Image Processing & Applications

Spatial audio systems have received nowadays great attention not only from the academic sector but also from the market. Many techniques have been studied for the creation of an immersive scenario, but the development of a large zone, where more than one listener can perceive an immersive audio experience, is based on the use of multichannel approaches and sound field technologies. In this context, the presented study aims at an objective analysis of the listeners' sound experience based on their physiological responses and facial expressions to recognize variations in the subject mental and physical status and in their emotional reactions. Several experiments have been conducted using an immersive loudspeakers sphere installed in a semi-anechoic chamber exploiting Ambisonics technology and monitoring the listeners with a camera-based system equipped with a Facial Coding (FAC) and an advanced multimodal acquisition system to measure different physiological signals.

Keywords: facial coding , immersive audio , physiological monitoring , spatial sound

The analysis of molecular interactions between antigens and antibodies is crucial for understanding the immunological mechanisms underlying the immune response and for developing effective therapies against various diseases. In this context, the ability to distinguish between protein interfaces that form stable and unstable complexes is a key step in the design of therapeutic antibodies and vaccines. In recent years, deep learning models have provided advanced tools for biomedical research. This work introduces a novel approach to analyzing antibody-antigen interactions, and in particular SARS-CoV-2 spike protein-targeting antibodies, using a Siamese Neural Network specifically designed to integrate depth maps with geometric descriptors of molecular surfaces. By combining these representations, the model captures geometrical shape complementarity to differentiate between stable and unstable protein complexes. The network was trained using image-based representations of antigens and antibodies interfaces enriched with geometric descriptors, using data that include binders and non-binders of the SARS-CoV-2 spike protein receptor-binding domain. The deep learning network operates by comparing feature vectors representing these molecular surfaces; pairs with closer vectors in feature space are associated with stable interactions, while those with more distant vectors suggest instability. Extensive testing with different configurations achieved an accuracy of 90%, demonstrating the robustness of this approach to predict interactions. This innovative integration of artificial intelligence, depth maps and geometric descriptors offers promising applications for designing novel antibodies and vaccines.

Keywords: Antibodies, Viral , COVID-19 , Deep Learning , Humans , Neural Networks, Computer , Protein Binding , SARS-CoV-2 , Spike Glycoprotein, Coronavirus

Optical tool tracking is the process of determining the 6DoF pose of an object in real time using visual sensor streams and image processing algorithms. It enables spatial localization in applications such as robotics, medical imaging, augmented reality, and precision manufacturing. However, existing solutions often involve tight coupling between hardware and software, complicating the management and benchmarking of different tracking systems. This paper presents the Modular Optical Tool Tracking (MOTT) framework, a unified platform for implementing, integrating, and benchmarking optical tracking solutions. A requirement-based design approach was adopted, using Quality Function Deployment (QFD) to systematically derive technical features and identify design drivers that guided the architecture of the framework. The resulting software framework standardizes the concept of an optical tracking method, featuring a flexible and extensible architecture based on object-oriented principles. Two marker-based tracking methods using an RGB camera as video source were evaluated through the developed framework. The presented case studies showcased the use of the framework for method implementation and comparison within a unified pipeline, reporting computational metrics such as frame rate, CPU and memory usage, and providing pose visualizations. The proposed framework enables standardized evaluation and benchmarking of optical tracking systems, and provides a foundation for future extensions involving non-optical tracking modalities and large-scale comparative studies. The implementation is openly available at https://github.com/tooltip-optical-tracking/mott-framework.

