Image Processing & Applications

This paper introduces a Unity-based Virtual Reality (VR) system designed for real-time, incremental acquisition, processing, and interactive spatial filtering of LiDAR-generated point clouds on the Meta Quest 3 platform. Using an Intel RealSense L515 sensor and SLAM-inspired alignment techniques, the system enables dynamic, user-defined spatial segmentation through bounding boxes to refine collected data. The proposed framework supports multi-pass acquisitions, merging, and cleaning of point clouds, effectively addressing the limitations of traditional batch processing workflows. Additionally, the system provides an efficient modular structure that integrates live visualization using the Pcx Unity asset and optimizes performance through binary PLY file conversion. A case study on capturing a Meta Quest 3 protective case demonstrates the system's robustness and ability to preserve geometric fidelity across incremental scans, with minimal redundancy and automatic alignment. The results highlight the potential for future expansion into real-time volumetric reconstruction, automated point cloud registration, and advanced spatial computing applications. By combining modular software architecture with accessible hardware configurations, this work contributes a practical framework for immersive 3D data handling, aligning with ongoing research in real-time 3D reconstruction, interactive VR applications, and advanced point cloud processing.

Keywords: 3D Scanning , Point Cloud Processing , Virtual Prototyping , Virtual Reality

Burn scars represent a significant clinical challenge, owing to their morphological complexity and their functional and aesthetic impact on the patient. Traditionally, the assessment of such scars relies on subjective scales, which are prone to inter-observer variability. This study proposes an innovative approach that integrates colorimetric analysis with artificial intelligence models to provide an objective and reproducible assessment of burn scars. To this end, two novel colorimetric indices—the Erythema Index (EI) and the Vascularity Index (VI)—were introduced. These indices, when combined with machine learning algorithms, enable the automation of scar analysis, yielding promising results in terms of accuracy and reliability.

Keywords: Burn scars , Colorimetric analysis , Colorimetric indices , Machine learning , Objective assessment

Physical ergonomics is critical for worker well-being, safety, and productivity, in Industry 4.0 and in the emerging Industry 5.0 paradigm, which emphasize human-machine collaboration. Advances in motion capture technologies, especially marker-less systems using RGB cameras, offer new possibilities for real-time ergonomic assessments. These systems effectively address the intrusiveness issues inherent to marker-based approaches and partially overcome challenges such as lighting variations, occlusions and calibration commonly associated with RGB-D camera-based systems. Nonetheless, their accuracy can be suboptimal due to limitations arising from perspective distortions. To overcome this issue, the present paper introduces an “inductive” RGB camera-based approach to ergonomic assessment, leveraging Deep Learning models trained on a synthetic dataset of 14,400 unique postures captured from 24 viewpoints in Unity 3D. Unlike traditional methods reliant on measured angles, this approach predicts ergonomic evaluations directly from posture observations. We implemented the proposed approach on a prototypal system that allowed us to experiment and collect quantitative results. We analysed the system's performance through comparative testing, juxtaposing predictions from our system with measurements obtained from an inertial system worn by individuals while performing tasks in a controlled laboratory environment. Experimentations highlighted the satisfactory performance of the proposed inductive approach and related technology, highlighting its potential for delivering more accurate real-time ergonomic assessments.

Keywords: Deep learning , Ergonomics , Human centred manufacturing , RGB motion capture , Synthetic datasets

