Learning a high-order connectional brain template (CBT) endowed with cognitive capacities such as visual or auditory memory is crucial for identifying cognition-related biomarkers and distinguishing between control and clinical populations. Higher-order CBTs provide a population-level representation that captures not only structural or topological regularities but also the multi-regional interactions and cognitive processes that conventional pairwise models fail to reflect. Because the brain operates through complex, coordinated dynamics, estimating CBTs that incorporate such higher-order and cognitively meaningful organization is essential for advancing our understanding of neural function and dysfunction. While recent machine-learning and graph-neural-network approaches have improved CBT estimation, they remain limited by their focus on pairwise interactions and purely structural features, overlooking both higher-order organization and cognitive properties. This gap raises a central question: How can we learn a high-order CBT that is well-centered at the population level and also endowed with cognitive capacities? We tackle this challenge using reservoir computing (RC), a biologically inspired framework that mimics how the brain processes information. RC exhibits dynamic properties similar to those of the prefrontal cortex, an area associated with working memory and features a fading memory mechanism, known as the Echo State Property (ESP), which mirrors the brain's short-term memory function. Building on these properties, we introduce HyperCOCO, a novel framework for generating high-order cognitively enhanced CBTs in two stages. First, BOLD signals are processed through a random reservoir to generate high-order individual functional connectomes, which are then aggregated into a population-level template. Second, this template is instantiated into a hyper-cognitive reservoir and stimulated with multi-sensory inputs (visual, auditory, and linguistic). Finally, we measure the memory capacity of the resulting CBT as a proxy for its ability to encode and retain cognitive information. Our source code is available at https://github.com/basiralab/HyperCOCO.

Message passing is a core mechanism in Graph Neural Networks (GNNs), enabling the iterative update of node embeddings by aggregating information from neighboring nodes. Graph Convolutional Networks (GCNs) exemplify this approach by adapting convolutional operations for graph structures, allowing features from adjacent nodes to be combined effectively. However, GCNs encounter challenges with complex or dynamic data. Capturing long-range dependencies often requires deeper layers, which not only increase computational costs but also lead to over-smoothing, where node embeddings become indistinguishable. To overcome these challenges, reservoir computing has been integrated into GNNs, leveraging iterative message-passing dynamics for stable information propagation without extensive parameter tuning. Despite its promise, existing reservoir-based models lack structured convolutional mechanisms, limiting their ability to accurately aggregate multi-hop neighborhood information. To address these limitations, we propose RGC-Net (\emph{Reservoir-based Graph Convolutional Network}), which integrates reservoir dynamics with structured graph convolution. Key contributions include: (i) a reimagined convolutional framework with fixed-random reservoir weights and a leaky integrator to enhance feature retention; (ii) a robust, adaptable model for graph classification; and (iii) an RGC-Net-powered transformer for graph generation with application to dynamic brain connectivity. Extensive experiments show RGC-Net achieves state-of-the-art performance in classification and generative tasks, including brain graph evolution, with faster convergence and mitigated over-smoothing. Our source code is available at https://github.com/basiralab/RGC-Net.

Diseases affecting tomatoes represent a significant threat to agricultural productivity and food security. Identifing diseases early and accurately is crucial to manage crops optimally. In this article, we propose a hard-attention-enhanced VGG16 model for recognizing diseased tomatoes. By incorporating a hard-attention mechanism into the VGG16 architecture, the model effectively focuses on upon most relevant regions image, thereby improving its ability to distinguish between healthy and infected tomatoes. We evaluate Our proposed model is evaluated using one publicly available dataset and demonstrates its superior performance compared to the standard VGG16 architecture and other advanced models. The results obtained from the dataset show that our model achieves accuracy of 96.0% surpassing the baseline VGG16 model, which achieves 92.3% accuracy. These findings underscore the value of hard-attention mechanisms in enhancing plant disease identification systems, including precision interpretability. One of the key strengths of our model lies in its explainability, providing clearer insights to the model’s decision-making process of the model

This paper introduces a novel framework for context-driven need detection in home care, with a strong emphasis on the role of Bidirectional Encoder Representations from Transformers (BERT) in providing nuanced contextual understanding. Our approach leverages BERT to process unstructured textual communications, which then guides probabilistic reasoning through the OwlMEBN Jena API, while dynamically enriching a knowledge graph represented using the Web Ontology Language (OWL). This process integrates multi-modal data via an adaptive weighted fusion of wearable sensor data, caregiver observations, and textual communications. Rigorous experiments, in a simulated environment utilizing real-world data from the e-SAAD platform, demonstrate a significant performance improvement, achieving an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.93 and an F1-score of 0.87, significantly outperforming baseline methods. This enhanced performance underscores BERT’s crucial role in enabling more accurate and personalized home care services by providing deep contextualization of beneficiary needs, while the OwlMEBN framework manages the uncertainty and provides structured probabilistic inferences.

