Search

Convolutional Neural Networks (CNNs) are opening for unprecedented scenarios in fields where designing effective features is tedious even for domain experts. This is the case of medical imaging, i.e. procedures acquiring images of a human body interior for clinical proposes. Despite promising, we argue that CNNs naive use may not be effective since “medical images are more than pictures”. A notable example is breast Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI), in which the kinetic of the injected Contrast Agent (CA) is crucial for lesion classification purposes. Therefore, in this work we introduce a new GAN like approach designed to simultaneously learn how to disentangle the CA effects from all the other image components while performing the lesion classification: the generator is an intrinsic Deforming Autoencoder (DAE), while the discriminator is a CNN. We compared the performance of the proposed approach against some literature proposals (both classical and CNN based) using patient-wise cross-validation. Finally, for the sake of completeness, we also analyzed the impact of variations in some key aspect of the proposed solution. Results not only show the effectiveness of our approach ( + 8 % AUC w.r.t. the runner-up) but also confirm that all the approach’s components effectively contribute to the solution.

Some applications have the property of being resilient, meaning that they are robust to noise (e.g. due to error) in the data. This characteristic is very useful in situations where an approximate computation allows to perform the task in less time or to deploy the algorithm on embedded hardware. Deep learning is one of the fields that can benefit from approximate computing to reduce the high number of involved parameters thanks to its impressive generalization ability. A common approach is to prune some neurons and perform an iterative re-training with the aim of both reducing the required memory and to speed-up the inference stage. In this work we propose to face CNN size reduction from a different perspective: instead of reducing the network weights or look for an approximated network very close to the Pareto frontier, we investigate whether it is possible to remove some neurons only from the fully connected layers before the network training without substantially affecting the network performance. As a case study, we will focus on “fine-tuning”, a branch of transfer learning that has shown its effectiveness especially in domains lacking effective expert-designed features. To further compact the network, we apply weight quantization to the convolutional kernels. Results show that it is possible to tailor some layers to reduce the network size, both in terms of the number of parameters to learn and required memory, without statistically affecting the performance and without the need for any additional training. Finally, we investigate to what extent the sizing operation affects the network robustness against adversarial perturbations, a set of approaches aimed at misleading deep neural networks.

Background and objective The indirect immunofluorescence (IIF) on HEp-2 cells is the recommended technique for the detection of antinuclear antibodies. However, it is burdened by some limitations, as it is time consuming and subjective, and it requires trained personnel. In other fields the adoption of deep neural networks has provided an effective high-level abstraction of the raw data, resulting in the ability to automatically generate optimized high-level features. Methods To alleviate IIF limitations, this paper presents a computer-aided diagnosis (CAD) system classifying HEp-2 fluorescence intensity: it represents each image using an Invariant Scattering Convolutional Network (Scatnet), which is locally translation invariant and stable to deformations, a characteristic useful in case of HEp-2 samples. To cope with the inter-observer discrepancies found in the dataset, we also introduce a method for gold standard computation that assigns a label and a reliability score to each HEp-2 sample on the basis of annotations provided by expert physicians. Features by Scatnet and gold standard information are then used to train a Support Vector Machine. Results The proposed CAD is tested on a new dataset of 1771 images annotated by three independent medical centers. The performances achieved by our CAD in recognizing positive, weak positive and negative samples are also compared against those obtained by other two approaches presented so far in the literature. The same system trained on this new dataset is then tested on two public datasets, namely MIVIA and I3Asel. Conclusions The results confirm the effectiveness of our proposal, also revealing that it achieves the same performance as medical experts.

Cancer is one of the leading causes of death in the western world, with medical imaging playing a key role for early diagnosis. Focusing on breast cancer, one of the emerging imaging methodologies is Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI). The flip side of using DCE-MRI is in its long acquisition times that often causes the patient to move. This results in motion artefacts, namely distortions in the acquired image that can affect DCE-MRI analysis. A possible solution consists in the use of Motion Correction Techniques (MCTs), i.e. procedures intended to re-align the post-contrast image to the corresponding pre-contrast (reference) one. This task is particularly critic in DCE-MRI, due to brightness variations introduced in post-contrast images by the contrast-agent flowing. To face this problem, in this work we introduce a new MCT for breast DCE-MRI leveraging Physiologically Based PharmacoKinetic (PBPK) modelling and Artificial Neural Networks (ANN) to determine the most suitable physiologically-compliant transformation. To this aim, we propose a Neural Registration Network relying on a very task-specific loss function explicitly designed to take into account the contrast agent flowing while enforcing a correct re-alignment. We compared the obtained results against some conventional motion correction techniques, evaluating the performance on a patient-by-patient basis. Results show that the proposed approach results to be the best performing even when compared against other techniques designed to take into account for brightness variations.

The growing availability of cheap and reliable fingerprint acquisition scanners is resulting in an increasing spread of Fingerprint-based Authentication Systems (FAS) in consumer electronics. This has giving rise to a new wave in research on both smarter spoofing attacks, aimed to bypass a FAS by using a counterfeit fingerprint, and on more effective Liveness Detectors (LD), aimed to discern authentic (live) fingerprints from fake ones. As in many other computer vision tasks, deep Convolutional Neural Networks (CNN) demonstrated to be very effective also for fingerprint liveness detection. However, we showed that it is possible to adapt adversarial perturbation approaches to mislead CNN-based LD. In this paper, we want to make a step further toward the design of a black-box attack by investigating whether it is possible to transfer a perturbation across different CNN liveness detectors in the case of a target LD very different from the one used to compute the perturbations. To this aim, we designed an attack scenario where a shadow LD (i.e. an adaptation of the substitute technique for the liveness detection application) is used to generate an adversarial fingerprint in a white-box setting before submitting it to the real target LD, invoked in a total back-box manner. Finally, we analysed the impact that such attack has on the authentication system, also analysing if and to what extent the scanner and the spoofing material combinations affect the success of the attack.

