Abstract The rapid advancement of Deep Learning (DL) techniques has led to the widespread proliferation of DeepFake (DF) images, raising real-world concerns. In this work, DeepDect is proposed and evaluated in a real-world scenario. The platform has been developed using a human-centered approach, integrating insights (requirements) from both common and DF expert users. The platform can detect both face-swapping and generated deepfake faces, which are the most common cases of DF in the current context. Two benchmarks have been conducted to evaluate state-of-the-art DF detection models for face-swap and AI-generated images. The best-performing models (ResNet-50 and Random Forest for face-swapping detection, Capsule Forensics v2, and CNN for AI-generated images) have been integrated into the platform as the detection engine. An Explainable AI (XAI) module has been implemented and integrated into DeepDect to provide visual (Grad-CAM heatmaps) and textual explanations, enhancing interpretability and user trust. A real-world evaluation involving 108 participants was performed to assess DeepDect’s effectiveness compared to human detection. DeepDect has achieved 81% detection accuracy, outperforming human users, underscoring the need for such a tool in real-world applications. These findings have highlighted the importance of accessible, explainable, and high-performing AI solutions, offering a balance between technical robustness and User-Centric Design (UCD).

In the digital age, deepfakes pose a growing threat to information security and integrity, with a particular focus on audio fakes. This paper explores and compares different machine-learning techniques for audio deepfake detection, analyzing both deep and shallow learning-based approaches. In particular, CombinedNet, a novel model obtained by combining two deep learning networks selected from the literature, is presented. Performance evaluation was conducted on public datasets through benchmarking based on standard metrics such as accuracy, precision, and F1-score. The experimental results show that CombinedNet outperforms the benchmark models, highlighting the potential of hybrid solutions in audio deepfake detection.

The widespread use of artificial intelligence deep neural networks (DNNs) in fields such as medicine and engineering necessitates understanding their decision-making processes. Current explainability methods often produce inconsistent results and struggle to highlight essential signals influencing model inferences. This paper introduces the Evolutionary Independent Deterministic Explanation (EVIDENCE) theory, a novel approach offering a deterministic, model-independent method for extracting significant signals from black-box models. EVIDENCE theory, grounded in robust mathematical formalization, is validated through empirical tests on diverse datasets, including COVID-19 audio diagnostics, Parkinson's disease voice recordings, and the George Tzanetakis music classification dataset (GTZAN). Practical applications of EVIDENCE include improving diagnostic accuracy in healthcare and enhancing audio signal analysis. For instance, in the COVID-19 use case, EVIDENCE-filtered spectrograms fed into a frozen Residual Network with 50 layers (ResNet50) improved precision by 32 % for positive cases and increased the Area Under the Curve (AUC) by 16 % compared to baseline models. For Parkinson's disease classification, EVIDENCE achieved near-perfect precision and sensitivity, with a macro average F1-Score of 0.997. In the GTZAN, EVIDENCE maintained a high AUC of 0.996, demonstrating its efficacy in filtering relevant features for accurate genre classification. EVIDENCE outperformed other Explainable Artificial Intelligence (XAI) methods such as Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Gradient-weighted Class-Activation Mapping (GradCAM) in almost all metrics. These findings indicate that EVIDENCE not only improves classification accuracy but also provides a transparent and reproducible explanation mechanism, crucial for advancing the trustworthiness and applicability of AI systems in real-world settings.

In today’s software development landscape, the DevSecOps approach has gained traction due to its focus on the software development process and bolstering security measures in projects, a task in light of the ever-evolving cybersecurity threats. This study aims to address the lack of metrics for quantitatively assessing its efficacy from both security and business logic perspectives. To tackle this issue, the research introduces the Framework of Business Index Concerning Security (FOBICS), a set of metrics designed to enable transparent evaluations of project security. FOBICS considers various perspectives relevant to DevSecOps practices. It includes factors such as project duration and financial outcomes, making it appealing for implementation in business settings. The effectiveness of FOBICS is validated theoretically and empirically via its application in two real-world projects: the results from these implementations show a correlation between FOBICS metrics and the security strategies employed as the development methodologies adopted by diverse teams throughout the projects. Hence, FOBICS emerges as a tool for assessing and continuously monitoring project security, offering insights into areas of strength and areas that may require enhancement. FOBICS is shown to be effective in assessing the level of DevSecOps implementation. The ease of calculating FOBICS metrics makes them easily interpretable and continuously verifiable. Moreover, FOBICS summarizes most of the other quantitative and qualitative metrics in the literature. • DevSecOps challenges were analysed to identify metrics to assess project performance • A framework of metrics is proposed. The degree of Security and Testing is evaluated • FOBICS is compared with other metrics, showing how it can summarize many of them • The framework is applied to two real projects and the values obtained are evaluated

