Image Processing & Applications

The field of robotic manipulation has seen ever-growing interest in recent years, with the aim of increasing process efficiency and safety for human operators. The main challenge lies in grasping objects with variable mechanical characteristics such as dimensions and geometries. In fact, the gripping solutions currently available tend to be highly task-specific, with limited adaptability to different operational scenarios. To bridge this gap, the present study proposes a four-finger gripper with a reconfigurable architecture to ensure versatility. The device has been validated through the use of a virtual prototype, demonstrating the ability to grasp objects with different shapes and masses, as well as multiple objects, thanks to five different grip configurations and the presence of soft elements. The fingertip design includes the possibility of integrating sensors, with the goal of enabling advanced sensing and feedback functions during the manipulation phase. This aspect opens the door to the implementation of adaptive control strategies and to improved operational safety in dynamic environments. The potential applications range from automated logistics to industrial assembly, from collaborative robotics to service robotics, including sectors where it is essential to manipulate heterogeneous objects in a precise and non-destructive manner. In particular, the ability to handle delicate or sensitive components has made the device suitable for applications in highly specialized fields such as agri-food and pharmaceuticals.

Keywords: Adaptive Grasping , Reconfigurable Gripper , Robotic Manipulation

Progressive stiffness has been proposed as a solution to the common trade-off in the design of series elastic actuators: higher torque-tracking performance, achieved by stiffer springs, or safer physical interaction with the environment and enhanced torque transparency, offered by softer springs. In this work, we investigate the application of series elastic actuators with stiffening characteristic in the context of dynamic legged locomotion. A planar one-legged robot with an articulated leg is modeled and simulated using a multi-body dynamics framework, with a control strategy based on the Spring-Loaded Inverted Pendulum template. The actuators incorporate compact, monolithic compliant mechanisms based on flexure beams that exhibit a progressive torque deflection characteristic. The design of these mechanisms is parameterized in terms of length, thickness, and number of beams, and then optimized with the goal of reducing both the joints' torque requirements and the mechanical cost of transport during a running task. Simulations results show that, compared to stiff actuation and linear series elastic actuation configurations, properly tailored nonlinear stiffness profiles can improve actuation efficiency and reduce energy requirements in dynamic legged locomotion.

Keywords: Compliant Mechanism , Energy Efficiency , Legged Robot , Nonlinear Series Elastic Actuators , SLIP

Purpose: This mini-review provides an updated overview of SPECT-based radiomics in oncology, summarizing clinical applications, highlighting SPECT-specific technical considerations, and identifying priorities for future research. Methods: A structured search of MEDLINE and Scopus was performed using the terms “SPECT,” “radiomics,” and “oncology.” Only original clinical studies published in English within the last five years were included. Eligible articles were screened and grouped by oncologic topic. Results: Eleven studies met the inclusion criteria. The reported applications included differentiation of metastatic versus benign bone lesions using bone SPECT/CT, prediction of survival in neuroblastoma with 123I-mIBG SPECT/CT, and characterization of indeterminate renal masses with 99mTc-sestamibi. Early work in radiodosiomics showed potential for estimating absorbed doses during peptide receptor radionuclide therapy and predicting response to selective internal radiation therapy. Conclusions: SPECT radiomics and radiodosiomics show promise for quantitative oncologic imaging, but the evidence base remains limited by small, retrospective, single-center studies and the complete absence of external validation. Standardizing acquisition and reconstruction protocols, and multicenter studies are essential to enable translation of SPECT-based radiomics into routine clinical practice.

Keywords: Oncologic imaging , Quantitative imaging , Radiodosiomics , Radiomics , SPECT , SPECT/CT , Texture analysis

This work explores the automation of AutoCAD workflow using Python scripting, with a focus on the different Python libraries available for such tasks. The article analyzes four prominent Python packages—win32com, pyautocad, PyAutoGUI, and ezdxf—assessing their strengths and weaknesses in the context of AutoCAD automation. To evaluate the performance of these libraries, a Python application was developed for the simple task of prefixing layer names in an AutoCAD drawing, testing each library's effectiveness in accomplishing this task. The findings indicate that each Python library offers distinct advantages depending on the automation requirements, with ezdxf standing out in aspects such as versatility, robustness, and documentation quality.

Keywords: applied programming , AutoCAD , automation , computer aided design , Python

Analysing the particle size distribution (PSD) of nanomaterials is essential for many scientific and industrial applications, as PSD offers important details about the distribution, average size, and range of particles within a sample. All these properties have a huge impact on the characteristics and behaviour of the materials. In this work we investigated image-based PSD recognition of nanoparticles from synthetic SEM images. For the task we considered an image classification pipeline based on five conventional (hand-crafted) texture descriptors and five pre-trained convolutional neural networks (CNN). Fusion strategies for combining the best performing feature sets were also investigated. We found that both the conventional and the CNN-based features achieved good performance (best accuracy respectively 87.6% and 88.2%), whereas the optimal performance (90.2%) was obtained by combining the two best performing CNN-based feature sets. Altogether the results are promising and demonstrate the feasibility of the method.