Recent studies show that most Work-Related Musculoskeletal Disorders (WMSDs) of the hand and wrist are the result of repeated stresses, caused by repetitive activities over a long time. Tools like ergonomics risk indexes, have been developed to assess upper limb overload in repetitive tasks: part of this evaluation is based on the observation of the number and frequency of Technical Hand Grips (THGs), differing by the number of fingers involved, the exerted forces and most affected regions. Ergonomists typically assess THGs manually via video review, which is time-consuming: a need arises for the automation of these procedures. The rapid and functional creation of AI models capable of classifying efficiently THGs presents some particular challenges: (1) labeling a huge amount of images at ergonomist-level, since, to the best of our knowledge, there are no useful datasets available; (2) considering object occlusions as an intrinsic requirement of realistic study of THGs; (3) moving towards the shared request of data anonymization. This work proposes an Active Learning system in which an initial model labels THGs from videos and consults an Oracle (ergonomist) only when uncertain. An implemented AI (GNN) model is trained on these carefully labeled data to replace the first rudimentary model. The process repeats until the desired model accuracy or data set size is reached. This has guaranteed: (1) a rapid labeling strategy and scaling of the THGs classification model that learns from selected “good” data; (2) implement an occlusion-robust model, by applying selective masking strategies; (3) anonymization through graph-structured data.

Keywords: Active Learning , Ergonomics , Graph Neural Network , OCRA Index , Technical Hand Grip Classification

While research on facial expressions has traditionally focused on basic emotions, other spontaneous reactions, especially those elicited through emotional contagion, are far more common in daily social interactions, with facial movements playing a key role in conveying emotions. Here, we explore spontaneous facial expressions of laughter, yawning, and mirror pain through a contagion experiment involving 32 participants. We identified key facial action units and used landmark-based distances as morphometric features to capture their dynamics. Our analysis revealed distinct patterns that separate these expressions from each other and from neutral faces, particularly in the lower face. This work provides a curated set of action units and features for the assessment of laughter, yawning, and mirror pain expressions, offering a valuable resource for future spontaneous facial expression recognition studies.

Keywords: psychology , social sciences

Cutting guides are widely used in cranio-maxillofacial surgery by providing mechanical references for precise bone resections and reducing intraoperative variability. Nevertheless, their rigid and patient-specific design requires dedicated CAD modeling and fabrication, making them time-consuming to produce and difficult to adapt when anatomical conditions or bone surfaces change. This work presents a hybrid surgical cutting guide that combines a physically adjustable device with Augmented Reality (AR) feedback to support intraoperative alignment in maxillofacial osteotomies. The concept merges the tactile reliability of conventional guides with the adaptability of digital visualization, enabling surgeons to fine-tune the cutting plane directly through AR tracking. Registration was performed using cephalometric landmarks and an inside-out tracking approach with HoloLens 2, allowing precise superimposition of virtual cutting planes onto 3D-printed mandibular models. The system was evaluated by both expert and novice operators under three feedback conditions: no AR, holographic overlay, and real-time distance guidance. Results showed that AR feedback was associated with improved positional accuracy, with mean linear deviations of 1.27±0.71mm and angular errors of 4.46±3.27°. Operator experience influenced overall performance, yet enhanced feedback compensated part of this variability. Combining physical and digital guidance can yield more adaptable, precise, and reusable osteotomy tools, paving the way for flexible surgical assistance in clinical settings.

Keywords: Adjustable surgical guide , Augmented Reality , Cranio-maxillofacial surgery , Hololens 2 , Hybrid surgical guide , Pose estimation

Deafness poses significant challenges to effective communication, particularly in contexts where access to sign language interpreters is limited. Hand configuration recognition represents a fundamental component of sign language understanding, as configurations constitute a core cheremic element in many sign languages, including Italian Sign Language (LIS). In this work, we address configuration-level recognition as an independent classification task and propose a machine vision framework based on RGB-D sensing. The proposed approach combines MediaPipe-based hand landmark extraction with normalized three-dimensional geometric features and a Support Vector Machine classifier. The first contribution of this study is the formulation of LIS hand configuration recognition as a standalone, configuration-level problem, decoupled from temporal gesture modeling. The second contribution is the integration of sensor-acquired RGB-D depth measurements into the landmark-based feature representation, enabling a direct comparison with estimated depth obtained from monocular data. The third contribution consists of a systematic experimental evaluation on two LIS configuration sets (6 and 16 classes), demonstrating that the use of real depth significantly improves classification performance and class separability, particularly for geometrically similar configurations. The results highlight the critical role of depth quality in configuration-level recognition and provide insights into the design of robust vision-based systems for LIS analysis.