Abstract This study introduces a Deep Neural Network architecture called LightAudioCNN. Its main purpose is to examine cord vibration patterns to improve the diagnosis of Parkinsons’ disease (PD) and differentiate it from similar conditions. LightAudioCNN represents a step in developing more objective and precise diagnostic tools, especially crucial in the early stages of PD, unlike the conventional symptom-based methods known for their arbitrary and unreliable nature. By analyzing vowel sounds (“a” and “i”) from a dataset of 83 participants, this study evaluates LightAudioCNN’s effectiveness while ensuring the reliability of its outcomes using a patient separation method. LightAudioCNN demonstrates high diagnostic accuracy and efficiency, achieving an Area Under the Curve (AUC) score of 0.99 in binary classification tasks and 0.96 in multiclass classification tasks with corresponding accuracy rates of 95% and 81%. These results were obtained through comparisons with Deep Neural Networks trained on Mel Spectrograms and contemporary transformer models processing Mel spectrograms or raw audio data. Additionally, the application of LightAudioCNN to the Italian Parkinson Speech dataset further substantiates its high diagnostic capability. On this dataset, LightAudioCNN achieved a mean accuracy of 97.69%, a precision of 97.88%, and an AUC score of 0.9873, illustrating its ability to capture complex speech patterns associated with Parkinson’s disease. The model’s performance was in line with the other deep learning models. Furthermore, the study highlights the versatility of LightAudioCNN beyond Parkinsons’ disease by proving its superiority in identifying COVID-19 by analyzing breath patterns and cough sounds. In this comparison, LightAudioCNN surpasses deep learning and traditional machine learning models by achieving a mean accuracy of 78.81% in the same scenarios. This proves the model’s potential for quickly and accurately diagnosing COVID-19, demonstrating its relevance across conditions. The model also has a small footprint of about 3.1 M parameters, which is about 7 times less than standard computer vision architectures such as ResNet50, allowing the integration of this technology locally into smartphone applications with the aim of managing and treating not just Parkinson’s’ Disease but also emerging health threats, like COVID-19.

Deep learning (DL) has been demonstrated to be a valuable tool for analyzing signals such as sounds and images, thanks to its capabilities of automatically extracting relevant patterns as well as its end-to-end training properties. When applied to tabular structured data, DL has exhibited some performance limitations compared to shallow learning techniques. This work presents a novel technique for tabular data called adaptive multiscale attention deep neural network architecture (also named excited attention). By exploiting parallel multilevel feature weighting, the adaptive multiscale attention can successfully learn the feature attention and thus achieve high levels of F1-score on seven different classification tasks (on small, medium, large, and very large datasets) and low mean absolute errors on four regression tasks of different size. In addition, adaptive multiscale attention provides four levels of explainability (i.e., comprehension of its learning process and therefore of its outcomes): 1) calculates attention weights to determine which layers are most important for given classes; 2) shows each feature's attention across all instances; 3) understands learned feature attention for each class to explore feature attention and behavior for specific classes; and 4) finds nonlinear correlations between co-behaving features to reduce dataset dimensionality and improve interpretability. These interpretability levels, in turn, allow for employing adaptive multiscale attention as a useful tool for feature ranking and feature selection.