Social media platforms serve as massive repositories of textual data, reflecting diverse human interactions and preferences. However, the unstructured nature of this content poses significant challenges for extracting semantically rich insights. This paper introduces a novel methodology for the automated construction of knowledge graphs (KGs) from social media discourse, specifically focusing on Twitter tweets. Our approach synergistically integrates large language models (LLMs), specifically a fine-tuned BERT model, with an ontology-driven framework. First, we define a detailed ontology of online communication concepts. The pre-trained BERT model is then fine-tuned using a multi-task learning approach on a curated dataset of anonymized and segmented Twitter discussions, thereby aligning its semantic representations with the predefined ontology. The fine-tuned LLM is leveraged for several critical tasks including entity and relation extraction, sentiment analysis, intention classification and the inference of contextual information and discussion styles. Furthermore, a mechanism is introduced to infer inter-user relationships and shared interests using graph neural networks (GNNs), analyzing patterns in interaction and language use. This multi-faceted extracted and inferred data is subsequently employed to build a knowledge graph, stored and queried via the Neo4j graph database management system. This study presents several contributions such as the integration of a ontology with an LLM method and the innovative user relationship and shared interest extraction using graph neural networks. The proposed methodology was rigorously evaluated using a real-world dataset of Twitter discussions, showcasing its ability to capture semantic content, and elucidate inter-user relationships effectively and revealing shared interests of the users involved. Furthermore, an ablation study is included which further demonstrates each of the method component contribution and demonstrates the importance of such integrations. Our findings highlight the potential for various downstream applications such as in community structure analysis and sentiment analysis to improve information management within online social networks.

This paper introduces OntoMed-KGTransformer, a novel framework for integrating unstructured clinical text with structured medical knowledge graphs (KGs) and ontologies (specifically, UMLS). Our approach addresses key limitations in clinical decision support by combining: (1) an ontology-guided hierarchical attention mechanism that enforces domain-specific semantic constraints within a Transformer architecture; (2) a dynamic graph-centric tokenization strategy that bridges textual and KG representations; and (3) a dual-encoder architecture with ontology-driven contrastive learning. Evaluations on the MIMIC-III dataset and a custom Hetionet-derived KG demonstrate significant improvements over state-of-the-art baselines (e.g., KG-BERT, BioBERT) in diagnostic prediction, with a 15% relative increase in F1-score. Furthermore, the framework enhances interpretability through the extraction of clinically relevant knowledge paths, promoting trust and transparency in clinical decision-making.

This paper introduces a novel methodology for diagnosing COVID-19 leveraging Generative Adversarial Networks for Conditional Tabular data (GANCT). GANCT constructs a sophisticated conditional generative adversarial framework specialized for precise modeling of statistical distributions endemic to COVID-19 tabular datasets. Extensive experiments quantify the impact on predictive performance from augmenting the original data with synthetic GANCT samples. Results prove superior COVID-19 screening accuracy and ROC AUC across classifiers when supplementing real data with GANCT synthesizations. Moreover, we identified the optimal augmentation range that markedly improves performance while avoiding overfitting. The key contributions are: 1) A customized GAN architecture for COVID-19 tabular data synthesis, 2) Comprehensive evaluations demonstrating the benefits of GANCT augmentation, 3) Identification of optimal augmentation levels, and 4) Substantial improvements in COVID-19 prediction. GANCT can help overcome data scarcity to develop more reliable AI diagnostic systems for the pandemic.

A Smart-Grid (SG) represents an advanced electrical network designed for intelligent and efficient management across its entire infrastructure, facilitating seamless communication and coordination among its interconnected components, including IoT-enabled devices, and leveraging real-time data transfer mechanisms. This paper focuses on examining the local level within the SG, primarily tasked with monitoring energy consumption. To gain a comprehensive understanding of these concepts and the strategies employed for effective energy management, an energy management system has been devised. The primary objective of this system is twofold: first, to minimize the overall energy consumption within the SG, ensuring it remains within or below the received energy quantity; and second, to optimize the utility of appliances while ensuring they do not surpass the total energy capacity supplied by the Photovoltaic Panels (PVPs). To achieve this, we employ the Knapsack algorithm, wherein the locally produced energy serves as the algorithm's capacity, and the devices seeking energy consumption are treated as objects. Each device's weight, fixed consumption, and utility values are considered as attributes defining these objects, aiding in the algorithm's decision-making process. Through this approach, we aim to develop a robust energy management framework capable of efficiently allocating resources while maximizing overall utility within the SG ecosystem.