Keywords: Italian Sign Language (LIS) , machine learning , MediaPipe , RGB-D cameras , sign-language recognition

Musculoskeletal disorders are frequent workplace injuries, especially during manual lifting activities. They are influenced by posture, lifting technique, and repetitive movements. Various ergonomic assessment methods exist, but each has limitations: observational methods can be slow and prone to error, while contact sensor-based methods, although more accurate, tend to be invasive and expensive. Recent developments have focused on non-contact sensors, such as RGB and RGB-D cameras, combined with Deep Learning algorithms and observational methods, to improve efficiency and reliability. This study proposes a solution combining a skeleton-based Deep Learning algorithm for Human Pose Estimation with an observational method for postural assessment. Using an RGB camera, four lifting techniques (stoop, squat, semi-squat, and weightlifter) were analyzed, evaluating their impact on worker posture through the REBA score. Among handle-assisted lifts, the stoop and weightlifter techniques showed the lowest average maximum REBA scores (5.375 and 6.125), while the squat and semi-squat techniques scored highest at 7. The semi-squat without handles showed the greatest postural risk (7.875). Future work will integrate 3D data and validate the approach with a larger, more diverse population.

Keywords: Deep learning , Ergonomics , Industry 5.0 , Postural analysis

Artificial intelligence-powered assistants are revolutionizing the music industry by transforming how music is produced and experienced, opening new frontiers for creativity and innovation. Their application in supporting learners to play a musical instrument, however, has yet to be fully explored. Most current applications primarily focus on sound data to distinguish between different notes. This approach excludes the correct hand form, which is essential for learning to play an instrument. Furthermore, sound can be subject to background noise or compression. The current article presents a pipeline for guitar chord recognition based on 3-D data acquisition and color images. The developed method utilizes MediaPipe to estimate hand landmarks from color images and leverages depth images to retrieve real-depth information. In this study, various machine learning (ML) algorithms were compared to perform chord recognition from hand landmark information. The number of chords was limited to four, plus one class for unknown gestures. The classifier was trained with an RGB-D video data set (red, green, and blue plus depth) that features the hands of 18 subjects who performed the selected chords. The random forest (RF) classifier demonstrated remarkable performance in the classification task, achieving a balanced accuracy (BA) that exceeded 85% on unseen data under the same acquisition conditions and improving the state-of-the-art performance. Future developments will focus on expanding the number of supported chords so that the proposed approach could be used in real-world applications. Such applications might include not only guitar education but also transcription, control of sound synthesis, identification in ensembles, and other contexts where audio input alone is insufficient.

Keywords: Audio acoustics , Color , Color image processing , Computer music , Computer vision , Data acquisition , Learning algorithms , Learning systems , Palmprint recognition

Early and reliable identification of vertebral metastases on computed tomography remains a major challenge in oncologic imaging due to the morphological complexity of metastatic lesions and the high inter-patient variability of spinal anatomy. In this study, an end-to-end interpretable radiomic-based framework was developed to automatically distinguish healthy from metastatic vertebrae using segmented DICOM data, coupled with an interactive virtual reality (VR) visualization module implemented in Unity 3D. The proposed framework integrates radiomic feature extraction and selection, informed undersampling to address class imbalance, and automatic machine learning-based classification. To facilitate interpretation, patient-specific 3D models with overlapped classifier outputs were integrated into a VR desktop application, enabling advanced exploration of patient-specific spinal models, with color-coded visualization of algorithmic predictions and expert-defined suspicious lesions. The final classification model, trained using a Random Forest algorithm and optimized via stratified 5-fold cross-validation, achieved an overall accuracy of 0.86, an Area Under the Receiver Operating Characteristic Curve of 0.91, and an F1-score of 0.81 for the metastatic class on the independent test set, achieving competitive diagnostic performance while preserving transparency and clinical interpretability. This study represents a foundational step toward intelligent, interactive, and clinically interpretable tools for the diagnosis and follow-up of spinal metastatic disease.

