This study investigates the classification of two prominent movement disorders: Parkinson’s Disease (PD) and Essential Tremor (ET) using a comprehensive machine learning framework. A novel dataset was meticulously created which contains handwriting samples collected at Habib Bourguiba Hospital in Sfax, Tunisia, specifically designed for differentiating between PD and ET. Preprocessing techniques such as image resizing, normalization, and data augmentation were employed to enhance robustness. Feature extraction was performed using the ResNet50 model, effectively capturing essential image characteristics through global average pooling. Recursive Feature Elimination (RFE) was then applied to identify the most significant features, followed by the training and validation of two classification models Random Forest and SVM using these selected features. The performance of these models is rigorously assessed through various metrics, revealing that the Random Forest model attained an accuracy of $92.83 \% \pm 2$, while the SVM model achieved an average accuracy of $94.66 \% \pm 1$. Visualizations such as confusion matrices and ROC curves provide deeper insights into model performance. Overall, the findings demonstrate the potential of machine learning techniques to enhance diagnostic accuracy in distinguishing between PD and ET, ultimately contributing to improved clinical decision-making.

The recognition of Arabic and Latin text for autonomous cars involves developing systems and algorithms capable of accurately detecting and understanding Arabic characters and words from images. This technology is crucial in enabling autonomous vehicles to interpret and respond to Arabic and Latin traffic signs, road markings, and other textual information on the roads. The development of a reliable identification system, particularly for Arabic is challenging if a dataset contains differences in text size, typefaces, colors, orientation, illumination and noise. These problems become more difficult to solve. By exploiting the benefits of both CTC (Connectionist Temporal Classification) and Attention mechanisms to improve the accuracy and robustness of the recognition system, a hybrid CTC and Attention model is used in the prediction stage for Arabic-Latin image text recognition. Current research is focusing intensively on text panels in Latin, while other scripts, such as Arabic, remain little valued. For this reason, the NaSTSArLaTR (Natural Scene Traffic Sign and Panel Guide Arabic-Latin Text Recognition) dataset has been set up to validate our experiments. Our tests found that the suggested hybrid CTC and Attention model outperformed either Attention or CTC alone in the prediction phase. The dataset is publicly available in IEEE DataPort https://dx.doi.org/10.21227/phyg-_jc98.

Contactless fingerprint authentication has gained popularity as a field of research in biometrics in recent years. Unlike traditional fingerprint recognition systems that require direct contact of the person’s finger with the sensor, contactless fingerprint systems offer several advantages, among them ease of capture and cost-effectiveness. Despite the progress made in this field, poor contrast, background noise, and limited image information continue to pose difficulties for fingerprint recognition in contactless environments. Furthermore, the number of images in published fingerprint biometric datasets for each person is restricted, and there is insufficient data for efficient training. Nevertheless, Convolutional Neural Network (CNN) architectures have been widely used, necessitating large databases. To address these issues, this paper introduces a Siamese network designed for the purpose of identity Verification, using the contactless thumb fingerprint modality to enhance recognition results. The Siamese network is able to extract pertinent features from noisy images with low contrast and limited information, even if they have limited information and low contrast. Additionally, this work proposes the use of the contactless thumb fingerprint modality instead of the contactless index fingerprint modality, which is more commonly used in related works. Consequently, the Mobile FingerPrint (MFP) dataset is introduced and constructed for evaluation. Experimental results demonstrate the efficiency of the proposed method, achieving an accuracy of 98.68% for thumb fingerprint Verification.

The real risk in the privacy of a plain image transmitted over an unsecured connection is content alteration or illegal eavesdropping. The architecture proposed in this paper provides Encryption and Authentication (EA) in their comprehensively for protecting a transmitted image from unauthorized with an average execution time of one second for each 256*256 color image. At the origin point (sender side), it encrypts and signs the image’s content. On the receiver side, it decrypts and verifies the cipher image, achieving two stages of confidentiality and two stages of verification for the image content without depending upon any third party and reinforced with two key chaos equal size color images. Every image has a unique hash value (signature or identity) with QR Code watermarking used to detect forgeries, even if the change is slight (even by one bit). EA has features that make it more resistant to security attacks, as shown by the experiments in this paper.