Real-world adversarial physical patches were shown to be successful in compromising state-of-the-art models in a variety of computer vision applications. Existing defenses that are based on either input gradient or features analysis have been compromised by recent GAN-based attacks that generate naturalistic patches. In this paper, we propose Jedi, a new defense against adversarial patches that is resilient to realistic patch attacks. Jedi tackles the patch localization problem from an information theory perspective; leverages two new ideas: (1) it improves the identification of potential patch regions using entropy analysis: we show that the entropy of adversarial patches is high, even in naturalistic patches; and (2) it improves the localization of adversarial patches, using an autoencoder that is able to complete patch regions from high entropy kernels. Jedi achieves high-precision adversarial patch localization, which we show is critical to successfully repair the images. Since Jedi relies on an input entropy analysis, it is model-agnostic, and can be applied on pre-trained off-the-shelf models without changes to the training or inference of the protected models. Jedi detects on average 90% of adversarial patches across different benchmarks and recovers up to 94% of successful patch attacks (Compared to 75% and 65% for LGS and Jujutsu, respectively).

Signatures are one of the most relevant and convenient methods for user verification, despite their vulnerability to skilled tampering. In this work, we propose a user-independent offline signature verification technique based on the Siamese architecture. For the feature extraction step, we use an improved residual network by incorporating channel wise attention. Experiments conducted on the CEDAR database have validated the effectiveness and strength of our approach.

Brain tumor segmentation from MRIs has been a continuing challenge for radiology and neurology specialists. Therefore, a reliable method that captures the growing and refined data in this regard, is much needed. Among many deep learning techniques having been proposed for medical image analysis, U-Net-based variants are found to the most used models in multimodal medical image segmentation. Moreover, knowing that brain tumors can have various shapes, sizes, and appearances, the segmentation task, using simple block architectures, does not capture all the intricacies of the tumor boundaries and internal structures. Therefore, more complex architectures, such as U-Net and its 3D extensions, are better suited to handle these variations and provide better segmentation results. In this paper, we introduce an enhanced automated 3D brain tumor segmentation network built on the foundation of the 3D U-Net architecture. Through the combination of the advantages of UNet, ResNet, we created a new 3D UNet model with block modifications, based on simple ResNet block, BNet’s, with batch normalization. We used the 2020 Multi-modal Brain Tumor Segmentation Challenge (BraTS2020) data sets to train and validate the proposed model. The BraTS2020 validation data set yielded dice values of 0.86, 0.82 and 0.85 for the Whole Tumor (WT), for Enhancing Tumor core (ET), and for Tumor Core (TC), respectively. The experimental results show that our model significantly outperforms the standard brain tumor segmentation methods.

Information extraction (IE) from a receipt-based document requires understanding the contextual and visual semantics of texts. However, prior studies have predominantly utilized pre-trained language models and Optical Character Recognition (OCR) engines to extract textual features from the document and conduct entity extraction. Besides, this paper presents a novel approach for information extraction from receipt-based documents by leveraging both visual and textual features. The proposed approach employs a convolutional neural network to capture visual features and a word embedding model to capture textual features. These features are then combined and fed into a fully connected layer for entity classification. The experimental results demonstrate that the proposed approach outperforms the baseline and achieves state-of-the-art results in information extraction from receipts.

Noninvasive medical neuroimaging has yielded many discoveries about the brain connectivity. Several substantial techniques mapping morphological, structural and functional brain connectivities were developed to create a comprehensive road map of neuronal activities in the human brain -namely brain graph. Relying on its non-euclidean data type, graph neural network (GNN) provides a clever way of learning the deep graph structure and it is rapidly becoming the state-of-the-art leading to enhanced performance in various network neuroscience tasks. Here we review current GNN-based methods, highlighting the ways that they have been used in several applications related to brain graphs such as missing brain graph synthesis and disease classification. We conclude by charting a path toward a better application of GNN models in network neuroscience field for neurological disorder diagnosis and population graph integration. The list of papers cited in our work is available at https://github.com/basiralab/GNNs-in-Network-Neuroscience.