Keywords: automatic lesion classification , bone lesions , machine learning , medical 3D visualization , spinal metastases , virtual reality

Digital Image Correlation applied to rotating objects is a challenging task due to the extremely large displacements to be detected and to the rotation of the subsets during revolutions. For these reasons, conventional DIC algorithms are limited to about 7° rotations before losing tracking capabilities. Nevertheless, a full 360° rotation is to be tracked in practical scenarios, so that a new paradigm in optical tracking of rotating objects remains urgent. In this paper, a global DIC algorithm is proposed to overcome the limitations of conventional methods. A simple analytical representation of the roto-translation is proposed, and an optimization algorithm is developed to find the rotation angle and center to minimize the discrepancies between the reference and rotated image. The critical issue of the procedure is represented by the selection of a reliable first guess, which, due to the presence of several local minima and a really sharp global minimum, needs to be extremely accurate for the optimization to converge. To this extent, a robust procedure is proposed based on image unwrapping and spatial-FFT analysis of the unwrapped image to infer a reliable first guess to be fed to the subsequent final optimization process. The approach was validated through experimental analysis of a rotating PC fan.

Keywords: Digital Image Correlation , Image unwrapping , Optical vibration measurements , Rotating DIC

Full-field methods such as Electronic Speckle Pattern Interferometry (ESPI) and Digital Image Correlation (DIC) emerge as promising alternatives to strain rosette in the hole-drilling method, each with its own advantages and limitations. In particular, DIC stands out as a method requiring minimal (and cheap) specimen preparation; however, its sensitivity is approximately one order of magnitude lower than that of conventional strain gauge techniques. We address two fundamental questions. First, how can the inverse problem of the hole-drilling method be systematically approached outside the well-established framework of ASTM E837, which provides users with comprehensive guidance, including automated regularization procedures? Second, thanks to full-field measurement techniques, is it possible to retrieve stress distributions at a spatial resolution finer than the drilling step? We revisit the theoretical development of the inverse problem, aiming to directly identify the residual stresses from raw images, without intermediate pre-processing of the displacement fields. We introduce the L-curve method – well established in other scientific domains – as a rational tool for the choice of the regularization parameter. Finally, we show that, in principle, it would be possible to identify the entire stress distribution from a single drilling increment. We perform a deep rolling treatment on an aluminum specimen and compare the residual stresses identified using both a strain gauge rosette and DIC. We demonstrate that the proposed framework yields results comparable to those obtained using strain gauge rosettes, while minimizing user-dependent arbitrariness. Furthermore, we show that depth super-resolution of at least a factor of two is already achievable with current technological capabilities.

Keywords: Digital image correlation , Full-field , Hole-drilling , Regularization , Residual stress

Digital Image Correlation (DIC) is a widely used non-contact method for full-field displacement measurement under static and dynamic conditions. However, conventional algorithms struggle with large rotations, typically limited to 10-15°, making them unsuitable for rotating objects like wheels, turbines, or gears. Current workarounds involve using databases of static images at various angles to find a suitable reference image, but this approach lacks flexibility. This research presents a global DIC algorithm capable of tracking rotating objects without relying on image databases. A grid of points is defined on the reference image and numerically rotated around a specified center. Bicubic spline interpolation is used to estimate grey levels at non-integer positions in the deformed image. These are compared with reference values to compute an error, minimized through an optimization process based on a Gauss-Newton scheme. The method was validated on experimental images of a rotating PC fan, showing effective tracking for angles up to 90°.

