results in different aspects of techniques and systems for intelligent signal and image processing; with the main focus on theoretical aspects, methodologies, technologies, and applications in the contexts of industrial, environmental and biometric applications.
Original adaptive methods (mostly based on deep learnin, artificial neural networks, statistical approaches, and knowledge extraction techniques) permitted to obtain efficient and optimized solutions in intelligent systems for information processing. In this field, he mostly focused on machine learning approaches for designing robust artificial neural networks, and on data analysis methods based on fuzzy logic.
The high robustness of the realized adaptive methods permitted to considerably reduce the operational and environmental constraints, guaranteeing high accuracy at the same time. The research performed in this field permitted to obtain relevant improvements of the human-machine interaction in the context of industrial applications, as well as in biometric recognition systems. In particular, he realized biometric recognition methods able to sensibly reduce the operational constraints with respect to the state of the art, and innovative techniques for the privacy protection of biometric data (based on cryptographic algorithms). These methods can increase the usability and acceptability of the systems in a relevant manner, without decreasing the recognition accuracy, thus representing enabling technologies for high-security applications and ambient intelligence scenarios.
Furthermore, he applied the obtained results to realize innovative solutions for the industrial and environmental monitoring, also in the context of projects financed by the European Commission.
Computational intelligence techniques, architectures, and applications
The research activity regarded computational intelligence methods, with particular attention to the artificial neural networks. In this context, he analyzed different approaches from the theoretical, architectural and applicative points of view, realizing innovative methods for identification, classification, control, prediction, and multidimensional signal processing.
In particular, the realized innovative methods have permitted to create advanced industrial applications for monitoring production processes, environmental control, and quality evaluation.
- Theoretical contributions to the design and optimization of systems
He analyzed intelligent methods for implementing and configuring optimized digital architectures, from the theoretical, architectural, and methodological points of view. In this context, he mainly considered applications based on computational intelligence, as artificial neural networks and fuzzy logic. In particular, he analyzed techniques for classification, approximation, prediction, estimation of three-dimensional models, control, and multidimensional signal processing (one dimensional signals, images, and three-dimensional models). He mainly focused on five topics: 1) adaptive methods for enhancement and feature extraction designed for multidimensional signals; 2) classification methods for particularly noisy signals acquired in non-ideal conditions; 3) measurement methods based on heterogeneous signals acquired using low-accuracy sensors; 4) three-dimensional reconstruction methods based on images acquired using multiple cameras; 5) prediction methods for time series characterized by high level of noise and missing data. - Industrial applications
With the goal of acquiring the experience needed to design innovative and general methods, and to test their efficiency in contexts characterized by particularly noisy data, he analyzed, experimentally evaluated and implemented real application scenarios based on multidimensional signal processing. He applied design methodologies for intelligent systems for the analysis, feature extraction, fusion, and classification of complex data in industrial applications of computer vision. He realized a deep learning approach to detect anomalities in iron production. He studied models for designing bridges. He realized methods to process multidimensional signals collected from heterogeneous sources, like wireless antennas and radars. He realized an innovative and fully automatic system for the real-time classification of wood samples based on the spectral fluorescence analysis. He proposed neural techniques for contactless measuring the volume of heterogeneous objects by analyzing multiple-view images. He realized innovative methods for three-dimensional particle size measurement based on computational intelligence techniques and image processing. He also realized a virtual environment for simulating falling particles, which has been used to design and optimize the final measurement system. Moreover, he used computational intelligence techniques to analyze big sets of data acquired from heterogeneous sources. In particular, he realized expert systems based on fuzzy logic for the analysis of geographical data, and proposed innovative methods based on artificial neural networks for predicting the production of renewable energy. Furthermore, the research activity regarded the automatic classification of cells affected by leukemia in microscopic images of blood samples. In this context, he collaborated in creating a public database of images, which become a reference point for the international research community studying image processing techniques for detecting the presence of cells affected by leukemia. - Environmental applications
He realized innovative algorithms, computational intelligence methods, and simulation techniques for environmental monitoring systems. The main focus consisted of systems for monitoring wildfires based on low-cost cameras.
Biometric systems
The research activity regarded the realization of innovative methods and algorithms for biometric systems applied for verification and identification, considering physiological as well as behavioral traits. He realized original methods for multidimensional signal processing and computational intelligence techniques for touchless biometric systems based on three-dimensional models. He considered the following biometric traits: fingerprint, iris, face, cardiac signals, voice, and soft-biometric traits (e.g., height, weight, and age). He also proposed innovative methods for protecting the privacy of biometric data, continuous identity verification techniques, and methods for improving the accuracy and usability of biometric technologies used in automatic border control systems, in ambient intelligence scenarios, and in forensic applications.
- Touchless fingerprint recognition and 3D-based methods
He realized innovative biometric recognition methods based on the analysis of fingerprint images acquired using touchless techniques. In particular, he realized original hardware and software for high-security systems, which reconstruct and process three-dimensional finger models. He also implemented innovative techniques for extracting sweat pore features, and recognition techniques integrable in mobile devices. He validated the implemented methods using real data as well as innovative virtual environments for simulating three-dimensional fingerprint samples. Furthermore, he proposed quality assessment methods for touchless fingerprint samples.