Lung ultrasound imaging is reaching growing interest from the scientific community. On one side, thanks to its harmlessness and high descriptive power, this kind of diagnostic imaging has been largely adopted in sensitive applications, like the diagnosis and follow-up of preterm newborns in neonatal intensive care units. On the other side, state-of-the-art image analysis and pattern recognition approaches have recently proven their ability to fully exploit the rich information contained in these data, making them attractive for the research community. In this work, we present a thorough analysis of recent deep learning networks and training strategies carried out on a vast and challenging multicenter dataset comprising 87 patients with different diseases and gestational ages. These approaches are employed to assess the lung respiratory status from ultrasound images and are evaluated against a reference marker. The conducted analysis sheds some light on this problem by showing the critical points that can mislead the training procedure and proposes some adaptations to the specific data and task. The achieved results sensibly outperform those obtained by a previous work, which is based on textural features, and narrow the gap with the visual score predicted by the human experts., Comment: 7 pages

The exponential growth of connected devices (i.e. laptops, smartphones or tablets) has radically changed communications means, also making it faster and impersonal by using On-line Social Networks and Instant messaging through several apps. In this paper we discuss about the cyberbullying problem, focusing on the analysis of the state-of-the-art approaches that can be classified in four different tasks (Binary Classification, Role Identification, Severity Score Computation and Incident prediction). In particular, the first task aims to predict if a particular action is aggressive or not based on the analysis of different features. In turn, the second and the third task investigate the cyberbullying problem by identifying users’ role in the exchanged message or assigning a severity score to a given users or session respectively. Nevertheless, information heterogeneity, due to different multimedia contents (i.e. text, emojis, stickers or gifs), and the use of datasets, which are typically unlabeled or manually labelled, create continuous challenges in addressing the cyberbullying problem.

With the growing number of digital medical imaging records, the need for an automatic procedure to retrieve only data of interest is of increasing importance. A Picture Archiving and Communication System (PACS) provides effective storage and retrieval based on TAGs but does not allow us for query by example. A possible solution is to use a Content-Based Image Retrieval (CBIR) system, namely a system able to retrieve images from a database based on the similarity to a given reference image. The features used to describe the images strongly affect both the performance and the applicability of CBIR to medical images, motivating for the finding of a suitable set of feature for realizing an effective CBIR based PACS. In recent years, Deep Learning (DL) approaches outperformed classical machine learning methods in many computer vision applications, thanks to their ability to learn compact hierarchical features of input data that well fit the specific task to solve. In this paper we introduce a simple yet effective modular architecture to implement a “Smart PACS”, namely a PACS exploiting a deep-based CBIR compatible with the classical Hospital Information System (HIS) infrastructure. The feature extraction relies on Convolutional Neural Networks, a DL approach commonly applied in image processing, while the image indexing and look-up are based on Apache Solr. As application case-study, we analysed the need for a physician to obtain all the images of past studies having similar traits with the patient under analysis.

We address the problem of clustering a set of images, according to their source device, in the absence of any prior information. Image similarity is computed based on noise residuals, regarded as single-image estimates of the camera’s photo-response non-uniformity (PRNU) pattern. First, residuals are grouped by correlation clustering, and several alternative data partitions are computed as a function of a running decision boundary. Then, these partitions are processed jointly to extract a single, more reliable, consensus clustering and, with it, more reliable PRNU estimates. Finally, both clustering and PRNU estimates are progressively refined by merging pairs of the same- PRNU clusters, selected on the basis of a maximum-likelihood ratio statistic. Extensive experiments prove the proposed method to outperform the current state of the art both on pristine images and compressed images downloaded from social networks. A remarkable feature of the method is that it does not require the user to set any parameter, nor to provide a training set to estimate them. Moreover, through a suitable choice of basic tools, and efficient implementation, complexity remains always quite limited.

We perform a systematic review of ECG biometric databases in the literature.We compare the acquisition hardware used to acquire ECG biometric databases.We compare the acquisition protocols adopted by ECG biometric databases.We highlight that an exhaustive database for biometric studies still lacks. Computer-based biometric systems (CBBSs) individual recognition are expert and intelligent systems that are gaining increasing interest in many areas, such as securing financial systems, telecommunications and healthcare applications. The electrocardiogram (ECG) has been used as biometric feature for its low circumvention, large acceptability and uniqueness, thus being at the basis of several CBBSs. As ECG databases collected for clinical applications are not adequate for biometric applications, we have assisted to the development of other repositories of ECG, each one different from the others and highlighting certain issues of ECG-based biometric recognition. Through a systematic framework presented here, we quantitative analyse, evaluate and compare the acquisition hardware and the acquisition protocols of ECG databases available in literature and suited to develop CBBSs. Although the most recent ones, namely CYBHI and UofTDB, result the best for the acquisition hardware and the acquisition protocols, respectively, our survey shows that none is exhaustive for developing a robust and general enough CBBSs. The analysis also highlights the current lack of standardization in this field and the difficulty of performing an effective benchmarking activity. Since a publicly available database is essential for the research community in ECG-based CBBS to correctly assess the performance of existing algorithms or even commercial expert systems, we also discuss here the main features that an "optimal" repository for the intelligent application at hand.