This paper provides an in-depth exploration of the Crop Module within the "EarTH Observation for the Early forecasT of Irrigation needS (THETIS)" project, specifically addressing challenges in precision agriculture. The study unfolds in the "Fortore" irrigation district (Southern Italy), focusing in particular on the 6/B district. The Crop Module, rooted in AquaCrop crop model architecture, emerges as a pivotal component in simulating and predicting crop growth, development, and water dynamics. It operates across leaf development, crop growth and productivity, and water balance levels, ensuring adaptability to daily temperature variations for real-time simulations. In interaction with the Soil Water Balance Module (SWB) and leveraging insights from satellite imagery, the Crop Module undergoes meticulous calibration and validation. The expected outcomes encompass increased precision in irrigation scheduling, early anticipation of water demand, and improved seasonal forecasting. This comprehensive approach positions stakeholders for informed decision-making, fostering sustainability and efficiency in agricultural practices.

This paper reports on a Spatial Decision Support System (SDSS) for the early, medium, and short-term forecast of irrigation needs in a semi-arid Mediterranean environment. The SDSS is developed in the context of the "EarTH Observation for the Early forecasT of Irrigation needS (THETIS)" project supported by the Italian Space Agency (ASI). THETIS integrates hydrologic and crop growth models with advanced Earth Observation (EO) products, Artificial Intelligence (AI) and a WEBGIS interface to provide basin-scale information for efficient planning of irrigation resources. The study describes initial results concerning the irrigated area of the Apulian Tavoliere (AT) served by the Reclamation Consortium of the Capitanata, Foggia, Italy.

Human Activity Recognition (HAR) is an essential area of research related to the ability of smartphones to retrieve information through embedded sensors and recognize the activity that humans are performing. Researchers have recognized people's activities by processing the data received from the sensors with Machine Learning Models. This work is intended to be a hands-on survey with practical’s tables capable of guiding the reader through the sensors used in modern smartphones and highly cited developed machine learning models that perform human activity recognition. Several papers in the literature have been studied, paying attention to the preprocessing, feature extraction, feature selection, and classification techniques of the HAR system. In addition, several summary tables illustrating HAR approaches have been provided: most popular human activities in the literature with paper references, the most popular datasets available for download (Analyzing their characteristics, such as the number of subjects involved, the activities recorded, and the sensors with online-availability), co-occurrences between activities and sensors, and a summary table showing the performance obtained by researchers. The paper's goal is to recommend, through the discussion phase and thanks to the tables, the current state of the art on this topic.

Abstract This study establishes a correlation between computer science and psychology, specifically focusing on the incorporation of smartphone sensors and users' personality index. A limited number of state-of-the-art approaches have considered these factors, while no existing dataset currently encompasses this correlation. In this study, an Android application was developed to implement a questionnaire on bullying and cyberbullying, using smartphone sensors to predict Personal Index. Sensor data are collected in the “UNIBA HAR Dataset” and were analyzed using AI algorithms to find a correlation between the categorization class of the questionnaire (Personality Index) and the prediction of ML behavioral models. The results indicate that the Bayesian Bridge with "Bullying bully vs. Victimization bullying" and "Total bullying vs. Total victimization" performs better on average 0.94 accuracy, and the LSTM with the last categorization performs 0.89 accuracy. These results are crucial for future development in the same research area. Graphical abstract

This paper reviews the recent literature on technologies and methodologies for quantitative human gait analysis in the context of neurodegenerative diseases. The use of technological instruments can be of great support in both clinical diagnosis and severity assessment of these pathologies. In this paper, sensors, features and processing methodologies have been reviewed in order to provide a highly consistent work that explores the issues related to gait analysis. First, the phases of the human gait cycle are briefly explained, along with some non-normal gait patterns (gait abnormalities) typical of some neurodegenerative diseases. Then the paper reports the most common processing techniques for both feature selection and extraction and for classification and clustering. Finally, a conclusive discussion on current open problems and future directions is outlined.