Mandibular reconstruction, particularly with temporomandibular joint (TMJ) involvement, is a challenging procedure where precise evaluation of condylar displacement is critical for assessing functional and anatomical outcomes. The complexity of this evaluation stems from anatomical variability, surgical techniques, and post-surgical changes, making consistent and reliable assessments difficult. To address this issue, we developed an innovative, standardized procedure for the evaluation of condylar displacement following mandibular reconstruction. The proposed procedure starts from biomedical image analysis and integrates precise segmentation with an innovative standardized alignment technique, ensuring reproducibility across different reconstructive approaches, regardless of the surgical method used. This procedure was validated through a retrospective study of 18 patients, employing a consistent 3D model reconstruction and analysis workflow. The results demonstrate that the developed procedure provides a reliable framework for comparing different reconstructive methods, regardless of whether they employ CAD/AM techniques or conventional approaches, offering a repeatable and adaptable approach for assessing mandibular reconstruction outcomes. This methodology lays the groundwork for future improvements in surgical planning and post-operative evaluation, ultimately contributing to the optimization of both functional and aesthetic results in mandibular reconstruction.

Keywords: 3D model reconstruction , condylar displacement evaluation , mandibular reconstruction , standardized alignment procedure

Biofouling, the phenomenon involving the growth and accumulation of marine organisms on submerged surfaces, poses a significant challenge for the conservation of underwater structures and historical artifacts. Its presence can accelerate material degradation and complicate potential restoration efforts, especially in cases where direct recovery from the marine environment is not feasible. In this context, the early identification of biofouling is crucial for developing targeted intervention strategies and minimizing long-term damage. This study aims to lay the groundwork for creating a training dataset for a deep learning-based segmentation network. Images of submerged specimens were captured in both the visible spectrum and under UV illumination, which reveals biofouling often invisible to the naked eye. Using semi-automatic thresholding methods, segmentation masks were generated from UV images to approximate biofouling regions. RGB, HSV, Lab, and YCbCr color spaces were systematically compared to determine the most stable and reliable color model for segmentation. Results identified the HSV space as offering the most consistent thresholding performance across the dataset. These segmentation masks provide a foundation for training deep learning models aimed at automatically detecting biofouling in visible-light images, where manual annotation is challenging.

Keywords: Biofouling detection , Computer vision , Image segmentation , Underwater cultural heritage

Purpose: To compare semiquantitative striatal indices produced by two widely used quantification platforms — BasGanV2™ and NeuroTrans3D (Oasis

Keywords: BasGanV2 , Dopamine transporters , Movement disorders , Neurology , Oasis© , Parkinson's disease , Personalized medicine , Quantitative analysis , Rehabilitation , ¹²³I-Ioflupane

Currently, Additive Manufacturing is revolutionizing the production of complex and customized components across various industries, offering significant advantages in material efficiency, design flexibility, and rapid prototyping. Concurrently, Machine Learning models have become increasingly crucial in Additive Manufacturing, improving decision-making, process efficiency, predictive accuracy, and defect recognition. However, the practical implementation of Machine Learning models in production environments presents significant challenges, such as managing incremental knowledge without causing catastrophic forgetting of old knowledge. The complexity of the additive process and the large number of parameters often require training the model on new data without forgetting previously acquired knowledge. Continual Learning is an emerging practice for incremental knowledge management in Artificial Intelligence. This paper presents a novel Continual Learning approach for class-incremental learning tasks, called Progressive Online Ridge Regression (PORR), based on an extended version of Ridge Regression that fine-tunes a pre-trained Convolutional Neural Network, MobileNetV3. The method is applied to an image analysis problem to recognize defects in Powder Bed Fusion of Polymers. An open-access dataset has been analyzed to validate the approach. The experimental results demonstrate that the proposed approach reduces catastrophic forgetting by optimizing computational resource allocation with respect to accuracy, training time, CPU utilization, and maximum RAM usage.