With the increasing spread of easy and effective frameworks, in recent years Deep Learning approaches are becoming more and more used in several application fields, including computer vision (such as natural and biomedical image processing), automatic speech recognition (ASR) and time-series analysis. If, on one hand, the availability of such frameworks allows developers to use the one they feel more comfortable with, on the other, it raises questions related to the reproducibility of the designed model across different hardware and software configurations, both at training and at inference times. The reproducibility assessment is important to determine if the resulting model produces good or bad outcomes just because of luckier or blunter environmental training conditions. This is a non-trivial problem for Deep Learning based applications, not only because their training and optimization phases strongly rely on stochastic procedures, but also because of the use of some heuristic considerations (mainly speculative procedures) at training time that, although they help in reducing the required computational effort, tend to introduce non-deterministic behavior, with a direct impact on the results and on the model’s reproducibility. Usually, to face this problem, designers make use of probabilistic considerations about the distribution of data or focus their attention on very huge datasets. However, this kind of approach does not really fit some application field standards (such as medical imaging analysis with Computer-Aided Detection and Diagnosis systems – CAD) that require strong demonstrable proofs of effectiveness and repeatability of results across the population. It is our opinion that in those cases it is of crucial importance to clarify if and to what extent a Deep Learning based application is stable and repeatable as well as effective, across different environmental (hardware and software) configurations. Therefore, the aim of this work is to quantitatively analyze the reproducibility problem of Convolutional Neural Networks (CNN) based approaches for the biomedical image processing, in order to highlight the impact that a given software framework and hardware configurations might have when facing the same problem by the same means. In particular, we analyzed the problem of breast tissue segmentation in DCE-MRI by using a modified version of a 2D U-Net CNN, a very effective deep architecture for semantic segmentation, using two Deep Learning frameworks (MATLAB and TensorFlow) across different hardware configurations.

The use of biometric-based authentication systems spread over daily life consumer electronics. Over the years, researchers’ interest shifted from hard (such as fingerprints, voice and keystroke dynamics) to soft biometrics (such as age, ethnicity and gender), mainly by using the latter to improve the authentication systems effectiveness. While newer approaches are constantly being proposed by domain experts, in the last years Deep Learning has raised in many computer vision tasks, also becoming the current state-of-art for several biometric approaches. However, since the automatic processing of data rich in sensitive information could expose users to privacy threats associated to their unfair use (i.e. gender or ethnicity), in the last years researchers started to focus on the development of defensive strategies in the view of a more secure and private AI. The aim of this work is to exploit Adversarial Perturbation, namely approaches able to mislead state-of-the-art CNNs by injecting a suitable small perturbation over the input image, to protect subjects against unwanted soft biometrics-based identification by automatic means. In particular, since ethnicity is one of the most critical soft biometrics, as a case of study we will focus on the generation of adversarial stickers that, once printed, can hide subjects ethnicity in a real-world scenario.

Fingerprint-based Authentication Systems (FAS) usage is increasing over the last years thanks to the growing availability of cheap and reliable scanners. In order to bypass a FAS by using a counterfeit fingerprint, a Presentation Attack (PA) can be used. As a consequence, a liveness detector able to discern authentic from fake biometry becomes almost essential in each FAS. Deep Learning based approaches demonstrated to be very effective against fingerprint presentation attacks, becoming the current state-of-the-art in liveness detection. However, it has been shown that it is possible to arbitrarily cause state-of-the-art CNNs to misclassify an image by applying on it a suitable small peturbation, often even imperceptible to human eyes. The aim of this work is to understand if and to what extent adversarial perturbation can affect FASs, as a preliminary step to develop an adversarial presentation attack. Results show that it is possible to exploit adversarial perturbation to mislead both the FAS liveness detector and the authentication system, by giving rise to images that are even almost imperceptible to human eyes.

Nowadays, Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) is increasingly succeeding as a complementary methodology for breast cancer, with Computer Aided Detection/Diagnosis (CAD) systems becoming essential technological tools to provide early detection and diagnosis of tumours. Several CADs make use of machine learning, resulting in a constant design of hand-crafted features aimed at better assisting the physician. In recent years, Deep learning (DL) approaches raised in popularity in many pattern recognition tasks thanks to their ability to learn compact hierarchical features that well fit the specific task to solve. If, on one and, this characteristic suggests to explore DL suitability for biomedical image processing, on the other, it is important to take into account the physiological inheritance of the images under analysis. With this goal in mind, in this work we propose "3TP U-Net", an U-Shaped Deep Convolutional Neural Network that exploits the well-known Three Time Points approach for the lesion segmentation task. Results show that our proposal is able to outperform not only the classical (non-deep) approaches but also some very recent deep proposal, achieving a median Dice Similarity Coefficient of 61.24%.

Supporting the early diagnosis of skin cancer is crucial for the sake of any kind of treatment or surgery. This work proposes to improve the outcome of automatic diagnoses approaches by using an ensemble of pre-trained deep convolutional neural networks and a suitable voting strategy. Moreover, a novel patching approach has been deployed. The proposal has been fairly evaluated with the literature proposals demonstrating good preliminary results.

Nowadays, Machine Learning techniques are more and more pervasive in several application fields. In order to perform an evaluation as reliable as possible, it is necessary to consider the reproducibility of these models both at training and inference time. With the introduction of Deep Learning (DL), the assessment of reproducibility became a critical issue due to heuristic considerations made at training time that, although improving the optimization performances of such complex models, can result in non-deterministic outcomes and, therefore, not reproducible models. The aim of this paper is to quantitatively highlight the reproducibility problem of DL approaches, proposing to overcome it by using statistical considerations. We show that, even if the models generated by using several times the same data show differences in the inference phase, the obtained results are not statistically different. In particular, this short paper analyzes, as a case study, our ICPR2018 DL based approach for the breast segmentation in DCE-MRI, demonstrating the reproducibility of the reported results.

Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) is a diagnostic method for the detection and diagnosis of breast cancer. Requiring the acquisition of images before and after the injection of a paramagnetic contrast agent, it provides a large amount of data that can hardly be analyzed without the use of a Computer Aided Diagnosis (CAD) system, whose aim is to support radiologists in the interpretation of medical images. Among the major issues in developing a CAD for the breast DCE-MRI there is the lesion diagnosis, namely the classification of lesioned tissues according to the tumour aggressiveness. Several studies have been conducted so far to explore the applicability of Deep Learning (DL) approaches to the automatic breast lesions classification. However, we argue that solutions only relying on DL are not so effective since past learned experience in the radiomics field should also be kept in mind to better exploit the dynamics of contrast agent and its effect on the acquired images. To this aim, we propose an approach that exploits the well-known Three Time Points (3TP) idea to select the specific time points that best highlight the tissues under analysis. Our findings show that promising results can then be obtained by using transfer learning, resulting in an approach that is able to outperform both the classical (non-deep) and some very recent deep proposals.

Nowadays, every facet of people’s lifestyle is impacted by the continuous use of technologies and in particular the Internet. Social interactions have been radically changed since new technologies have removed communication barriers. The obstacles of time and space have been overcome, letting people from different places and cultures communicate. People tend to become part of a dense social network, where the distribution of information becomes an almost immediate process. This is the reason why companies, public institutions and business activities have opted for social networks as communication medium. As a consequence, people seem to incorporate information gained from the social networks into their decision-making processes.

International audience; The food economic chain of many rivers is based on an important control of the presence of predatory fish in the water. The assessment of the predation pressure on migratory species cannot be done manually. Therefore some automatic techniques are needed. In this paper we propose, for the first time, a deep neural architecture to estimate the catfish density from Aerial Images taken on the Loire river. The proposed architecture is adapted to the problem in hand and some variations to existing approaches, never applied in this application context, are designed to better fit the needs of the problem. Preliminary results show the appropriateness of the proposal and form the foundations for future developments on this new application.

Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) is a diagnostic method suited for the early detection and diagnosis of cancer, involving the serial acquisition of images before and after the injection of a paramagnetic contrast agent. Dealing with long acquisition times, DCE-MRI inevitably shows noise (artefacts) in acquired images due to the patient (often involuntary) movements. As a consequence, over the years, machine learning approaches showed that some sort of motion correction technique (MCT) have to be applied in order to improve performance in tumours segmentation and classification. However, in recent times classic machine learning approaches have been outperformed by deep learning based ones, thanks to their ability to autonomously learn the best set of features for the task under analysis. This paper proposes a first investigation to understand if deep learning based approaches are more robust to the misalignment of images over time, making the registration no longer needed in this context. To this aim, we evaluated the effectiveness of a MCT both for the classification and for the segmentation of breast lesions in DCE-MRI by means of some literature proposal. Our results show that while MCTs seems to be still quite useful for the lesion segmentation task, they seem to be no longer strictly required for lesion classification one.

Detecting contact lenses is required for reliable iris-based authentication.Iris and sclera information are both relevant for detection and classification.A reliable segmentation method is devised to extract iris and sclera.A bag-of-feature method based on dense descriptors is proposed.Classification performance improves significantly w.r.t. previous state-of-the-art. Detecting the presence of contact lenses and their type helps increasing the reliability of iris-based authentication systems. We propose a machine-learning approach for this task, based on expressive local image descriptors. The image is first segmented to extract the iris and sclera regions, then scale-invariant local descriptors (SID) are computed densely on both areas, and summarized through the Bag-of-Features paradigm. Classification is based on a properly trained linear SVM. The major contributions of our proposal concern the segmentation algorithm, the use of information drawn from the sclera, and the use of non-rectified data to preserve local structures. A number of variants of the proposed method are investigated, working on different areas of the image, with alternative local descriptors, and with different encoding techniques. Eventually, results are compared with the state-of-the-art in the field. The experimental analysis, carried out on several publicly available datasets, shows that the proposed classification method based on a dense scale invariant descriptor outperforms all the reference techniques.

Nowadays, Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) has demonstrated to be a valid complementary diagnostic tool for early detection and diagnosis of breast cancer. However, without a CAD (Computer Aided Detection) system, manual DCE-MRI examination can be difficult and error-prone. The early stage of breast tissue segmentation, in a typical CAD, is crucial to increase reliability and reduce the computational effort by reducing the number of voxels to analyze and removing foreign tissues and air. In recent years, the deep convolutional neural networks (CNNs) enabled a sensible improvement in many visual tasks automation, such as image classification and object recognition. These advances also involved radiomics, enabling high-throughput extraction of quantitative features, resulting in a strong improvement in automatic diagnosis through medical imaging. However, machine learning and, in particular, deep learning approaches are gaining popularity in the radiomics field for tissue segmentation. This work aims to accurately segment breast parenchyma from the air and other tissues (such as chest-wall) by applying an ensemble of deep CNNs on 3D MR data. The novelty, besides applying cutting-edge techniques in the radiomics field, is a multi-planar combination of U-Net CNNs by a suitable projection-fusing approach, enabling multi-protocol applications. The proposed approach has been validated over two different datasets for a total of 109 DCE-MRI studies with histopathologically proven lesions and two different acquisition protocols. The median dice similarity index for both the datasets is 96.60 % (±0.30 %) and 95.78 % (±0.51 %) respectively with p < 0.05, and 100% of neoplastic lesion coverage.

Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) has demonstrated, in recent years, a great potential as a complementary diagnostic method for early detection and diagnosis of breast cancer. However, due to the large amount of data, DCE-MRI manual inspection is error prone and can hardly be handled without the use of a Computer Aided Diagnosis (CAD) system. In a typical CAD processing, the segmentation of the breast parenchyma is a crucial stage aimed to reduce computational effort and to increase reliability. In the last years, deep convolutional networks have outperformed the state-of-the-art in many visual tasks, such as image classification and object recognition. However, very few proposals based on a deep learning approach have been applied so far for segmentation tasks in the biomedical field. The aim of this work is to apply a suitably modified convolutional neural network for fully-automating the non-trivial breast tissues segmentation task in 3D MR data, in order to accurately segment breast parenchyma from the air and other tissues (such as chest-wall). The proposed approach has been validated over 42 DCE-MRI studies. The median segmentation accuracy and Dice similarity index were 98.93 (±0, 15) and 95.90 (±0, 74) respectively with p< 0.05, and 100% of neoplastic lesion coverage.

The aim of this paper is to propose an algorithm based on convolutional neural networks (CNN) for iris sensor model identification. This task is important in forensics applications as well as to face the problem of sensor interoperability in large scale systems. When different sensor models are involved in a recognition system, in fact, the overall performance can strongly decrease. A possible solution consists in first identifying the sensor model and then mapping the features extracted from the image from one sensor to the other. To keep low both complexity and memory requirements we propose a simple network architecture and the use of transfer learning to speed-up the training phase and tackle the problem of limited training set availability. Experiments are carried out on several public iris databases. First, we show that the proposed solution outperforms the state-of-the art approaches used for the model identification task. Then, we test the performance of a biometric recognition system and show that improving the sensor model identification step can benefit the iris sensor interoperability.

Electrocardiography (ECG) has been recently proposed as biometric trait for identification purposes. Intra-individual variations of ECG might affect identification performance. These variations are mainly due to Heart Rate Variability (HRV). In particular, HRV causes changes in the QT intervals along the ECG waveforms. This work is aimed at analysing the influence of seven QT interval correction methods (based on population models) on the performance of ECG-fiducial-based identification systems. In addition, we have also considered the influence of training set size, classifier, classifier ensemble as well as the number of consecutive heartbeats in a majority voting scheme. The ECG signals used in this study were collected from thirty-nine subjects within the Physionet open access database. Public domain software was used for fiducial points detection. Results suggested that QT correction is indeed required to improve the performance. However, there is no clear choice among the seven explored approaches for QT correction (identification rate between 0.97 and 0.99). MultiLayer Perceptron and Support Vector Machine seemed to have better generalization capabilities, in terms of classification performance, with respect to Decision Tree-based classifiers. No such strong influence of the training-set size and the number of consecutive heartbeats has been observed on the majority voting scheme.

We propose to use local binary patterns (LBP) on image residual for iris liveness detection.Low-complexity interpolation-free implementation enables use on mobile devices.Performance is promising for both print-based and screen-based attacks. Iris recognition is well suited to authentication on mobile devices, due to its intrinsic security and non-intrusiveness. However, authentication systems can be easily tricked by attacks based on high-quality printing. A liveness detection module is therefore necessary. Here, we propose a fast and accurate technique to detect printed-iris attacks based on the local binary pattern (LBP) descriptor. In order to improve the discrimination ability of LBP and better explore the image statistics, LBP is performed on a high-pass version of the image with 3 i? 3 integer kernel. In addition a simplified interpolation-free descriptor is considered and finally a linear SVM classification scheme is used. The detection performance, measured on standard databases, is extremely promising, despite the resulting very low complexity, which makes possible the implementation for the relatively small CPU processing power of a mobile device.

Biometric authentication systems are quite vulnerable to sophisticated spoofing attacks. To keep a good level of security, reliable spoofing detection tools are necessary, preferably implemented as software modules. The research in this field is very active, with local descriptors, based on the analysis of microtextural features, gaining more and more popularity, because of their excellent performance and flexibility. This paper aims at assessing the potential of these descriptors for the liveness detection task in authentication systems based on various biometric traits: fingerprint, iris, and face. Besides compact descriptors based on the independent quantization of features, already considered for some liveness detection tasks, we will study promising descriptors based on the joint quantization of rich local features. The experimental analysis, conducted on publicly available data sets and in fully reproducible modality, confirms the potential of these tools for biometric applications, and points out possible lines of development toward further improvements.

Presentation attacks such as using a contact lens with a printed pattern or printouts of an iris can be utilized to bypass a biometric security system. The first international iris liveness competition was launched in 2013 in order to assess the performance of presentation attack detection (PAD) algorithms, with a second competition in 2015. This paper presents results of the third competition, LivDet-Iris 2017. Three software-based approaches to Presentation Attack Detection were submitted. Four datasets of live and spoof images were tested with an additional cross-sensor test. New datasets and novel situations of data have resulted in this competition being of a higher difficulty than previous competitions. Anonymous received the best results with a rate of rejected live samples of 3.36% and rate of accepted spoof samples of 14.71%. The results show that even with advances, printed iris attacks as well as patterned contacts lenses are still difficult for software-based systems to detect. Printed iris images were easier to be differentiated from live images in comparison to patterned contact lenses as was also seen in previous competitions.

We explore the use of convolutional neural networks for the semantic classification of remote sensing images. Two recently proposed architectures, CaffeNet and GoogLeNet, are adopted, with three different learning modalities. Besides conventional training from scratch, we resort to pre-trained networks that that only need to be fine-tuned on the target data, so as to avoid overfitting problems and reduce design time. Experiments on three remote sensing datasets, with markedly different characteristics, testify on the effectiveness and wide applicability of the proposed solution, which provides a state-of-the-art performance.