Driver monitoring at nighttime is a tremendous area of research because of its crucial role to save lives and decrease traffic crashes injuries. However, this task is highly complex because of the high amount of naturalistic driving issues and the low visibility. In the gist of this paper, a novel nighttime driver action recognition network named multi-convolutional stream for hybrid network are proposed, which effectively fuses multimodal data to efficiently classify driver's actions in low visibility and a cluttered driving scene. Using the unique public driver action dataset recorded at nighttime, up to our knowledge, in two separate viewpoints, our proposed methodology beats state-of-the-art methodologies in classification performance, with an advancement of up to 10% over the best-practice methods.

Reasoning on an ontology is presently limited to the logical one. However, in the case of inconsistent knowledge or unreliable and incomplete information, it is difficult for a system to make a decision or measure the degree of truth of a hypothesis. Several approaches have been made and focused mainly on how to represent probabilistic information in ontologies and then use them for reasoning. Although, this requires a great effort for systems already designed and based on ontological or Bayesian knowledge bases. In these cases the users must integrate the probabilistic information of Bayesian networks (BN) in ontology manually. We propose a new approach to integrate the BN (more particularly Object-Oriented Bayesian Networks (OOBn)) with OWL by providing a Jena API (Application Programming Interface). OOBNs is an extension of the standard BN using the paradigm object. Our approach (i) uses the BN knowledge base for ontological enrichment, (ii) allows the adjustment of BN structures through structural and parametric learning based on ontological knowledge bases, (iii) integrate probabilistic information into SPARQL queries. In addition, we propose a solution in which we use a selection of instances according to the needs of the user described by a set of SPARQL requests.

Several areas have evolved with the evolution of technology. Telemedicine and home care are one of them. While there are many studies on home care, the overall problem of decision-making is not sufficiently addressed. In previous research, we proposed an ontological-based intervention system. Despite its abilities, reasoning on an ontology is currently limited to logic. However, in the case of inconsistent knowledge or unreliable and incomplete information, it is difficult for a system to make a decision or assess the degree of truth of a hypothesis. Several approaches have been made and focused mainly on how to represent probabilistic information in ontologies and then use them for reasoning. Although, this requires a great effort for systems already designed and based on ontological or Bayesian knowledge bases. In this article, we propose two approaches to combining probabilistic and logical reasoning within a home-based medical intervention system. The first, using probabilistic ontology PR-OWL. The second, providing a Jena API (Application Programming Interface).

In order to contribute to safe sharing of 3D mul-tiresolution meshes, a new approach of crypto-watermarking is proposed. This approach is based on the use of RSA algorithm and spherical harmonics transform. The aim being is to increase the integration rate while maintaining mesh quality on the one hand and extract correctly inserted image on the other hand. To embed data, host mesh is decomposed using spherical harmonics transform. Resulting coefficients are watermarked twice to insert the grayscale image already encrypted using RSA algorithm. Finally, watermarked mesh is reconstructed through the use on 10% of coefficients already calculated. Found results prove that our approach is able to insert a high amount of data without influencing the mesh quality. The application of the most popular attacks does not prevent a correct extraction of data already inserted. Our algorithm is,then,robust against these attacks.

This paper provides an informative overview about the structure, the operating principle, and the mathematical model of the Deep Belief Network (DBN) which is one of the best known and trusted deep learning models as well as a brief outline of the well-known and, likewise, the suggested pattern recognition applications that make wide use of this specific kind of neural network based on probability model. The application particularly targeted in this paper is the recognition of actions of violence. Hence, various experimentations have been conducted by varying several parameters using DeeBNet object-oriented MATLAB toolbox and the proposed approach returned an accuracy of 65.5 %.

Artificial intelligence methods have aided the advancement of several disciplines of science and technology. Furthermore, they have had a significant influence on smart grid management. One of the most significant information for optimal management in Smart Grid is the ability to predict its parameters: electricity consumption and meteorological factors. It is mostly utilized to develop improved control ways for building appliances (such as lighting and heating/cooling systems). Several methodologies for load and weather data characterization prediction have recently been proposed. The work discussed in this paper is aimed at the development and the comparison of three artificial intelligence forecasting approaches used to manage the Smart Grid by integrating load and climate data predictions. To achieve this objective, we primarily looked on the predicting accuracy of some artificial intelligence methodologies: a neural network, a neuro-fuzzy and a deep learning prediction algorithms are applied and compared to forecast the smart grid parameters (temperature, solar radiation, wind speed and the energy consumption). The simulation results are checked based on real database of wind speed, temperature, solar radiation, and consumption data. The findings of the simulation give us an idea about the most appropriate and performant algorithm to use in this aim.