Keywords: Additive manufacturing , Continual learning , Defect recognition , Machine learning , Powder bed fusion of polymers

The increasing use of geopolymers in the construction industry, driven by the need for eco-friendly production processes and the gradual replacement of Portland cement in such applications, leads to new scientific challenges. In particular, the availability of reliable tools to predict the mechanical response of this class of materials is becoming increasingly crucial for their application in a wider range of engineering solutions. In the context of this research work, and as a contribution to addressing these needs, a modeling framework based on a morphologically informed Representative Volume Element was developed and proposed. To this end, an X-ray tomographic technique was employed on geopolymer samples, and internal defects were quantified through morphometric image analysis. The resulting data were subsequently integrated into a Finite Element Method (FEM) code to build a Representative Volume Element incorporating a distribution of equivalent porosity. Therefore, numerical simulations were carried out using an explicit FEM code to reproduce a compressive mechanical test, and the results were compared with experimental data. Good agreement was observed between the experimental and numerical results, both in the linear elastic regime and during the damage initiation and failure phases of the specimens.

Keywords: CT scan , FEM , Geopolymers , morphometry , Representative Volume Element

Hollow Fiber Membrane Bioreactors (HFMBs) support mammalian cell culture at high cell density by enabling effective transport of nutrients, gases and metabolites to/from cells and by offering high membrane surface area-to-bioreactor volume ratios, scalability and flexibility in design. A recent study showed that shell-and-tube HFMBs with cells in the extracapillary space (ECS) for medicine may be designed to mimic the bone architecture for bone tissue engineering (TE). A more transport-efficient HFMB, the BRx-HFMB, consists of a 3D stack of alternating cross-woven mats of microporous and gas-permeable hollow fiber membranes with cells in the ECS and medium flowing inside the membranes. This design was shown to support the culture of a large mass of densely packed cells with high oxygen and nutrient metabolic demands similar to bone cells. Its biomimicry and rationale for good performance are poorly characterized, hindering exploitation of its characteristics for bone TE. Herein, we report the architectural and biomimetic characterization of pore/void distribution in the ECS of laboratory BRx-HFMBs using non-invasive non-destructive micro-computed tomography and advanced image analysis to minimize artifacts. The results suggest that the ECS pore architecture and specific surface area of BRx-HFMBs mimic those of bone tissue, favoring cell migration around, and adhesion on, membranes at clinical cell densities. The membrane network architecture enables cell perfusion with medium, and membranes act as spatially distributed oxygen sources promoting oxygen transport through the cell construct over longer distances than in static bioreactors. This supports the use of BRx-HFMBs to develop cellular models of bone tissue for drug testing and precision medicine. This journal is

Non-invasive assessment of spinal morphology traditionally relies on subjective manual palpation or data-driven deep learning models that often lack metrological interpretability. This study presents a fully automated, purely geometric framework for the localization and morphometric characterization of spinous processes directly from 3D dorsal surface scans. Leveraging continuous differential geometry and the scale-invariant properties of the shape index, the proposed algorithm operates independently of training data, physical markers, and ionizing radiation. Validation against expert clinical palpation demonstrated high spatial reliability, yielding a mean spatial deviation of 3.7 mm and a 98.6% Confidence Index in matching manually identified landmarks. Notably, the computational framework exhibited heightened morphological sensitivity, consistently extracting mathematically valid geometric signatures in spinal regions conservatively bypassed by the clinician. This resulted in a theoretical algorithmic False Negative Rate bounded between 13.5% and 14.6%, compared to a 33.1% manual rate. Furthermore, the system autonomously evaluated morphometric parameters, such as basal width, symmetry index, and local apical prominence, successfully classifying 327 apophyses into “prominent” (19%) and “masked” (81%) morphotypes across complex, soft-tissue-mediated topographies. While these findings highlight the method's capability to standardize and objectify dorsal surface assessment, definitive anatomical fidelity requires future radiological correlation. Ultimately, this robust topographic profiling lays the foundation for autonomously quantifying structural asymmetries in clinical spinal deformities.

Keywords: 3D surface topography , Automated detection , Differential geometry , Morphometrics , Non-invasive spinal assessment , Spinous process localization