Keywords: Benchmarking , Modular framework , Optical tracking , Pose estimation , QFD , Tool tracking

Hand rehabilitation requires consistent, repetitive exercises that can often reduce patient motivation, especially in home-based therapy. This study introduces ReHAb Playground, a deep learning-based system that merges real-time gesture recognition with 3D hand tracking to create an engaging and adaptable rehabilitation experience built in the Unity Game Engine. The system utilizes a YOLOv10n model for hand gesture classification and MediaPipe Hands for 3D hand landmark extraction. Three mini-games were developed to target specific motor and cognitive functions: Cube Grab, Coin Collection, and Simon Says. Key gameplay parameters, namely repetitions, time limits, and gestures, can be tuned according to therapeutic protocols. Experiments with healthy participants were conducted to establish reference performance ranges based on average completion times and standard deviations. The results showed a consistent decrease in both task completion and gesture times across trials, indicating learning effects and improved control of gesture-based interactions. The most pronounced improvement was observed in the more complex Coin Collection task, confirming the system's ability to support skill acquisition and engagement in rehabilitation-oriented activities. ReHAb Playground was conceived with modularity and scalability at its core, enabling the seamless integration of additional exercises, gesture libraries, and adaptive difficulty mechanisms. While preliminary, the findings highlight its promise as an accessible, low-cost rehabilitation platform suitable for home use, capable of monitoring motor progress over time and enhancing patient adherence through engaging, game-based interactions. Future developments will focus on clinical validation with patient populations and the implementation of adaptive feedback strategies to further personalize the rehabilitation process.

Keywords: gesture recognition , hand rehabilitation , home-based rehabilitation , serious game

Shoulder pain is a common complaint in clinical practice, often requiring MRI for diagnosis and treatment planning. Accurate segmentation of shoulder bones from MRI scans is crucial for 3D anatomical assessment. Deep learning (DL) models have shown promise in automating segmentation, yet previous studies have primarily focused on the humerus and scapula, with limited investigation of the clavicle. This study compares DL models for automatic segmentation of the humerus, scapula, and clavicle from MRI. We explore multiple MRI sequences, including PDW SPAIR (coronal and sagittal) and PD-TSE, to assess their impact on segmentation accuracy. DL architectures, including 3D U-Net and 3D Attention U-Net, were trained, and their hyperparameters were optimized. Performance was assessed using the Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), and volume variation. Additionally, morphological measurements were analyzed to quantify segmentation errors. The best segmentation performance was achieved using a combination of coronal and sagittal PDW SPAIR sequences with Attention U-Net, yielding DSC values of 0.878 (clavicle), 0.843 (scapula), and 0.931 (humerus), and minimal HD. Our study highlights the importance of multimodal MRI input and model optimization for shoulder segmentation. The integration of morphological assessments enhances clinical applicability, providing reliable automated segmentation for surgical planning and medical training.

Keywords: Convolutional Neural Network , Deep Learning , Image Segmentation , Magnetic Resonance Imaging , Shoulder

Tourists are increasingly seeking to satisfy their cultural needs, which is leading to a growing demand for more engaging experiences in heritage tourism. This study investigated whether visitors' emotional engagement and experiences at an art crafts museum could be used to profile them. The ultimate goal, following the digitalization of the collection, is to develop an app that enhances the visitor's experience. A survey was conducted with 218 visitors (113 females) who completed a questionnaire at the end of the visit. A two-step cluster analysis based on visitors' emotions and engagement identified three distinct segments, while regression analyses explored predictors of overall visit satisfaction. The following typologies resulted from cluster profiling based on motivations for visiting, background variables such as prior experience with handcrafts and museums, and sociodemographic variables: ‘Explorers': visitors with lower emotional involvement and a dominant educational motivation; ‘Committed Competent': tourists, passionate about handcrafts and mainly motivated by interest in the area and its culture; ‘Experience Seekers': highly emotionally involved tourists who are satisfied with their visit and mainly motivated by sightseeing. The findings provide simple and easy-to-implement insights for identifying visitor profiles at the museum entrance and for developing an app that can provide personalized tour recommendations.