Parkinson’s disease (PD) is a common and progressive neurodegenerative disorder with motor symptoms and a variety of non-motor symptoms. Experts regularly include handwriting as one of the Parkinsonian motor symptoms of PD and as a valuable tool that can aid in diagnosing and tracking the disease’s progression. PD patients have two periods. ‘On’ time is when levodopa is working well and your symptoms are controlled. ‘Off’ time is when levodopa is no longer working well and symptoms such as tremor, rigidity and slow movement re-emerge. To our knowledge, all existing publicly available datasets allow PD to be identified using only one period. No publicly available online handwriting datasets are dedicated to the analysis of PD using these two periods. Therefore, in this paper, we present our new online Arabic handwriting dataset for analysing PD, which we will make publicly available so that it could potentially be used for diagnosis, screening and monitoring the progression of PD. Our dataset was collected from 30 healthy controls and 30 PD patients in both “off” and “on” at the Neurology Department, Habib Bourguiba Hospital, Sfax, Tunisia. All participants performed five different handwriting tasks. The tasks included drawing repetitive ellipses, a spiral, repetitive digits and Arabic word writing. We hope that our new dataset will help researchers in the early detection of Parkinson’s disease, inpatient rehabilitation and quantification of therapeutic effects.

Facial Expressions Recognition (FER) has become an active area of research. To accomplish emotion recognition, several of machine learning algorithms have been employed. However, these models require a significant amount of data, a large training memory and an important time for training. This paper presents a Facial Expression Recognition method based on ArcFace features and TinySiamese network to solve these problems. The proposed method consists of two parts: the first part is for feature extraction and the second is for feature recognition. In the first part, a feature extractor was used to extract features from the embedded image. The feature extractor was based on a ResNet-50 network and ArcFace loss function. In the second part, the TinySiamese network was used for feature classification and verification. The proposed method was evaluated on three popular datasets: FER2013, RAF-DS, and ExpW. The experimental part shows that the proposed method achieved competitive results compared to related works with classification rates equal to 60.43%, 85.14%, 65.13% and verification rates equal to 73.31%, 84.29%, 76.08% on the FER2013, RAF-DS and ExpW datasets respectively.

People can be recognized by a lot of unique biometric features either physiological or behavioral. Therefore, a biometric security method for a smartphone is proposed in this paper for easier use and to improve recognition results by using a new knuckle modality: the major knuckle of the thumb finger modality instead of the major knuckle of the middle finger modality which is the most used in related works. In finger knuckle recognition domains, different CNN (Convolutional Neural Network) architectures have been applied, requiring a huge database. To deal with this problem, this paper presents a Siamese network for identity verification using contactless major knuckle of thumb finger modality. Actually, the Siamese network has been well used for one-shot learning that aims at learning and recognizing from little data. The FMK (Finger Major Knuckle) dataset was proposed and constructed for testing and evaluation. In fact, this study achieved good results with accuracy of 82% for the major knuckle of the left thumb finger and 79.63% for the major knuckle of the left middle finger.

Compared to traditional biometric systems, in-air signatures are considered more robust and secure than classical pen paper. A few datasets capturing in-air signatures have been introduced, utilizing various devices such as the Leap Motion and the Microsoft Kinect sensor camera. However, these devices are not exempt from shortcomings and exhibit certain limitations. The expenses associated with their implementation and the requirement for technical proficiency in operating them present notable challenges for in-air signature analysis. Additionally, users may encounter difficulties in adapting their finger movements to fit within the device's limited field of view, particularly if they lack familiarity with these devices. To address these concerns, this paper proposes the creation of three in-air signature datasets using solely the camera of a laptop or a smartphone, eliminating the need for any additional specialized equipment. Our datasets were collected in three ways. The first is the In-Air Signature dataset (IAS dataset) and the second is the In-Air Signature dataset using a transparent Glass Plate (IASGP dataset) while the third is the In-Air Signature dataset using Smart Phone (IASSP dataset). Forty volunteers participated in the construction of these datasets. Their ages ranged from 21 to 40 years. Each volunteer signs in the air five signatures and imitates five signatures of five other volunteers. Our in-air signatures datasets are publicly available and can be used for various research tasks like in-air signature verification and identification.

Unconstrained handwriting text recognition is a stimulating field in the branch of pattern recognition. This field is still an open search due to the wide variability of human writing. Recent trends show a potential improvement of recognition by adoption a novel representation of extracted features. In the present paper, we propose a novel feature extraction model by learning a Bag of Features Framework for handwritten text recognition based on Deep Sparse Auto-Encoder. The Hidden Markov Models are then used for sequences modeling. For features learned quality evaluation, our proposed system was tested on two handwritten text datasets IFN/ENIT word images benchmark and MNIST handwritten digits. Our method achieves promising recognition on both datasets.