In the field of brain tumor image analysis, automatic brain tumor image segmentation remains a challenging task due to the varying sizes, shapes, and textures of type of these masses. Deep Learning algorithms have improved the speed and quality of segmentation for certain tasks in medical imaging, especially for Glioma image segmentation. This work present to design and evaluate an algorithm capable of segmenting MRI images. In this regard, U-Net is the most prominent deep network, we recall that it has been the most popular architecture in the medical image segmentation. In this paper, we propose modifications to the U-Net architecture in order efficiently handle a multi-modal MRI input. Based on these modifications, we develop a novel architecture, HDFU-Net as a potential successor to the U-Net architecture by introducing and presenting an improved dice loss that can address the demand for more accurate segmentation in medical images and modify network architecture to improve segmentation performance. The improved dice loss is called Hybrid Dice Focal loss (HDF loss). We then evaluate the presented approach on the BraTS 2020 dataset and discuss the results.

Computer-aided design has been widely used in modern industry for several decades, resulting in the huge databases of 3D CAD models that specialist companies currently own. Therefore, developing a solution to retrieve a reusable 3D CAD model becomes a strategic need for companies specializing in the modern design and manufacturing industry. Recently, some research works have been launched with the aim of recognizing 3D CAD objects based on design similarity and reusability. In this context, the use of probabilistic graphical models for information retrieval was always of great importance, especially when the context is characterized by the large volume of data and the uncertainty of the result. In this paper, authors will present a new approach proposed for modeling 3D CAD objects into reusable subparts. This approach is based on the Hidden Markov Model (HMM). This model has shown improved accuracy and efficiency in recognizing reusable 3D CAD objects, compared to other previously proposed solutions.

Segmentation is one of the most important subjects in image analysis due to its good performance in a wide range of applications. It is the task of clustering parts of an image together, which belong to the same object class. Using medical images for tumor growth modeling involves the improvement of all tasks of image processing, most importantly the segmentation. We introduce the tumor segmentation framework based on traditional machine learning and community detection algorithm. In This paper, we propose a novel approach for image segmentation, which is based on community detection algorithms existing in social networks. In this regard, we propose a method based on super-pixels and algorithms for community detection in graphs. The super-pixel method reduces the number of nodes in the graph while community detection algorithms provide more accurate segmentation than traditional approaches. We compare our method with the image segmentation method based on the deep learning approach and our previous work. Experimental results have shown that our method provides more precise segmentation.

While machine learning applications are getting mainstream owing to a demonstrated efficiency in solving complex problems, they suffer from inherent vulnerability to adversarial attacks. Adversarial attacks consist of additive noise to an input which can fool a detector. Recently, successful real-world printable adversarial “patches” were proven efficient against state-of-the-art neural networks. In the transition from digital noise based attacks to real-world physical attacks, the myriad of factors affecting object detection will also affect adversarial patches. Among these factors, view angle is one of the most influential, yet under-explored. In this paper, we study the effect of view angle on the effectiveness of an adversarial patch. To this aim, we propose the first approach that considers a multi-view context by combining existing adversarial patches with a perspective geometric transformation in order to simulate the effect of view angle changes. Our approach has been evaluated on two datasets: the first dataset which contains most real world constraints of a multi-view context, and the second dataset which empirically isolates the effect of view angle. The experiments show that view angle significantly affects the performance of adversarial patches, where in some cases the patch loses most of its effectiveness. We believe that these results motivate taking into account the effect of view angles in future adversarial attacks, and open up new opportunities for adversarial defenses.

In this paper, we propose a new Conditional Region Growing (CRG) approach with the ability of finding the accurate MC boundaries starting from selected seed points. The starting seed points are determined based on regional maxima detection and superpixel analysis. The region growing step is controlled by a set of criteria derived from prior knowledge to characterize MCs. The key feature is to highlight below each MC to estimate the appropriate criteria and not to use the same parameters for all of them. Defined criteria can be divided into two categories. The first one concerns the neighbourhood searching size. The second one deals with the gradient information and shape evolution within the growing process. Experimental results show the benefits of used criteria in terms of improving the MC delineation qualities.