Keywords: craftsmanship , emotion-based segmentation , Heritage tourists , personalized tours , profiling visitors

The scarcity of medical imaging data poses a significant challenge for the development of automated segmentation models. This study addresses this limitation by focusing on 3D reconstruction of shoulder bones. We evaluated manual and automatic segmentation methods: DL-based using MONAILabel and commercial software (Materialise Mimics), using only 14 CT scans manually labeled for the humerus, scapula, and clavicle. To mitigate data limitations, we applied data augmentation techniques and trained convolutional neural networks optimized for medical imaging, systematically tuning hyperparameters to assess their impact. Segmentation performance was measured using Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD). Our DL model achieved DSC scores of 0.96 (humerus), 0.90 (scapula), and 0.82 (clavicle). Fractures were preserved, enabling accurate 3D reconstruction and measurement. The commercial software achieved similar accuracy. This study highlights the feasibility of using DL for accurate shoulder segmentation in data-constrained scenarios and supports the use of open-source tools as viable alternatives in clinical workflows. Future work will focus on expanding the dataset and further refining the CNN models.

Keywords: Computed Tomography , Convolutional Neural Network , Deep Learning , Image Segmentation , Shoulder

The study of an automated system for intralogistics requires an important use of time and resources, starting from the input data analysis up to the definition of the technical solution. While many commercial tools are available for testing and optimizing the plant performance during the advanced design stages, little work has been done concerning the workflow to be followed at the pre-sales design phase. In this context, the present paper focuses on the definition of the best practices for the correct preliminary definition of a robotic cell for palletization. To simplify and speed up the pre-sales feasibility study and estimate the performance of the proposed robotic system, an engineering approach based on a simplified theoretical model is reported and integrated within a dynamic calculation table. As the main output, the proposed tool calculates the robot saturation which is a key index for the plant preliminary definition.

Keywords: Design Tool , Industry 4.0 , Palletizing Robotic System , Performance Definition , Pre-Sales Design

The Fourth Industrial Revolution conceptualizes the rapid change of industries resulting from the convergence of technologies such as artificial intelligence, genetic editing, and advanced robotics that enable increasing interconnectivity and machines that can analyze and diagnosing problems without human intervention, through intelligent automation. In this scenario, the use of augmented reality technologies is of great interest. The paper aims to explore the use of augmented reality in support of traditional inspections for assisting textile experts in fabric defect detection. The contribution of this study consists of three main phases, necessary for the future development of the system: (1) the analysis of possible automatic defect detection techniques; (2) the analysis of hardware solutions for the realization of a system based on important criteria such as operator comfort, system footprint, and so on; (3) the proposal of a possible comprehensive solution. Considering these aspects this paper identifies and investigate the best scenario for the introduction of artificial intelligence and augmented reality technologies to help the operator in the detection of textile defects.

Keywords: Artificial intelligence , augmented reality , machine vision , textile defect detection

Inverse modelling with deep learning algorithms involves training deep architecture to predict device's parameters from its static behaviour. Inverse device modelling is suitable to reconstruct drifted physical parameters of devices temporally degraded or to retrieve physical configuration. There are many variables that can influence the performance of an inverse modelling method. In this work the authors propose a deep learning method trained for retrieving physical parameters of Level-3 model of Power Silicon-Carbide MOSFET (SiC Power MOS). The SiC devices are used in applications where classical silicon devices failed due to high-temperature or high switching capability. The key application of SiC power devices is in the automotive field (i.e. in the field of electrical vehicles). Due to physiological degradation or high-stressing environment, SiC Power MOS shows a significant drift of physical parameters which can be monitored by using inverse modelling. The aim of this work is to provide a possible deep learning-based solution for retrieving physical parameters of the SiC Power MOSFET. Preliminary results based on the retrieving of channel length of the device are reported. Channel length of power MOSFET is a key parameter involved in the static and dynamic behaviour of the device. The experimental results reported in this work confirmed the effectiveness of a multi-layer perceptron designed to retrieve this parameter.