In the Big-Data era, many engineering tasks have to deal with extracting valuable information from large amount of data. This is supported by different methodologies, many of which strongly rely on curve fitting (both linear and non-linear). One of the most common approach to solve this kind of problems is the use of least squares method, usually by iterative procedures that can cause slowness when applied to problems that require to repeat the fitting procedure many times. In this work we propose a method to speed-up the curve fitting evaluation by means of a Look-up Table (LuT) approach, exploiting problems resilience. The considered case study is the fitting of breast Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) data to a pharmacokinetic model, that needs to be fast for clinical usage. To validate the proposed approach, we compared our results with those obtained by using the well-known Levenberg-Marquardt algorithm (LMA). Results show that the proposed approach and LMA are not statistically different in terms of MSE (with respect to the observed data) and in terms of Sum of Differences in the parameter and in the solution spaces. Considering the computational effort, the proposed LuT approach is an order of magnitude faster than the LMA on the same dataset.

Camera model identification has great relevance for many forensic applications, and is receiving growing attention in the literature. Virtually all techniques rely on the traces left in the image by the long sequence of in-camera processes which are specific of each model. They differ in the prior assumptions, if any, and in how such evidence is gathered in expressive features. In this work we study a class of blind features, based on the analysis of the image residuals of all color bands. They are extracted locally, based on co-occurrence matrices of selected neighbors, and then used to train a classifier. A number of experiments are carried out on the well-known Dresden Image Database. Besides the full-knowledge case, where all models of interest are known in advance, other scenarios with more limited knowledge and partially corrupted images are also investigated. Experimental results show these features to provide a state-of-the-art performance.

Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) is gaining popularity as a complementary diagnostic method for early detection and diagnosis of breast cancer. However, due to the large amount of data, DCE-MRI can hardly be inspected without the use of a Computer Aided Diagnosis (CAD) system. Among the major issues in developing CAD for breast DCE-MRI there is the classification of regions of interest according to their aggressiveness. For this task newer hand-crafted features are continuously proposed by domain experts. On the other hand, deep learning approaches have gained popularity in many pattern recognition tasks, being able to outperform classical machine learning techniques in different fields, by learning compact hierarchical representations of an image which well fit the specific task to solve. The aim of this work is to explore the applicability of Convolutional Neural Networks (CNN) in automatic lesion malignancy assessment for breast DCE-MRI data. Our findings show that while promising results in treating DCE-MRI can be obtained by using transfer learning, CNNs have to be carefully designed and tuned in order to outperform approaches specifically designed to exploit all the available data information.

Implementing hardware accelerators of multiple classifier systems assures an improving in performance: on one hand, the combination of multiple classifiers outcomes is able to improve classification accuracy, with respect to a single classifier; on the other hand, implementing the prediction algorithm by means of an integrated circuit enables classifier systems with higher throughput and better latency compared with a pure software architecture. Although, this approach requires a very high amount of hardware resources, limiting the adoption of commercial configurable devices, such as Field Programmable Gate Arrays. In this paper, we exploit the application of Approximate Computing to trade classification accuracy off for hardware resources occupation. Specifically, we adopt the bit-width reduction technique on a multiple classifier system based on the Random Forest approach. A case study demonstrates the feasibility of the methodology, showing an area reduction ranging between 8.3 and 72.3%.

Biometric authentication systems are pervasive in modern society, but they are quite vulnerable to spoofing attacks. Research on spoofing (or liveness) detection is therefore very active. A number of methods have been proposed in the literature, sometimes with very promising results, but limited robustness with respect to the large variety of biometric traits, sensors, and attacks encountered in real-life. Recently, methods based on Convolutional Neural Networks (CNNs) are drawing great attention, given their success in many other image processing tasks. However, despite some promising results, they seem to suffer the same robustness problem, requiring heavy training to work properly. Here, we propose a new CNN architecture for biometric spoofing detection. Thanks to domain-specific knowledge, accounted for through a suitable loss function, a compact architecture is obtained, allowing reliable training also in the presence of small-size datasets. Experiments prove the proposal to provide state-of-art performance on several widespread datasets for face and iris liveness detection.

The Photo Response Non-Uniformity (PRNU) noise can be regarded as a camera fingerprint and used, accordingly, for source identification, device attribution and forgery localization. To accomplish these tasks, the camera PRNU is typically assumed to be known in advance or reliably estimated. However, there is a growing interest for methods that can work in a real-word scenario, where these hypotheses do not hold anymore. In this paper we analyze a PRNU-based framework for forgery localization in a blind scenario. The framework comprises four main steps: PRNU-based blind image clustering, parameter estimation, device attribution, and forgery localization. Each of these steps impacts on the final outcome of the analysis. The aim of this paper is to assess the overall performance of the proposed framework and how it depends on the individual steps.