In this extended version of this paper, an automatic video diagnosis system for dementia classification is presented. Starting from video recordings of patients and control subjects, performing sit-to-stand test, the designed system is capable of extracting relevant patterns for binary discern patients with dementia from healthy subjects. The original system achieved an accuracy 0.808 by using the rigorous inter-patient separation scheme especially suited for medical purposes. This separation scheme provides the use of some people for training and others, different, people for testing. The implementation of features from the kinematic theory of rapid human movement and its sigma-lognormal model together with classic features increased the overall accuracy of the system to 0.947 F1 score. In addition, multi-class classification was performed with the aim of classifying neurodegenerative disease severities. This work is an original and pioneering work on sit-to-stand video classification for neurodegenerative diseases, its novelties are on phases segmentation, experimental setup and the application of kinematic theory of rapid human movements to sit-to-stand videos for neurodegenerative disease assessment.

This work considers the Internet of Things (IoT) and machine learning (ML) applied to the agricultural sector within a real-working scenario. More specifically, the aim is to punctually forecast two of the most important meteorological parameters (solar radiation and the rainfall) to determine the amount of water needed by a specific plantation under different contour conditions. Three different state-of-the-art ML approaches, coupled with boosting techniques, have been adopted and compared to obtain hourly forecasting. Real-working conditions are referred to the situation in which training data are missing for a specific weather station near the specific field to be irrigated. A simple but effective approach, based on correlation between available weather stations, is considered to cope with this problem. Results, evaluated considering different metrics as well as the execution time, demonstrate the viability of the proposed solution in real IoT working scenario in which these forecasting are input data to successively evaluate irrigation needing.

On-line signatures can be acquired adopting specifically devoted stylus-pad system as well as with general purpose tablet/smartphone. The resulting acquired signal is strongly influenced by the device itself as well as by the user interaction modality. It is evident that it is desirable to identify the most stable features for verification aims. In this work, a set of 39 function feature domains is investigated in terms of stability adopting the weighted Direct Matching Points approach. Experiments have been performed on the e-biosign dataset providing many evidences related to the different scenarios.

In a society with increasing age, the understanding of human falls it is of paramount importance. This paper presents a Decision Support System whose pipeline is designed to extract and compute physical domain's features achieving the state of the art accuracy on the Le2i and UR fall detection datasets. The paper uses the Kinematic Theory of Rapid Human Movement and its sigma-lognormal model together with classic physical features to achieve 98% and 99% of accuracy in automatic fall detection on respectively Le2i and URFD datasets. The effort made in the design of this work is toward recognition of falls by using physical models whose laws are clear and understandable.

Although secular, handwritten signature is one of the most reliable biometric method used by most countries. In the last ten years, the application of technology for verification of handwritten signatures has evolved strongly, including forensic aspects. Some factors, such as the complexity of the background and the small size of the region of interest - signature pixels - increase the difficulty of the targeting task. Other factors that make it challenging are the various variations present in handwritten signatures such as location, type of ink, color and type of pen, and the type of stroke. In this work, we propose an approach to locate and extract the pixels of handwritten signatures on identification documents, without any prior information on the location of the signatures. The technique used is based on a fully convolutional encoder-decoder network combined with a block of refinement layers for the alpha channel of the predicted image. The experimental results demonstrate that the technique outputs a clean signature with higher fidelity in the lines than the traditional approaches and preservation of the pertinent characteristics to the signer's spelling. To evaluate the quality of our proposal, we use the following image similarity metrics: SSIM, SIFT, and Dice Coefficient. The qualitative and quantitative results show a significant improvement in comparison with the baseline system.

On-line signature verification involves the use of many different features or domains. The most stable domains for a signer are analysed in this paper. For this purpose, stable domains are calculated with the weighted Direct Matching Points (ωDMP), which is a relaxed version of the classical DMP technique. In addition to the direct coupling, ωDMP also considers the information contained in the 1:N couplings from the Dynamic Time Warping algorithm. Using the ωDMP technique, state-of-the-art verification results are obtained, showing the capacity to outperform previous DMP techniques to calculate the local stability model of signers.