Keywords: channel length modeling , Deep network , Power MOS inverse modeling

Implementing an alternative reality for Metaverse implies modeling optimized digital content to guarantee real-time interaction and high-quality rendering. Even if 3D reconstruction based on 3D scanning techniques provides a good replica of real objects, the output files are challenging to use in this application. On the other hand, manually developed optimized 3D models require much time and effort. This aspect becomes crucial in scenarios including thousands of 3D models with which humans should interact. This paper proposes a method to automate the 3D modeling process of items whose shapes can be classified according to predefined geometrical categories. The dataset for this study relates to products, which present a wide variety of shapes but are attributable to just a few formal archetypes. In the proposed pipeline, metric orthographic images of the object to be digitally reproduced are analyzed by Convolutional Neural Networks (CNN)s. Subsequently, the same images are analyzed with Computer-Vision (CV) algorithms to extrapolate the characteristic dimensions related to the assigned archetypes. The method has been tested on different items, and the results proved the effectiveness of the whole approach in terms of correct archetypes recognition, parameter extraction, and creation of the 3D model, which are comparable with digitized 3D models with high-quality scanning tools but much lighter in model size.

Keywords: 3D Object Database , Artificial Intelligence , Automatic 3D Modeling , Computer Vision , Convolutional Neural Network , Process Optimization

Background & objective: Human vertebrae are analysed and measured for various purposes, including to study anatomy, to diagnose illness, and to evaluate therapies. Achieving an accurate morphological and dimensional characterisation of three-dimensional (3D) vertebrae relies on the precise recognition of their features. Traditionally, these features, lacking defined edges, have been identified manually through a time-consuming, poorly repeatable, and poorly reproducible process. To the authors' knowledge, there is only one method published in the literature able to automatically recognise all the most important vertebral features: the algorithmic feature recognition method (AFRM). It requires a high-density point cloud as input, the presence of all morphological features, even if incomplete, and incurs high computational costs. This research aims to propose an improved version of the AFRM to overcome its limitations. Methods: The proposed approach combines the robustness of AFRM to provide the semantic segmentation of a 3D healthy human vertebra with the ability of the enhanced statistical shape model (eSSM) to transfer information among different models. Specifically, AFRM provides the semantic segmentation of the mean shape of the eSSM, while the latter transfers this information to target shapes. Results: The eSSM was developed using 20 training samples of healthy adult male L2 vertebrae. The test samples included five healthy vertebrae and four vertebras with large missing parts. None of the test shapes were included in the training set. The novel approach could accurately recognise morphological features without the constraints that affect the AFRM. Conclusion: The proposed method guarantees reliable and automated segmentation through AFRM, exploiting the eSSM's ability to provide results even when dealing with sparsely populated or partially incomplete target models, significantly reducing the computational load.

Keywords: 3D medical image analysis , Bio-informatics , Computer methods for vertebra analysis , Shape segmentation , Statistical Shape Analysis

In the last couple of decades, the research on 3D facial expression recognition has been fostered by the creation of tailored databases containing prototypical expressions of different individuals and by the advances in cost effective acquisition technologies. Though, most of the currently available databases consist of exaggerated facial expressions, due to the imitation principle which they rely on. This makes these databases only partially employable for real world applications such as human-computer interaction for smart products and environments, health, and industry 4.0, as algorithms learn on these ‘inflated' data which do not respond to ecological validity requirements. In this work, we present two novel 2D + 3D spontaneous facial expression databases of young adults with different geographical origin, in which emotions have been evoked thanks to affective images of the acknowledged IAPS and GAPED databases, and verified with participants' self-reports. To the best of our knowledge, these are the first three-dimensional facial databases with emotions elicited by validated affective stimuli.

Keywords: 3D facial expression , Affective database , Ecological validity , Facial expression recognition , Human-computer interaction , Spontaneous expressions

The purpose of the study was to evaluate the performance of radiomics analysis of MR images for the detection of prostate cancer. The radiomics analysis was conducted using axial T2-weighted images from 49 prostate cancers. The study employs a sophisticated hybrid descriptive-inferential method for the meticulous selection and reduction of features, followed by discriminant analysis to construct a robust predictive model. Among 71 radiomics features, original_glrlm_ShortRunLowGrayLevelEmphasis demonstrated exemplary performance in differentiating between the whole prostate gland and prostate cancer. It had an AUROC of 68.46 (95% CI 0.544 – 0.824; p = 0.022), sensitivity of 76.25%, specificity of 73.15%, and accuracy of 71.02%. Radiomic analysis of T2 weighted MR images was demonstrated to have clinical application in prostate cancer detection, paving the way for improved diagnostic procedures and tailor-made treatment plans for prostate cancer patients.