Deep neural networks have proved promising results in many applications and fields, but they are still assimilated to a black box. Thus, it is very useful to introduce interpretability aspects to prevent the blind application of deep networks. This paper proposed an interpretable morphological convolutional neural network called Morph-CNN for pattern recognition, where morphological operations were incorporated using counter-harmonic mean into the convolutional layer in order to generate enhanced feature maps. Morph-CNN was extensively evaluated on MNIST and SVHN benchmarks for digit recognition. The different tested configurations showed that Morph-CNN outperforms the existing methods.

In this paper we proposed a writer adaptation system based on an adaptive fuzzy inference system (AFIS) that can be plug-in for any writer-independent handwriting recognition systems. The AFIS starts with an empty rule set. Subsequently, a supervised incremental learning algorithm is operated. When the user reports a misclassification, rule are added or updated. The proposed learning algorithm is evaluated by the adaptation of a writer-independent recognition system (LipiTk). Moreover, the results using a benchmark database named LaViola prove the efficiency of the proposed system. The error rate reduction varies between 66.32% and 41.05%.

In this paper we proposed a writer adaptation system based on an adaptation module that is a plug-in for any writer-independent handwriting recognition systems. The adaptation module is a radial basis function neural network (RBF-NN) that is built using an incremental learning algorithm named GALTM-AM algorithm (Growing-Adjustment with Long-Term Memory). GALTM-AM train a new given data with some LTM data to suppress the interference. Therefore, we design two procedures to manage the LTM data. The first is produce and store. The second is retrieve and learn. This new learning algorithm is evaluated by the adaptation of a writer-independent handwriting recognition system. Moreover, the results using a benchmark database named LaViola prove the efficiency of the proposed GALTM-AM. Performance comparison of GALTM-AM algorithm over the existing approaches is presented.

In this paper we propose a new Hybrid Bat Algorithm to solve the traveling salesman problem (TSP) that has attracted many researchers applying exact and metaheuristic methods trying to solve it. The new proposed method is based on the basics of Bat Algorithm (BA) recently proposed as a new bio-inspired meta-heuristic algorithm. Accordingly, we use the concepts of Swap Operator (SO) and Swap Sequence (SS) to redefine respectively BA position and velocity operators for TSP. Additionally, based on ordered crossover and 3-Opt algorithm, we propose to redefine the Bat's local search method. We compare our algorithm to other state of the art methods from the literature by using benchmark datasets of symmetric TSP from TSPLIB library in order to test its effectiveness. Based on the recorded experiments our method outperforms most of the compared methods.

This paper is a tentative to describe the basics of dynamic optimization using swarm & evolutionary methods. Computational intelligence methods based on swarming, collaborative computing and related techniques showed their potentials at solving classical static problems; for dynamic problems new paradigms needs to be established, this concerns the methods, the test benches and the performance evaluation processes. A review of the key population based computational techniques is performed prior to set some perspective guidelines on how to handle the multi-objective dynamic problems using these technique.

In this paper we proposed a new deep neural network architecture which is composed from a radial basis function neural network (RBF NN) followed by two auto-encoders and softmax classifier and we presented some comparison between this architecture and other architecture on numeral recognition applications. We gave also a review about RBF and sparse auto-encoder neural networks in the literature. First we defined neural networks and their different type's especially radial basis function neural networks (RBF NN) due to their specificity. Second we focused on auto-encoders and sparse coding then we moved to sparse auto-encoders and finally we demonstrated the effectiveness of our deep architecture by showing our experimental results and some comparisons.

The writer adaptation arisen with the appearance and the excessive use of Handheld devices. These devices are conceived to be used in diverse user settings which can be stationary or mobile. Most of the works tackle the writer adaptation in the "sitting at a desk" environment, nevertheless we notice a lack of contributions in the multi-environment context. In this paper we present a multi-environment writer adaptation technique to improve accuracy of writer-independent recognition system. Our system is based on adaptation module (AM) which can greatly decrease error accuracy without changing the writer-independent system. The (AM) is built using IGAAM which is an incremental learning algorithm. First, we test the performance of the IGA-AM on Laviola dataset against GA-AM algorithm for writer adaptation. Second, we test the recognition accuracy by taking into account the writing style change proportionally to environment changes. Thus the system contains as much adaptation module as handled environments. In this paper we consider two stationary environments that are sitting at a desk and standing. Finally, results on multi environment dataset (REGIM-MEnv) are presented.