Education may play a key role in developing “cognitive reserve” against neurodegenerative dementia. In this work, we investigate for the first time if handwriting dynamics can serve as a quantitative indicator of this reserve. We carried out an exploratory study involving a sample of mild cognitive impairment (MCI) subjects, with high and low education respectively, and a sample of healthy elder controls. We asked them to perform three complex handwriting tasks on a digitizing tablet: drawing a clock; copying a check; writing a spontaneous sentence. Dynamic measures of the handwriting were then analyzed both with an unsupervised and a supervised machine learning approach. The results we obtained suggest that: (i) handwriting of MCI subjects with high reserve is quite similar to that of controls; (ii) handwriting of MCI subjects with lower reserve is easier to be distinguished from the other two. Dynamic handwriting analysis could provide a novel methodology to elucidate the still unknown mechanisms underlying brain resilience.

In the present paper, we propose an evolutionary approach to address interoperability issues in multi-device signature verification, based on transformation mappings automatically tuned by a genetic algorithm. These mappings are meant to decrease dissimilarities between signatures acquired through different devices and with different modalities (stylus/finger). The effectiveness of the proposed method was evaluated on the e-BioSign data set. Our proposal achieved an average relative improvement of 26% of EER, for the case of skilled forgeries, compared to baseline results.

Water scarcity is one of the main issues that agriculture must face since an increase is expected not only in developing countries but also in southern Europe with Italy featuring a long-term annual average estimated in 1.909 m3per inhabitant. To deal with this problem, the EcoLoop project presented in this work proposes an ICT system able to collect, aggregate and analyze IoT data, with the aim to foster reuse of wastewater and optimize water usage in agriculture. A Decision Support System (DSS) acts on wastewater plants managing the irrigation and fertilization strategies, the reservation queues and the network distribution exploiting smart sensors, semantic ontologies and machine learning technologies.

This letter investigates different velocity-based signal processing techniques to the aim of Parkinson's disease classification through handwriting. It is showed that combining new velocity-based features with classic features improves state-of-the-art performance on the PaHaW dataset.

This paper presents an experimental analysis on the use of semi-supervised learning in the handwritten digit recognition field. More specifically, two new feedback-based techniques for retraining individual classifiers in a multi-expert scenario are discussed. These new methods analyze the final decision provided by the multi-expert system so that sample classified with a confidence greater than a specific threshold is used to update the system itself. Experimental results carried out on the CEDAR (handwritten digits) database are presented. In particular, error rate, similarity index and a new correlation score among them are considered in order to evaluate the best retraining rule. For the experimental evaluation, an SVM classifier and five different combination techniques at abstract and measurement level have been used. Finally, the results show that iterating the feedback process, on different multi-expert systems built with the five combination techniques, one retraining rule is winning over the other respect to the best correlation score.

This paper presents a new approach for zoning design based on a class-adaptive technique in which the optimal zoning method is defined for each class. For this purpose, in the zoning design stage, a multi-objective genetic algorithm was used to determine, for each class, both the optimal number of zones and the optimal zones for the Voronoi-based zoning method. The experimental tests were carried out in the field of handwritten digit and character recognition. The results show that the new class-adaptive zoning methods proposed in this paper are superior to the set-adaptive methods presented in the literature.

In the networked society, in which a multitude of different devices can be used for signature acquisition, specific research is still needed to determine the extent to which features of an input signature depend on the characteristics of the signature acquisition process. In this paper an experimental investigation is carried out on constrained signatures, which were acquired using writing boxes with different areas and shapes. The paper discusses different behaviour of dynamic and static features with respect to the writing boxes.

This paper presents a new approach for online signature verification that exploits the potential of local stability information in handwritten signatures. Different from previous models, this approach classifies a signature using a multidomain strategy. A signature is first split into different segments based on the stability model of a signer. Then, according to the stability model, for each segment, the most profitable domain of representation for verification purposes is detected. In the verification stage, the authenticity of each segment of the unknown signature is evaluated in the most profitable domain of representation. The authenticity of the unknown signature is then determined by combining local verification decisions. The study was carried out on the signatures in the SUSIG database, and the experimental results, thus, obtained confirm the effectiveness of the proposed approach, when compared with others in the literature.

In recent years, biometric-based authentication systems have been widely used in many applications which require reliable identification scheme. Among others, handwritten signature is one of the most interesting biometric means, that is being considered with renewed interest. This paper presents some of the most relevant advances in the field of offline signature identification and highlights some directions for further research.