- Iris recognition in non-ideal conditions
The main focus of the research has been on adaptive algorithms for iris segmentation in ocular images acquired in uncontrolled conditions. He realized innovative methods robust to samples affected by poor illumination, occlusions (as eyelids, eyelashes, hairs, and glasses), gaze deviation with respect to the camera, low resolution, motion blur, and sensor noise. The achieved results permitted to participate to the international competition NICE.I (Noisy Iris Challenge Evaluation), sponsored by Elsevier Science and organized by the Soft Computing and Image Analysis Group (SOCIA Lab), Department of Computer Science, University of Beira Interior, Portugal (http://nicel.di.ubi.pt). The approach was ranked in the top 8 (out of 97), which were invited to submit a paper on the techniques to the special issue on “Iris Images Segmentation” of the “Image and Vision Computing” journal. The paper was subject to regular rigorous review process and was accepted. He also created a dataset of ocular images collected from websites and social media, which represent one of the biggest collections of iris samples coupled by the corresponding manually created segmentation masks.
- Face analysis
The goals of the research activity consisted of designing novel approaches for identifying persons deceased in traumatic conditions and in techniques to extract soft biometric characteristics from face samples. The considered characteristics are the beauty, age, and expression. All the studies based on face images are based on deep neural networks.
- Continuous authentication and recognition of cardiac signals
The main goal consisted of realizing innovative biometric recognition methods based on strong biometric traits acquirable continuously during time. He proposed methods for identity verification, identification, and continuous verification based on electrocardiographic as well as pletismographic signals. He also realized a system for estimating the driver attention level from PPG signals.
- Privacy protection of biometric data
He analyzed different aspects of privacy and security, realizing innovative techniques to perform biometric recognitions in the encrypted domain and approaches for deidentifying the iris characteristics in high-resolution images colled from websites and social media.
- Biometric systems for automated border controls
He analyzed in depth the biometric systems currently used at international border controls with the goal of improving their usability and accuracy. In particular, he realized original adaptive methods to improve the usability and accuracy of fingerprint recognition systems, and algorithms for increasing the accuracy of biometric systems already deployed in real scenarios, which are compliant with the current international regulations and guidelines for privacy protection.
- Biometric systems for ambient intelligence scenarios
With the purpose of increasing the applicability of the biometric technologies in ambient intelligence scenarios, he realized novel methods for the analysis of physiological and behavioral characteristics, which can be used in uncontrolled scenarios characterized by uncooperative users. In particular, he proposed methods for the speaker identification integrable in devices with limited computational resources, methods for the analysis of soft-biometric traits from face images characterized by strong non-idealities, and methods for estimating the height and weight from frame sequences of walking individuals.
Industrial research and development
He realized innovative biometric recognition methods based on the analysis of fingerprint images acquired using touchless techniques. In particular, he realized original hardware and software for high-security systems, which reconstruct and process three-dimensional finger models. He also implemented innovative techniques for extracting sweat pore features, and recognition techniques integrable in mobile devices. He validated the implemented methods using real data as well as innovative virtual environments for simulating three-dimensional fingerprint samples. Furthermore, he proposed quality assessment methods for touchless fingerprint samples.
The main focus of the research has been on adaptive algorithms for iris segmentation in ocular images acquired in uncontrolled conditions. He realized innovative methods robust to samples affected by poor illumination, occlusions (as eyelids, eyelashes, hairs, and glasses), gaze deviation with respect to the camera, low resolution, motion blur, and sensor noise. The achieved results permitted to participate to the international competition NICE.I (Noisy Iris Challenge Evaluation), sponsored by Elsevier Science and organized by the Soft Computing and Image Analysis Group (SOCIA Lab), Department of Computer Science, University of Beira Interior, Portugal (http://nicel.di.ubi.pt). The approach was ranked in the top 8 (out of 97), which were invited to submit a paper on the techniques to the special issue on “Iris Images Segmentation” of the “Image and Vision Computing” journal. The paper was subject to regular rigorous review process and was accepted. He also created a dataset of ocular images collected from websites and social media, which represent one of the biggest collections of iris samples coupled by the corresponding manually created segmentation masks.
The goals of the research activity consisted of designing novel approaches for identifying persons deceased in traumatic conditions and in techniques to extract soft biometric characteristics from face samples. The considered characteristics are the beauty, age, and expression. All the studies based on face images are based on deep neural networks.
The main goal consisted of realizing innovative biometric recognition methods based on strong biometric traits acquirable continuously during time. He proposed methods for identity verification, identification, and continuous verification based on electrocardiographic as well as pletismographic signals. He also realized a system for estimating the driver attention level from PPG signals.
He analyzed different aspects of privacy and security, realizing innovative techniques to perform biometric recognitions in the encrypted domain and approaches for deidentifying the iris characteristics in high-resolution images colled from websites and social media.
He analyzed in depth the biometric systems currently used at international border controls with the goal of improving their usability and accuracy. In particular, he realized original adaptive methods to improve the usability and accuracy of fingerprint recognition systems, and algorithms for increasing the accuracy of biometric systems already deployed in real scenarios, which are compliant with the current international regulations and guidelines for privacy protection.
With the purpose of increasing the applicability of the biometric technologies in ambient intelligence scenarios, he realized novel methods for the analysis of physiological and behavioral characteristics, which can be used in uncontrolled scenarios characterized by uncooperative users. In particular, he proposed methods for the speaker identification integrable in devices with limited computational resources, methods for the analysis of soft-biometric traits from face images characterized by strong non-idealities, and methods for estimating the height and weight from frame sequences of walking individuals.
From June to November 2008, he worked for the company Harding-IT S.R.L., Crema, Italy, performing research activities on micrometric laser measurement systems. He has been involved in the study, design, experimentation, and implementation of advanced algorithms for processing signals acquired by different kinds of sensors in micrometric laser measurement machines, mainly designed for mechanic and advanced manufacturing In this collaboration, he implemented a software for the evaluation of the measurement systems uncertainty, cooperating with Prof. Giulio Barbato of Politecnico di Torino, Italy.