This paper investigates a Hybrid Naïve Possibilistic Classifier (HNPC) to detect the presence of heart disease from the heterogeneous data (numerical and categorical) of the Cleveland dataset. The proposed classifier stands for the hybridization of two versions of Naïve Possibilistic Classifier (NPC) which have been recently applied on numerical and categorical data, respectively. To estimate possibility beliefs from data, each one of these two versions calls the probability-possibility transformation method of Dubois et al. Later, two fusion steps are performed to make decision. In the first fusion, possibility values are combined within each classifier using the product and the minimum operators for numerical and categorical data, respectively. Then, these two rules are investigated in the second fusion step to combine possibilities assigned to each class. The obtained results show that the proposed HNPC outperforms the main classification techniques which have been used in recent related work.

This paper discusses the use of intelligent technology to solve the problem of grasp planning known as a difficult problem. The scope aims to find points of contact between a five-fingered hand and an object. In this paper, we applied a new hierarchical approach for distributed Multi-Objective Particles Swarms Optimization, based on dynamic subdivision of the population using Pareto fronts (pbMOPSO) for the optimization of the grasp planning problem. The problem is based on simultaneous optimization of two objectives functions. The first objective is to explore the space of skillful manipulation of a robot hand with five fingers and find the best configuration of the fingers by minimizing the distance between the center of mass of the object and the center of the contact polyhedron. The second evaluation function is to maximize another quality measure that is related to the angles defining a configuration of the hand. An experimental study done with the HandGrasp simulator has shown a better performance of our algorithm to solve the grasp planning problem.

This paper investigates a Naïve Bayes Style Possibilistic Classifier (NBSPC) to make decision from the categorical and subjective medical information included by the lymphography dataset of University of California Irvine (UCI). Main focus of the work is to improve the classification accuracy. NBSPC simultaneously relies on the structure of the Naïve Bayes classifier as a good classifier for categorical features, and on the possibility theory as an interesting framework to model and fuse subjective medical data. Possibilistic measures are estimated within the NBSPC using maximum likelihood estimation and then the probability-possibility transformation method of Dubois et al. Results show that the proposed classifier outperforms other classification techniques which have been already evaluated on the same data.

The adaptation module is a Radial Basis Function Neural Network (RBF-NN) that can be connected to the output of any recognition system and its aim is to examine the output of the writer-independent system and produce a more correct output vector close to the desired response. The proposed adaptation module is built using an incremental training named GA-AM algorithm (Growing-Adjustment Adaptation Module). Two adaptation strategies are applied : Growing and Adjustment. The growing criteria are based on the estimation of the significance of the new input and the significance of the nearest unit compared to the input. The adjustment consists of the update of two specific units (nearest and desired contributor) parameters using the standard LMS gradient descent to decrease the error at each time no new unit is allocated. This new training algorithm is evaluated by the adaptation of two handwriting recognition systems. The results, reported according to the cumulative error, show that the GA-AM algorithm leads to decreasing the classification error and to instantly adapting the recognition system to a specific user's handwriting. Performance comparison of GA-AM training algorithm with two other adaptation strategies, based on four writer-dependent datasets, are presented.

Generally real data sets are naturally defined in a fuzzy context. Moreover, in real applications there is no sharp boundary between classes. Therefore, fuzzy clustering is better suited for complex real data sets to determine the best distribution. In this paper we present a new fuzzy learning approach called FMIG (Fuzzy Multilevel Interior Growing Self-Organizing Maps). It is a fuzzy version of MIGSOM (Multilevel Interior Growing Self-Organizing Maps). The main contribution of FMIG is to define a fuzzy process of mappings and take in account the fuzzy criterion of real datasets. This new algorithm is able to auto-organize the map perfectly due to the fuzzy training property of the nodes. Experiment study with synthetic and real world data sets is made to compare FMIG to the crisp MIGSOM and GSOM. Thus, our new method shows improvement in term of quantization error and topology preservation.

This paper belongs to the field of communication and computer networks. Networks of low earth orbiting satellites are able to provide wireless connectivity to any part of the world while ensuring timely and better performance lower bit error rate. This type of technology has been growing interest towards the development of small satellites. Especially, when we talk about the execution of the service quality system, we must use some optimization techniques. However, these systems have the drawback of energy management which is the biggest problem to worry about. Therefore, optimization of the processing time and the effective implementation of information flow and storage on board must be discussed with respect to topology changes fast. In this paper we will discuss various routing algorithms of data used in small satellites and terrestrial networks. As a multiobjective problem, we try to solve the problem of routing data with multiobjective particle swarm optimization (MOPSO).