In this paper a new on-line signature verification technique is proposed. Differently from previous works, this approach classifies a signature using a multi-domain strategy. In particular, based on the stability model of each signer, the signature is split into different segments and for each segment the most profitable domain of representation for verification purpose is detected. In the verification stage, Dynamic Time Warping (DTW) is used to evaluate the genuinity of each segment of the unknown signature, using the specific domain of representation. The experimental results, carried out on signatures of the SUSIG database, demonstrate the effectiveness of the proposed approach when compared to other approaches in literature.

The aim of this paper is to explore the properties of a new zoning technique based on Voronoi tessellation for the task of handwritten digit recognition. This technique extracts features according to an optimal zoning distribution, obtained by an evolutionary-strategy based search. Extensive experiments have been conducted on the MNIST dataset to investigate strengths and weakness of the proposed approach. Comparisons with regular square zoning reveal that the presented zoning strategy achieves better results with any type of features. Furthermore, the proposed zoning method, jointly with a suitable choice of features, allows a low complexity classifier to reach excellent performances both in terms of accuracy and speed.

A new approach for static signature verification is presented in this paper. The approach uses optical flow to estimate local stability among signatures. In the enrollment stage, optical flow is used to define a stability model of the genuine signatures for each signer. In the verification stage, the stability between the unknown signature and each one of the reference signatures is estimated and consistency with the stability model of the signer is evaluated. The experimental results, carried out on the signatures in the GPDS database, demonstrate the effectiveness of the new approach.

This paper presents a new approach to optimal zoning design. The approach uses a multi-objective genetic algorithm to define, in a unique process, the optimal number of zones of the zoning method along with the optimal zones, defined through Voronoi diagrams. The experimental tests, carried out in the field of handwritten digit recognition, show the superiority of new approach with respect to traditional dynamic approaches for zoning design, based on single-objective optimization techniques.

Three different strategies in order to re-train classifiers, when new labeled data become available, are presented in a multi-expert scenario. The first method is the use of the entire new dataset. The second one is related to the consideration that each single classifier is able to select new samples starting from those on which it performs a missclassification. Finally, by inspecting the multi expert system behavior, a sample misclassified by an expert, is used to update that classifier only if it produces a miss-classification by the ensemble of classifiers. This paper provides a comparison of three approaches under different conditions on two state of the art classifiers (SVM and Naive Bayes) by taking into account four different combination techniques. Experiments have been performed by considering the CEDAR (handwritten digit) database. It is shown how results depend by the amount of the new training samples, as well as by the specific combination decision schema and by classifiers in the ensemble.

Classifier combination is a powerful paradigm to deal with difficult pattern classification problems. As matter of this fact, multi-classifier systems have been widely adopted in many applications for which very high classification performance is necessary. Notwithstanding, multi-classifier system design is still an open problem. In fact, complexity of multi-classifiers systems make the theoretical evaluation of system performance very difficult and, consequently, also the design of a multi-classifier system. This paper presents a new approach for the design of a multi-classifier system. In particular, the problem of feature selection for a multi-classifier system is addressed and a genetic algorithm is proposed for automatic selecting the optimal set of features for each individual classifier of the multi-classifier system. The experimental results, carried out in the field of handwritten digit recognition, demonstrate the effectiveness of the proposed approach.

In handwritten character recognition, zoning is one of the most effective approaches for features extraction. When a zoning method is considered, the pattern image is subdivided into zones each one providing regional information related to a specific part of the pattern. The design of a zoning method concerns the definition of zoning topology and membership function. Both aspects have been recently investigated and new solutions have been proposed, able to increase adaptability of the zoning method to different application requirements. In this paper some of the most recent results in the field of zoning method design are presented and some valuable directions of research are highlighted.

Recently, research in handwritten signature verification has been considered with renewed interest. In fact, in the age of e-society, handwritten signature still represents an extraordinary means for personal verification and the possibility of using automatic signature verification in a range of applications is becoming a reality. This paper focuses on some of the most remarkable aspects the field and highlights some recent research directions. A list of selected publications is also provided for interested researchers.