Keywords: Magnetic Resonance Imaging , Prostate Cancer , Radiomics , Texture Analysis

Protein–protein interaction occurs on surface patches with some degree of complementary geometric and chemical features. Building on this understanding, this study endeavors to characterize the spike protein of the SARS-CoV-2 virus at the morphological and geometrical levels in its Alpha, Delta, and Omicron variants. In particular, the affinity between different SARS-CoV-2 spike proteins and the ACE2 receptor present on the membrane of the human respiratory system cells is investigated. To achieve an adequate degree of geometrical accuracy, the 3D depth maps of the proteins in exam are filtered by developing an ad-hoc convolutional filter with a kernel implemented as a sphere of varying radius, simulating a ball rolling on the surface (similar to the 'rolling ball' filter). This ball ideally models a hypothetical molecule that could interface with the protein and is inspired by the geometric approach to macromolecule–ligand interactions proposed by Kuntz et al. in 1982. The aim is to mitigate the imperfections and to obtain a smoother surface that could be studied from a geometrical perspective for binding purposes. A set of geometric descriptors, borrowed from the 3D face analysis context is then mapped point-by-point onto protein depth maps. Following a feature extraction phase inspired by Histogram of Oriented Gradients and Local Binary Patterns, the final histogram features are used as input for a Support Vector Machine classifier to automatically classify the proteins according to their surface affinity, where a similarity in shape is observed between ACE2 and the spike protein of the SARS-CoV-2 Omicron variant. Finally, Root Mean Square Error analysis is used to quantify the geometrical affinity between the ACE2 receptor and the respective Receptor Binding Domains of the three SARS-CoV-2 variants, culminating in a geometrical explanation for the higher contagiousness of Omicron relative to the other variants under study.

Full-field methods can significantly improve measurement quality in experimental dynamic analysis, which is a critical issue in the industry. Optical methods based on digital image correlation (DIC) are widely used in this regard. This paper aims to present an experimental setup to perform full-field vibration measurements using off-the-shelf components costing tens or hundreds of euros rather than thousands. This allows for the construction of multiple measurement systems with a small investment, giving students hands-on experience. The system's inherent low cost, ease of assembly, and ease of use ensure that the student can handle the system from the design stage to the setup stage, and finally to the testing stage. A simple Matlab app was developed to set up and control the test, analyze data and display the results. The system's modularity allows it to further extend measurement capabilities over time, performing 2D measurements or more complex 3D measurements under single or multiple inputs. A prototype of the proposed system was assembled and tested on a planar specimen for 2D DIC measurements. The total cost of the equipment was less than 250 €. The setup was validated for geometrically complex torsional deformed shapes up to 660 Hz.

Keywords: Digital Image Correlation , Educational Test Bench , Full Field Vibration Measurement

Segmentation of anatomical components is a major step in creating accurate and realistic 3D models of the human body, which are used in many clinical applications, including orthopedics. Recently, many deep learning approaches have been proposed to solve the problem of manual segmentation, which is time-consuming and operator-dependent. In the present study, SegResNet has been adapted from other domains, such as brain tumors, for knee joints, in particular, to segment the femoral bone from magnetic resonance images. This algorithm has been compared to the well-known U-Net in terms of evaluation metrics, such as the Dice similarity coefficient and Hausdorff distance. In the training phase, various combinations of hyperparameters, such as epochs and learning rates, have been tested to determine which combination produced the most accurate results. Based on their comparable results, both U-Net and SegResNet performed well in accurately segmenting the femur. Dice similarity coefficients of 0.94 and Hausdorff distances less than or equal to 1 mm indicate that both models are effective at capturing anatomical boundaries in the femur. According to the results of this study, SegResNet is a viable option for automating the creation of 3D femur models. In the future, the performance and applicability of SegResNet in real-world settings will be further validated and tested using a variety of datasets and clinical scenarios.