In this paper we designed an adaptation module (AM) with the objective to increase the performance of a recognition system for a new user or new writing style. The developed adaptation module is added after the recognition system, and its role is to examine the output of the independent system and produce a more correct output vector close to the desired response of the user. To achieve this end, we conceive an adaptation module based on Radial Basis Function Neural Network (RBF-NN) which is built using an incremental training algorithm. Two adaptation strategies are applied for adaptation module training: increase the number of new hidden units and adjust the parameters of the nearest unit (weights and location of center) using the standard descent gradient. This new architecture is evaluated by the adaptation of two recognition systems, one for digit recognition and one for alphanumeric character recognition. The results, reported according to the cumulative error, show that the adaptation module (AM) leads to decreasing the classification error and is capable of fast adaptation to the users handwriting. Moreover, results are compared with those carried out using the weights updating strategy of the nearest center apart from the addition of new units. In fact, the adaptation module decreases an average of 50% the error rate with standard recognition systems.

Self-Organizing Maps (SOM) are widely used for their ability to preserve the topology in the projection. However, this topology is not perfectly preserved due to the static structure of SOM. Therefore, we show in this paper a novel architecture of SOM which organizes itself over time. The proposed method called MIGSOM (Multilevel Interior Growing Self-Organizing Maps) is generated by a growth process which allows to adds nodes where it is necessary. The network start with a minimum number of nodes, then nodes will be added from the boundary as well as the interior of the network. The MIGSOM algorithm adds the interior nodes in a superior level of the map. As a result, the map can have three-Dimensional structure with multi-levels oriented maps. To improve the performance of the proposed algorithm, comparison of MIGSOM to the Kohonen feature Map (SOM) and the Growing Grid (GG) is made. Our experiment results demonstrate that the MIGSOM constructs better mappings than the classic SOM and GG, especially, in terms of data quantification and topology preservation.

This paper introduces a new hierarchical architecture for multi-objective optimization. Based on the concept of Pareto dominance, the process of implementation of the algorithm consists of two stages. First, when executing a multiobjective Particle S warm Optimization (MOPSO), a ranking operator is applied to the population in a predefined iteration to build an initial archive Using ε-dominance. Second, several runs will be based on a dynamic number of sub-populations. Those populations, having a fixed size, are generated from the Pareto fronts witch are resulted from ranking operator. A comparative study with other algorithms existing in the literature has shown a better performance of our algorithm referring to some most used benchmarks.

This paper aims at presenting different strategies for the construction of beta basis function (BBF) fuzzy neural network. These strategies lead to the determination of the network architecture by determining the structure of the hidden layer and parameters of its centers based on data structure. For that, we use self organizing maps (SOM) clustering to construct a mapped structure of the real training data. By analyzing this structure, we proceed to neuron selection. Data sets were also analyzed with the fuzzy c-means (FCM) clustering technique to generate fuzzy membership values presenting fuzzy outputs for our fuzzy neural model. We propose to estimate the parameters of beta basis function in order to obtain better data coverage. Experimental results show that the use of the proposed technique produces better results.

In this paper, we introduce a new variant of growing self-organizing maps (GSOM) based on Alahakoon's algorithm for SOM training; so called 2IBGSOM (interior and irregular boundaries growing self-organizing maps). It's dynamically evolving structure for SOM, which allocates map size and shape during the unsupervised training process. 2IBGSOM starts with a small number of initial nodes and generates new nodes from the boundary and the interior of the network. 2IBGSOM represents the structure of the training data as accurately as possible. Our proposed method was tested on real world databases and showed better performance than the classical SOM and the growing grid (GG) algorithms. Three criteria were used to compare the above algorithms with our proposed method; the quantization error; the topological error and the labeling error to have more accuracy on the produced structure. Results report that 2IBGSOM shows a very good capacity of estimation for the training data based on the three tested factors.

We propose a new feature selection method based on distributed genetic algorithms and bi-coded genes. This solution uses homogeneous and heterogeneous population strategies to minimize the complexity and to accelerate the algorithm convergence. The importance rate is computed for each feature measure to estimate the contribution of each feature in the finale selected vector. A new fitness function was proposed to take into consideration the recognition rate relatively to the size of the selected features subset. Two genetic codes are used to represent each member; a binary code to represent when the corresponding feature was selected or not; the second real code was used to estimate the importance rate of the selected feature or the selection probability for the non selected feature.