This paper deals with the problem of the optimal sizing in the PVP/Battery system. To achieve this aim, an ANFIS estimation algorithm has been developed in order to estimate a data base of instantaneous photovoltaic power. The estimated instantaneous PV power has been used in an optimal algorithm for sizing a PVP / Battery power station to supply a 1.5 Kw AC load.The simulation of the proposal sizing system has been implemented in Matlab. The results of the simulation give a good performance of our proposal sizing system.

During the first generation of home intervention and support, complaints were usually triggered by telephone call. In most situations needs where described orally; there were no sufficient information neither on the patient's condition nor on his medical history. Thanks to the advance of technology, especially systems based on data mining, artificial intelligence time is saved, and immediate service is offered. Although home intervention and care have been the subject of numerous studies, the resolution of the overall decision-making problem is not sufficiently developed. Due to the lack of information such as the state of health of a patient, the set of parameters characterizing the daily life habits of the person analyzed in parallel with the evolution of physiological and environmental parameters. In addition to that, it is necessary to take into consideration the Medical Core, his location, his profile not only his professional status, but also his capacities and skills, which are not explicitly described in his curriculum. Different studies and systems exist in literature. Each them takes into account only part of the parameters. Indeed, these studies face either the monitoring of daily activities, the monitoring of physiological data or other environmental part. Either they take into account the specific features of the medical core profile. Either these systems use probabilistic data mining, which involves many interactions with experts to interpret the data, or an expert system based on the rules of inference defined by medical experts. In addition, most do not use controlled vocabulary, which provides semantics to the system. This complicates information sharing and collaborative work. We proposed in a previous work an ontology-based solution [3], the goal is to help the collaborator to make a decision and to draw new information. as the complexity of the data increases, so does the need to deal with uncertainty. Several approaches to the representation and reasoning of uncertainty in the semantic web have emerged. This article is a study for the integration of uncertainty in e-SAAD ontologies expressed in Web ontology language.

With the technological advancement in the field of Computer Aided Design such as the rapid development of scanning technologies, the reconstruction of complete and incomplete cylinders given noisy point clouds with form defects becomes an important issue. In fact, cylindrical surfaces are found in domestic to industrial contexts. In this paper, a comparative study of cylinder fitting algorithms manufactured in the LIPPS laboratory is proposed. The aim of the proposed approach is to determine the diameter of cylindrical feature for minimizing roundness error from experimental data-points. The roundness error is evaluated using two internationally defined methods: Minimum Circumscribed Cylinder (MCC) and Maximum Inscribed Cylinder (MIC). All algorithms give similar results in the case where the scanned cylinder is complete and without form defects, but in the case of missing data some algorithms give unacceptable results. The two reference cylinders have been independently analyzed, respecting six criteria (calculation complexity, damping parameter, initial guess, time, circularity error and complexity cylinder). The results of algorithms are also compared to help manufacturers and inspectors facilitate and improve the application of these methods and to select the appropriate algorithm for size and form evaluation.

For several years, the reconstruction of Computer Aided Design (CAD) models from a deformed mesh get more and more attention. This CAD model is used in order to visualize 3D objects that were scanned and approximate their shapes by mathematical formulations. It represents the geometric support used in many other activities (analysis, manufacturing, assembly, etc.). Surface reconstruction is the most difficult problem of CAD model reconstruction. There are two types of surfaces: primitive surfaces and complex surfaces. In this paper, we propose a method to reconstruct complex surfaces. Our algorithm is based on Thin Plate Spline (TPS) method to optimize locations of control points of a B-Spline surface. Once surfaces are approximated, the geometric model can be reconstructed. We evaluate every step of our approach using mechanical models and show that we can achieve good results and meaningful approximated control points comparing with other methods.

In this paper, we propose a global approach for speech emotion recognition (SER) system using empirical mode decomposition (EMD). Its use is motivated by the fact that the EMD combined with the Teager-Kaiser Energy Operator (TKEO) gives an efficient time-frequency analysis of the non-stationary signals. In this method, each signal is decomposed using EMD into oscillating components called intrinsic mode functions (IMFs). TKEO is used for estimating the time-varying amplitude envelope and instantaneous frequency of a signal that is supposed to be Amplitude Modulation-Frequency Modulation (AM-FM) signal. A subset of the IMFs was selected and used to extract features from speech signal to recognize different emotions. The main contribution of our work is to extract novel features named modulation spectral (MS) features and modulation frequency features (MFF) based on AM-FM modulation model and combined them with cepstral features. It is believed that the combination of all features will improve the performance of the emotion recognition system. Furthermore, we examine the effect of feature selection on SER system performance . For classification task , Support Vecto Machine (SVM) and Recurrent Neural Networks (RNN) are used to distinguish seven basic emotions. Two databases- the Berlin corpus, and the Spanish corpus- are used for the experiments. The results evaluated on the Spanish emotional database, using RNN classifier and a combination of all features extracted from the IMFs enhances the performance of the SER system and achieving 91.16% recognition rate. For the Berlin database, the combination of all features using SVM classifier has 86.22% recognition rate.