Now, data collection and analysis are becoming more and more important in a variety of application domains, as long as novel technologies advance. At the same time, we are experiencing a growing need for human&ndash, machine interaction with expert systems, pushing research toward new knowledge representation models and interaction paradigms. In particular, in the last few years, eHealth&mdash, which usually indicates all the healthcare practices supported by electronic elaboration and remote communications&mdash, calls for the availability of a smart environment and big computational resources able to offer more and more advanced analytics and new human&ndash, computer interaction paradigms. The aim of this paper is to introduce the HOLMeS (health online medical suggestions) system: A particular big data platform aiming at supporting several eHealth applications. As its main novelty/functionality, HOLMeS exploits a machine learning algorithm, deployed on a cluster-computing environment, in order to provide medical suggestions via both chat-bot and web-app modules, especially for prevention aims. The chat-bot, opportunely trained by leveraging a deep learning approach, helps to overcome the limitations of a cold interaction between users and software, exhibiting a more human-like behavior. The obtained results demonstrate the effectiveness of the machine learning algorithms, showing an area under ROC (receiver operating characteristic) curve (AUC) of 74.65% when some first-level features are used to assess the occurrence of different chronic diseases within specific prevention pathways. When disease-specific features are added, HOLMeS shows an AUC of 86.78%, achieving a greater effectiveness in supporting clinical decisions.

It is well known that some sort of motion correction technique (MCT) should be performed before DCE-MRI data analysis in order to reduce movement artefacts. However, it is not clear if a single MCT can produce optimum results for every single examination, since for example different movements can occur. In this paper we investigated the possibility of choosing the best MCT per each specific patient, before performing further data analysis (e.g. tumour segmentation). In particular, our aim is the proposal of some physiological model-based quality indexes (QIs) for ranking different MCT on a patient basis. Moreover, for practical feasibility, we investigated the performance of our proposal when only a small fraction of the available data was used. We performed tests on a dataset of patients with breast tumour. Specifically, for each patient we compared the “reference ranking” of different MCT obtained by using the results of tumour segmentation with the rankings produced with each QI. Our results indicate that the ranking obtained by using the QI based on the Extended Tofts-Kermode model (with the Parker arterial input function) are in accordance with the “reference ranking”. Moreover, computational load can be significantly reduced without affecting the overall performance by using only 5% of the available data.

Dynamic Contrast Enhanced - Magnetic Resonance Imaging (DCE-MRI) is gaining popularity as complementary diagnostic tool for breast cancer. In a typical Computer Aided Detection (CAD) processing, the identification and segmentation of the breast parenchyma is a crucial stage aimed to reduce computational effort and increase reliability, by reducing the number of voxels to analyse and removing foreign tissues and air. The aim of this work is to propose a fully-automated geometrical-based breast-mask extraction method in DCE-MRI, that combines three 2D Fuzzy C-Means clustering and geometrical breast anatomy characterization. In particular, seven well defined key-points have been considered in order to accurately segment breast parenchyma from air and chest-wall. The proposed approach has been validated on 30 DCE-MRI studies. The median segmentation accuracy and Dice similarity index were 97.86 (±0.49) and 92.66 (±1.48) respectively with p < 0.05, and 100% of neoplastic lesion coverage.

We propose a new algorithm for blind camera identification, based on the Photo-Response Non-Uniformity (PRNU) noise estimated by image residuals. Successful identification relies on the correct clustering of residuals coming from the same camera. We adopt a two-step strategy. First, residuals are efficiently grouped by correlation clustering, setting parameters so as to over-partition the data points and avoid any wrong associations. Then, basic clusters are progressively merged by an ad hoc refinement algorithm. Experiments on the Dresden database prove the effectiveness of the proposed method.

The valorization and promotion of worldwide Cultural Heritage by the adoption of Information and Communication Technologies represent nowadays some of the most important research issues with a large variety of potential applications. This challenge is particularly perceived in the Italian scenario, where the artistic patrimony is one of the most diverse and rich of the world, able to attract millions of visitors every year to monuments, archaeological sites and museums. In this paper, we present a general recommendation framework able to uniformly manage heterogeneous multimedia data coming from several web repositories and to provide context-aware recommendation techniques supporting intelligent multimedia services for the users--i.e. dynamic visiting paths for a given environment. Specific applications of our system within the cultural heritage domain are proposed by means of real case studies in the mobile environment related both to an outdoor and indoor scenario, together with some results on user's satisfaction and system accuracy.

It has been showed that, by employing fake fingers, the existing fingerprint recognition systems may be easily deceived. So, there is an urgent need for improving their security. Software-based liveness detection algorithms typically exploit morphological and perspiration-based characteristics separately to measure the vitality. Both such features provide discriminant information about live and fake fingers, then, it is reasonable to investigate also their joint contribution. In this paper, we combine a set of the most robust morphological and perspiration-based measures. The effectiveness of the proposed approach has been assessed through a comparison with several state-of-the-art techniques for liveness detection. Experiments have been carried out, for the first time, by adopting standard databases. They have been taken from the Liveness Detection Competition 2009 whose data have been acquired by using three different optical sensors. Further, we have analyzed how the performance of our algorithm changes when the material employed for the spoof attack is not available during the training of the system.

Typical pattern recognition applications require to handle both binary and multiclass classification problems. Several researchers have pointed out that obtaining a classifier that discriminates between two classes is much easier than building one that simultaneously distinguishes among all classes. This observation has motivated substantial research on using a pool of binary classifiers to address multiclass problems. Such an approach is also named as decomposition method. Anyway, the performance of a given classification system can be sometimes unsatisfactory for the needs of real applications, especially when these are characterized by large data variability and/or significant amount of noise. In these cases it is important that the classification system is able to estimate the reliability of its decision for each sample under test. This estimate could be used, for example, for deciding to reject a sample instead of running the risk of misclassifying it, so improving the overall system performance. Based on these motivations, this paper defines a reliability estimator for decomposition schemes belonging to the One-per-Class framework. The estimator is based on the reliabilities provided by each binary classifier, on the status of their outputs while it is independent of their design. The performance of the proposed approach has been assessed on private and public medical datasets, showing that it can be used to improve the classification performance of the One-per-Class scheme with respect to both multiclass classifiers and other well-known decomposition schemes.