This paper presents a new technique for the analysis of stability in static signature images. The technique uses an equimass segmentation approach to non-uniformly split signatures into a standard number of regions. Successively, a multiple matching technique is adopted to estimate stability of each region, based on cosine similarity. The GPDS database has been considered for the experimental test. The results demonstrate the validity of the novel approach and highlight some directions for further research.

Abstract This paper describes a Speaker Verification System based on the use of multi resolution classifiers in order to cope with performance degradation due to natural variations of the excitation source and of the vocal tract. The different resolution representations of the speaker are obtained by considering multiple frame lengths in the feature extraction process and from these representations a single Pseudo‐Multi Parallel Branch (P‐MPB) Hidden Markov Model is obtained. In the verification process, different resolution representations of the speech signal are classified by multiple P‐MPB systems: the final decision is obtained by means of different combination techniques. The system based on the Weighted Majority Vote technique considerably outperforms baseline systems: improvements are between 15% and 38%. The execution time of the verification process is also evaluated and it proves to be very acceptable, thus allowing the use of the approach for applications in real time systems.

This letter introduces a new score normalization technique - based on Dynamic Time Warping (DTW) - for output integration in multiclassifier systems. More precisely, DTW is used to match the score cumulative distribution of each individual classifier against a standard cumulative distribution. The warping function allows optimal alignment of the scores provided by the individual classifiers with the scores on the standard cumulative distribution. Furthermore, in order to adapt the normalization process to the behavior of the individual classifiers and to the decision fusion rule, a new class of fuzzy cumulative distributions is introduced and a genetic approach is used to select the optimal distribution to be used as standard cumulative distribution for score normalization. The experimental tests report better results for the fuzzy normalization technique than for those obtained with other approaches present in the literature.

In the field of handwritten character recognition, image zoning is a widespread technique for feature extraction since it is rightly considered to be able to cope with handwritten pattern variability. As a matter of fact, the problem of zoning design has attracted many researchers who have proposed several image-zoning topologies, according to static and dynamic strategies. Unfortunately, little attention has been paid so far to the role of feature-zone membership functions that define the way in which a feature influences different zones of the zoning method. The result is that the membership functions defined to date follow nonadaptive, global approaches that are unable to model local information on feature distributions. In this paper, a new class of zone-based membership functions with adaptive capabilities is introduced and its effectiveness is shown. The basic idea is to select, for each zone of the zoning method, the membership function best suited to exploit the characteristics of the feature distribution of that zone. In addition, a genetic algorithm is proposed to determine-in a unique process-the most favorable membership functions along with the optimal zoning topology, described by Voronoi tessellation. The experimental tests show the superiority of the new technique with respect to traditional zoning methods.

In pattern recognition tasks it is frequent that new (labeled) data became available as the specific application scenario evolves. When a multi expert system (ME) is adopted, the collective behavior of classifiers can be used to select the most profitable samples in order to update the knowledge base. More specifically a misclassified sample, for a particular classifier, is used to update that classifier only if that sample produces a misclassification by the ensemble of classifiers. This approach is compared to situation in which the entire new dataset is used for learning as well as the case in which specific samples are selected by the individual classifier. Successful results have been obtained by considering the CEDAR (handwritten digit) database, moreover it is also shown how they depend by the specific combination decision schema, as well as by data distribution.

In zoning-based classification, a membership function defines the way a feature influences the different zones of the zoning method. This paper presents a new class of membership functions, which are called fuzzy-membership functions (FMFs), for zoning-based classification. These FMFs can be easily adapted to the specific characteristics of a classification problem in order to maximize classification performance. In this research, a real-coded genetic algorithm is presented to find, in a single optimization procedure, the optimal FMF, together with the optimal zoning described by Voronoi tessellation. The experimental results, which are carried out in the field of handwritten digit and character recognition, indicate that optimal FMF performs better than other membership functions based on abstract-level, ranked-level, and measurement-level weighting models, which can be found in the literature.

In this paper a new system for dynamic signature verification is presented. It is based on the consideration that each region of an handwritten signature can convey personal characteristics in diverse domains. Therefore, a multi-expert approach is considered in which each stroke of the signature is evaluated in the most profitable domain of representation. The experimental results demonstrate the effectiveness of the proposed approach.