Brain cancer phenotyping and treatment is highly informed by radiomic analyses of medical images. Specifically, the reliability of radiomics, which refers to extracting features from the tumor image intensity, shape and texture, depends on the accuracy of the tumor boundary segmentation. Hence, developing fully-automated brain tumor segmentation methods is highly desired for processing large imaging datasets. In this work, we propose a cooperative learning framework for multi-label brain tumor segmentation, which leverages on Structured Random Forest (SRF) and Bayesian Networks (BN). Basically, we embed both strong SRF and BN classifiers into a multi-layer deep architecture, where they cooperate to better learn tumor features for our multi-label classification task. The proposed SRF-BN cooperative learning integrates two complementary merits of both classifiers. While, SRF exploits structural and contextual image information to perform classification at the pixel-level, BN represents the statistical dependencies between image components at the superpixel-level. To further improve this SRF-BN cooperative learning, we ‘deepen’ this cooperation through proposing a multilayer framework, wherein each layer, BN inputs the original multi-modal MR images along with the probability maps generated by SRF. Through transfer learning from SRF to BN, the performance of BN improves. In turn, in the next layer, SRF will also benefit from the learning of BN through inputting the BN segmentation maps along with the original multimodal images. With the exception of the first layer, both classifiers use the output segmentation maps resulting from the previous layer, in the spirit of auto context models. We evaluated our framework on 50 subjects with multimodal MR images (FLAIR, T1, T1-c) to segment the whole tumor, its core and enhanced tumor. Our segmentation results outperformed those of several comparison methods, including the independent (non-cooperative) learning of SRF and BN.

Nowadays, recent technological advances in the acquisition, modeling and processing of three-dimensional (3D) objects data lead to the creation of models stored in huge databases, which are used in various domains such as computer vision, augmented reality, game industry, medicine, CAD (Computer-aided design), 3D printing etc. On the other hand, the industry is currently benefiting from powerful modeling tools enabling designers to easily and quickly produce 3D models. The great ease of acquisition and modeling of 3D objects make possible to create large 3D models databases, then, it becomes difficult to navigate them. Therefore, the indexing of 3D objects appears as a necessary and promising solution to manage this type of data, to extract model information, retrieve an existing model or calculate similarity between 3D objects. The objective of the proposed research is to develop a framework allowing easy and fast access to 3D objects in a CAD models database with specific indexing algorithm to find objects similar to a reference model. Our main objectives are to study existing methods of 3D objects similarity calculation (essentially shape-based methods) by specifying the characteristics of each method as well as the difference between them, and then we will propose a new approach for indexing and comparing 3D models, which is suitable for our case study and which is based on some studied previously methods. Our proposed approach is finally illustrated by an implementation, and evaluated in a professional context.

Our work highlights event detection system in video surveillance sequences. This should mainly distinguish acts of violence. The survey discusses the current methods and techniques that are being applied for the task of automated violence recognition in the images derived from video surveillance sequences. To do this, we propose a fusion strategy after using a variety of feature extraction algorithms to obtain the points-of-interest from input images and each of the extracted feature vectors is submitted to a classifier. In a decision fusion strategy, different classifiers are used to classify a feature vector and to establish a most suitable decision to classify the input action as violent or non-violent. We study the performance of the mentioned approaches on 21 datasets of human interaction images. Experiments were implemented in Matlab computing environment. This paper aspires to be a contribution for researchers who wish to improve the study of violent activity recognition and gather inspiration on the main challenges to tackle in this emerging field.

Geometric model reconstruction from a set of points is a difficult problem, which has been tackled with many different approaches. The reconstruction of the mechanical part is a necessity to visualize parts, simulate, assembly and detect interferences... The reconstruction of geometric entities (curves, edges, surfaces, faces) of these parts introduces particular difficulties. The most difficult problem to obtain the 3D geometric model from a cloud of points is the reconstruction of the faces. Many methods have been proposed to simplify the reconstruction of surfaces. There are two types of surface reconstruction: one of primitive shapes and the other of complex shapes like deformed mechanical parts and objects containing complex surfaces. In this paper, we present an algorithm to reconstruct the computer-aided design model from a deformed mesh. Then, we address a solution to reconstruct a 3D surface from a cloud of points extracted from a deformed mesh.