In this paper, we critically review the issue of information fusion for computer security, both in terms of problem formulation and in terms of state-of-the-art solutions. We also analyze main strengths and weaknesses of currently used approaches and propose some research issues that should be investigated in the future.

In the age of Big Data, which is clearly affecting also the Healthcare sector, one of the most valuable challenge is the one connected with the information extraction from raw data that implies the automatic detection of significant facts in unstructured texts and their transformation into structured documents, which are indexable and queryable exactly like databases. The volume, variety, velocity, verification and value of data raise the necessity of managing information with the most sophisticated linguistic and computational architectures, which are able to approach the semantic dimension of words and sentences. The present paper introduces ABC, A knowledge Based Collaborative framework that consists in a double-faced system aiming to support clinical processes in a more effective way (diagnostics and therapy). All the ABC's area of intervention (improving security from clinical risk, enhancing services' quality/results and perfecting the effects of the spending management) help to make hospitals more compliant with national and international laws and observant of many standards. ABC is intended as way to perform auto-diagnoses of the safety and the quality in hospitals, so the medical staff is always prepared to successfully overcome authorities' inspections; it must not be viewed as a supervisory control instrument. Moreover, ABC guarantees protection from every illegal use of sensitive data regarding patients and health users.

Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. A. Morales, M. Falanga, J. Fierrez, C. Sansone and J. Ortega-Garcia, "Keystroke dynamics recognition based on personal data: A comparative experimental evaluation implementing reproducible research," 2015 IEEE 7th International Conference on Biometrics Theory, Applications and Systems (BTAS), Arlington, VA, 2015, pp. 1-6. doi: 10.1109/BTAS.2015.7358772, This work proposes a new benchmark for keystroke dynamics recognition on the basis of fully reproducible research. Instead of traditional authentication approaches based on complex passwords, we propose a novel keystroke recognition based on typing patterns from personal data. We present a new database made up with the keystroke patterns of 63 users and 7560 samples. The proposed approach eliminates the necessity to memorize complex passwords (something that we know) by replacing them by personal data (something that we are). The results encourage to further explore this new application scenario and the availability of data and source code represent a new valuable resource for the research community., A.M. is supported by a post-doctoral Juan de la Cierva contract by the Spanish MECD (JCI-2012-12357). M.F. was supported by a scholarship provided by University of Naples Federico II and Compagnia di San Paolo. This work has been partially supported by projects: Bio-Shield (TEC2012-34881) from Spanish MINECO, BEAT (FP7-SEC-284989) from EU, CECABANK and Cátedra UAM Telefónica.

Italy’s Cultural Heritage is the world’s most diverse and rich patrimony and attracts millions of visitors every year to monuments, archaeological sites and museums. The valorization of cultural heritage represents nowadays one of the most important research challenges in the Italian scenario. In this paper, we present a general multimedia recommender system able to uniformly manage heterogeneous multimedia data and to provide context-aware recommendation techniques supporting intelligent multimedia services for the users. A specific application of our system within the cultural heritage domain is proposed by means of a real case study in the mobile environment related to an outdoor scenario, together with preliminary results on user’s satisfaction.

This paper presents an analysis on the suitability of a low cost motion capture system to support the development of generic applications within medical scenarios. In this study we used the Microsoft Kinect (first generation) sensor as basic technology for its wide availability and the relatively low cost, which generated great interest of the scientific community in several application fields. Most of the studies found in literature however, do not address the quality and reliability of the raw data (objects position and depth) considering only the results of the integrated 3D recognition algorithms for the final skeleton/gesture analysis. This can be critical in rehabilitation, in which the knowledge of accuracy, reliability and performance in tracking patient body movements is required and assessed. To this aim, this article proposes an experimental protocol to analyze the Kinect measurement capabilities, via the definition of peculiar parameters. Here we also introduced an analysis of performance versus usage time. Results showed some weak- nesses of the sensor. Specifically, resolution was found in the range of tens of millimeters while degrading over time. The adoption of the Microsoft Kinect (first generation) in medical scenarios should therefore be considered with respect to the application it serves having clear in mind its performances and limitations.

We present a multimedia summarizer system for retrieving relevant information from some web repositories based on the extraction of semantic descriptors of documents. In particular, semantics attached to each document textual sentences is expressed as a set of assertions in the \(\langle subject,verb,object \rangle \) shape as in the RDF data model. While, images’ semantics is captured using a set of keywords derived from high level information such as the related title, description and tags. We leverage an unsupervised clustering algorithm exploiting the notion of semantic similarity and use the centroids of clusters to determine the most significant summary sentences. At the same time, several images are attached to each cluster on the base of keywords’ term frequency. Finally, several experiments are presented and discussed.

In recent years, the need for data collection and analysis is growing in many scientific disciplines. This is consequently causing an increase of research in automated data management and data mining to create reliable methods for data analysis. To deal with the need for smart environments and big computational resources, some previous works proposed to address the problem by moving on remote processing, with the aim of sharing supercomputer resources, algorithms and costs. Following this trend, in this work we propose an architecture for advanced remote data processing in a secure, smart and versatile client–server environment that is capable of integrating pre-existing local software. In order to assess the feasibility of our proposal, we developed a case study in the context of an image-based medical diagnostic environment. Our tests demonstrated that the pro- posed architecture has several benefits: increase of the system throughput, easy upgradability, maintainability and scalability. Moreover, for the scenario we have considered, the system showed a very low transmission overhead which settles on about 2.5 % for the widespread 10/100 mbps. Security has been achieved using client–server certificates and up-to- date standards.