The evaluation of combination methods for multi-classifier systems is a difficult problem. In many cases multi-classifier combination methods are too complex to be formally studied and the experimental approach is the unique possible strategy. Of course, in order to simulate a multitude of real working conditions, sets of artificial classifiers with diverse characteristics must be generated. This paper presents an effective technique for generating sets of artificial classifiers with different characteristics both at the individual-level (i.e. recognition performance) and at the collective-level (i.e. degree of similarity). In the experimental tests, sets of artificial classifiers simulating different working conditions are generated and the performances of abstract-level combination methods are estimated. The results points out the effectiveness of the new technique for generating sets of artificial classifiers with different characteristics and their usefulness in estimating the performances of combination methods.

This paper addresses the problem of multi-classifier system evaluation by artificially generated classifiers. For the purpose, a new technique is presented for the generation of sets of artificial abstract-level classifiers with different characteristics at the individual-level (i.e. recognition performance) and at the collective-level (i.e. degree of similarity). The technique has been used to generate sets of classifiers simulating different working conditions in which the performance of combination methods can be estimated. The experimental tests demonstrate the effectiveness of the approach in generating simulated data useful to investigate the performance of combination methods for abstract-level classifiers.

Classifier combination is a powerful strategy to support useful solutions in difficult classification problems. Notwithstanding, the effectiveness of a multi-classifier system strongly depends on the decision fusion strategies. In this field, one of the most significant aspects concerns output normalization,when classifiers decisions are provided at measurement level. This paper presents a new approach for output normalization that uses dynamic time warping (DTW). Some experimental tests have been carried out in the field of handwritten digit recognition. The proposed approach is superior to other output normalization algorithms in the literature.

Multi-classifier approach is a widespread strategy used in many difficult classification problems.Traditionally, in a multi-classifier approach, a classification decision based on the combination of a multitude of classifiers is expected to outperform the decisions of each individual classifier. Therefore, in a multi-classifier systems, the potential of the whole set of classifiers is only exploited at the level of the final decision, in which the contributions of all classifiers is used by combining their individual decisions.This paper shows a feed-back based multi-classifier system in which the multi-classifier approach is used not only for providing the final decision, but also for improving the performance of the individual classifiers, by means of a closed-loop strategy.The experimental tests have been carried out in the field of hand-written numeral recognition. The result demonstrates the effectiveness of the proposed approach and its superiority with respect to traditional approach.

In this paper an approach called multi-frame speaker models (MFS) is proposed, in order to cope with performance degradation generally observed over (short and medium) time and trials in speaker identification's task. The approach, based on generative models, uses multiple frame's length for speech processing in training and testing phase. A complete multi-expert system is also presented which is able to implement the proposed approach onthe whole set of speakers and to obtain a near optimum for the ER's reduction.

In recent years, along with the extraordinary diffusion of the Internet and a growing need for personal verification in many daily applications, automatic signature verification is being considered with renewed interest. This paper presents the state of the art in automatic signature verification. It addresses the most valuable results obtained so far and highlights the most profitable directions of research to date. It includes a comprehensive bibliography of more than 300 selected references as an aid for researchers working in the field.

Speaker recognition/identification is a challenge for the implementation of security applications. Unfortunately, degradation in performance is usually observed for high pitched speakers and also whenever average pitch varies significantly between enrolment and testing. In this paper, a study on the impact of the frame length used to extract features from speech signal on the performance of speaker identification is presented. Tests have been carried out on a text-dependent database. Results show that a combination of different frame sizes between the training and the recognition phases can cope with the degradation. A reduction between 40% and 65% in false rejections has been generally observed.

The introduction of a new currency in Europe has changed the way of writing both the courtesy and the legal amount on checks. This paper presents the most important modifications brought on the bank-check processing system in order to solve the related problems also by proposing the software tools that must be utilized. The Computer Aided Software Engineering tools provided by the "Khoros" system are used to support the improvement of the system prototype. A visual programming environment is used to assemble the bankcheck processing system that can be easily modified and extended. The experimental results allow the adjournment of the improved system, as the modifications are introduced.