Keywords: 3D modeling knee , automated segmentation , CNN , computer-aided design , deep learning , machine learning for engineering applications , MONAI framework , SegResNet , U-Net

The correct execution of scanning protocols is crucial to ensure the quality of radiographic examinations. Recent advances in machine learning methods have opened up new possibilities for medical imaging. However, the potential of pose estimation models in this field is still largely unexplored. This study aims to address this gap by investigating the performance and an application of pose estimation in the context of X-ray image acquisition. To this goal, a pose estimation model was selected from a pool of state-of-the-art models. It was then trained on a dataset of 213 images of humans undergoing X-ray imaging. Despite the limited size of the dataset, the model achieved an AP of 0.902 and a near real-time inference speed of 7 FPS on CPU. The detection of landmarks through pose estimation enables the automatic assessment of pose adherence to prescribed imaging protocols. This automation can reduce human errors and alleviate the mental workload on radiologists. The results of this study highlight the potential of convolutional neural network-based pose estimation models to assist radiologists in performing X-ray imaging tasks effectively.

Keywords: Convolutional Neural Networks , Deep Learning , Optical Imaging , Pose Estimation , Radiography

Survey missions in the underwater environment are key activities in several application fields, but the physical properties of such a harsh environment make it challenging for both navigation and data collection. In particular, in the field of underwater photogrammetry, optical sensors are mainly installed on remotely operated vehicles (ROVs). The maneuvering is performed remotely by expert pilots who can only rely on bidimensional visual feedback from one or more camera streams. Additional navigation data can aid pilots in assessing if the mission meets the requirements while undergoing it. To this end, the paper presents a SLAM-based solution that can be used to compute parameters in support of ROV pilots during photogrammetric surveys, when a stereo-camera is equipped. In order to assess the performance of the proposed solution, different tests were conducted. These tests involved comparing the parameters derived from SLAM with the data computed by the navigation system of a professional-grade ROV. The focus of this comparison was on three key parameters: position, velocity, and distance from the target.

Keywords: Multi-camera System , Underwater 3D Reconstruction , Underwater Photogrammetry , Underwater SLAM , Underwater Survey

Objectives: Nowadays, a wide variety of software for 3D reconstruction from CT scans is available; they differ for costs, capabilities, a priori knowledge, and, it is not trivial to identify the most suitable one for specific purposes. The article is aimed to provide some more information, having set up various metrics for the evaluation of different software's performance. Methods: Metrics include software usability, segmentation quality, geometric accuracy, mesh properties and Dice Similarity Coefficient (DSC). Five different software have been considered (Mimics, D2P, Blue Sky Plan, Relu, and 3D Slicer) and tested on four cases; the mandibular bone was used as a benchmark. Results: Relu software, being based on AI, was able to solve some very intricate geometry and proved to have a very good usability. On the other side, the time required for segmentation was significantly higher than other software (reaching over twice the time required by Mimics). Geometric distances between nodes position calculated by different software usually kept below 2.5 mm, reaching 3.1 mm in some very critical area; 75th percentile q75 is generally less than 0.5 mm, with a maximum of 1.11 mm. Dealing with consistency among software, the maximum DSC value was observed between Mimics and Slicer, D2P and Mimics, and D2P and Slicer, reaching 0.96. Significance: This work has demonstrated how mandible segmentation performance among software was generally very good. Nonetheless, differences in geometric accuracy, usability, costs and times required can be significant so that information here provided can be useful to perform an informed choice.

Keywords: 3D reconstruction , CT scans , DSC , Geometric accuracy , Mandible , Segmentation , Usability