Gait is a recent important research field among the computer vision community. It aims identifying humans by analyzing their walk. It has different advantage comparing to others biometrics technologies such as face recognition, iris recognition and fingerprint. It can be performed at distance and without subject cooperation. Also, it doesn't need high resolution of image. In this paper, we present a new discriminative method for gait recognition using hybrid conditional random fields (CRF). We use a Hidden CRF model to combine two classifiers; a spatial classifier which assigns a label to a local feature (SURF descriptors) and temporal classifier which uses a motion History Image (MHI). The proposed framework, firstly extracts the human silhouette. Secondly, it takes out spatial and temporal cues from each frame. Then, it applies the MLP classification to the two set of features to obtain the Hidden CRF input; the final step is recognizing person with HCRF. Experimental results showed the superiority of our proposed method over several state of arts.

Hidden Markov Models (HMM) have been widely used in classification tasks. Despite their efficiency in stochastic sequences labeling, they are overwhelmed by imperfect quality of used data in the learning and inference processes. In this paper, we try to evaluate the contribution of possibilistic theory in creating sequences of observations used by HMM models. Experimental results show that observation sequences, obtained by possibilistic reasoning significantly, improve the performance of HMM in the recognition of online e-learning activities.

For the segmentation of ancient digitized document images, it has been shown that texture feature analysis is a consistent choice for meeting the need to segment a page layout under significant and various degradations. In addition, it has been proven that the texture-based approaches work effectively without hypothesis on the document structure, neither on the document model nor the typographical parameters. Thus, by investigating the use of texture as a tool for automatically segmenting images, we propose to search homogeneous and similar content regions by analyzing texture features based on a multiresolution analysis. The preliminary results show the effectiveness of the texture features extracted from the autocorrelation function, the Grey Level Co-occurrence Matrix (GLCM), and the Gabor filters. In order to assess the robustness of the proposed texture-based approaches, images under numerous degradation models are generated and two image enhancement algorithms (non-local means filtering and superpixel techniques) are evaluated by several accuracy metrics. This study shows the robustness of texture feature extraction for segmentation in the case of noise and the uselessness of a demising step.

Recently, texture-based features have been used for digitized historical document image segmentation. It has been proven that these methods work effectively with no a priori knowledge. Moreover, it has been shown that they are robust when they are applied on degraded documents under different noise levels and types. In this paper an approach of evaluating texture-based feature sets for segmenting historical documents is presented in order to compare them. We aim at determining which texture features could be more adequate for segmenting graphical regions from textual ones on the one hand and for discriminating text in a variety of situations of different fonts and scales on the other hand. For this purpose, six well-known and widely used texture-based feature sets (autocorrelation function, Grey Level Co occurrence Matrix, Gabor filters, 3-level Haar wavelet transform, 3-level wavelet transform using 3-tap Daubechies filter and 3-level wavelet transform using 4-tap Daubechies filter) are evaluated and compared on a large corpus of historical documents. An additional insight into the computation time and complexity of each texture-based feature set is given. Qualitative and numerical experiments are also given to demonstrate each texture-based feature set performance.

Falls are a common problem for old people. They can result in dangerous consequences even death. Therefore, automatic tools for fall detection using camera vision can be very useful for helping the elderly. These methods are based on analyzing extracted features. Different features are used such as vertical and horizontal gradient, motion history of image, shape analysis and posture. In this paper, we try to do an investigation of many proposed methods for the fall detection and compare their performances.

Hidden Markov Models (HMMs) are now widely used for off-line Arabic handwriting recognition. Actually, classical HMMs are one-dimensional models, that is why to process an Arabic word image we have developed a discrete Dynamic Bayesian Network (DBN). The DBNs are an extension and a generalization of the classical HMMs, which can model the interaction between several observations and state sequences. In our study, we have represented words by factorizing two streams in different manners, where the interaction is achieved through the causal influence between observable variables in the first model and state variables in the second one. The aim of this is to consider the two flows of information together: The observations on the columns (as well as lines) are obtained by scanning the image horizontally (and also vertically) by a uniform sliding window. We have compared the two models on the recognition of off-line Arabic handwritten words. The experiments show that the first model is better and more adapted to our task than the second one.