Keywords: Deep Learning , Image processing , Nanomaterials , Scanning electron microscopy , Texture

This work aims to highlight the state-of-the-art framework for implementing digital twins to support the product-process development. More in detail, the literature re-view is discussed on the base of a precise industrial context offered by a research collaboration active between the authors and ABB S.p.A., Santa Palomba (Pomezia), Italy. With this aim, applications may be distinguished by the level of complexity and related methodologies, tools and hindrances. Thanks to their prior strategic organization, data collection, processing and analytics are sufficiently mature to allow for the development and testing of various artificial intelligence and digital twin related approaches. This study identifies the key technical and methodological constraints for implementing digital twins in industrial environments, providing a foundation for future structured development. A detailed discussion of the major implementation issues is then presented, with a focus on four key areas within the specific industrial context: Process, Quality, Maintenance, and Flows. The analysis highlights both the bottlenecks and the positive aspects that can influence the effectiveness of the next future implementations.

Keywords: Data Analytics , Industry 4.0 , Integrated Product-Process Development , Machine Learning , Model Based System Engineering (MBSE)

Industry X.0 robotic manufacturing demands higher accuracy and flexibility, enabling continuously adaptive processes designed and optimized through simulations and Digital Twins. To achieve this level of flexibility and productivity in high value-added processes, where limited robot position accuracy becomes a critical constraint, advanced engineering methods and digital tools are required. These solutions must predictively compensate for inevitable robot positional accuracy errors, eliminating the need for manual pose refinement and enabling the generation of “first-time-right” robot code. This work aims to address these challenges by introducing an engineering tool capable of predictively correcting robot positioning inaccuracies across the workspace, enabling accurate point-to-point motion generation. It is intended for tasks with limited process interaction forces, where positioning errors are dominated by geometric, compliance, and joint-related effects. The tool leverages a multi-parameter Machine Learning (ML) error predictor trained on a reduced experimental dataset, minimizing data acquisition time and production downtime. Realized as a Python-based framework, it can be seamlessly integrated into commercial offline programming environments to automatically generate validated robot programs. The paper details the framework structure, focusing on the definition of the ML-based position error predictor, and its implementation on a robotic cell equipped with a high-payload KUKA robot and a FARO laser tracker. A preliminary experimental analysis identified payload, approach direction, and point location as the key operational parameters, accessible at the code level, that influence positioning accuracy. These insights guided feature selection and the design of reduced training datasets. In particular, a uniform spatial grid of only 64 points, corresponding to about one hour of measurement time, was sufficient to achieve near-optimal model accuracy. Several ML algorithms were compared, with the Tabular Prior-data Fitted Network achieving superior generalization on small datasets. Experimental validation on the KUKA robot showed up to a 98.4 % reduction in positioning error and consistent performance across all tested points, confirming the tool robustness and suitability for deployment across different industrial environments. All datasets, source code, and implementation scripts are openly released to enable reproducibility and facilitate industrial deployment.

Keywords: Engineering tool , Industrial robot , Machine learning , Offline compensation , Position accuracy , Robot simulation

This paper presents a comprehensive overview of methodologies developed to support blind and visually impaired people (BVIP) in perceiving two-dimensional visual content through tactile representations. Over the past decade, the TIP group at the Department of Industrial Engineering of Florence has advanced both deterministic and AI-assisted techniques to generate tactile bas-reliefs from pictorial and technical images. The research spans various computerassisted methodologies for producing depth images or full 3D models of tactile supports. The research seeks to establish a codified language, its expression dictated by the ‘grammar' of semi-automatic procedures that offer limited scope for a manual user's free interpretation. Latest results achieved in the translation of abstract paintings are presented, including an effective pixel intensity-based approach for translating chromatic information into tactile depth. Applications include cultural heritage accessibility and inclusive technical education. Results demonstrate the growing viability of hybrid and AI-enhanced solutions to generate meaningful, perceivable tactile models, highlighting the importance of structured tactile languages and multimodal interfaces for improving accessibility.

Keywords: 3D modelling , bas-relief , Blind People , Depth image , tactile models

In semiconductor environment, the manufacturing process and the related quality control hold a fundamental position. This aspect has been enhanced with the introduction of new materials, such as Silicon Carbide (SiC), and the necessity to perform a reliable method to detect wafer defects. In this paper, the Electrical Wafer Sorting (EWS), a reliable technique for creating electrical wafer maps, is combined with a deep learning pipeline to improve the defects identification. With respect to the state of the art, the innovative aspect is in the combination of a supervised system with an unsupervised sub-system. Compared to benchmarking algorithms present in literature, this approach allows to correctly classify well-know defect patterns and to characterize new defect morphology, leading to the identification of new potential issues in the production line and, overall, to a production yield improvement.

Keywords: Deep Learning , Electrical Wafer Sorting , Silicon Carbide , Wafer Defect Maps

Accurate three-dimensional (3D) reconstruction of individual dental elements from cone-beam computed tomography (CBCT) data is critical for adaptive prosthetic design and occlusal simulation in digital dentistry. However, challenges persist due to the complex morphology of teeth, the similar radiodensities of hard tissues, and the limited resolution of CBCT in capturing interproximal surfaces between adjacent teeth. This study presents a semi-automated, open-source compatible workflow that integrates CBCT image segmentation with the reconstruction of missing interproximal molar regions using Non-Uniform Rational B-Spline (NURBS) surface modeling. Molar datasets from CBCT acquisitions were segmented into 3D Slicer open-source software and refined to isolate individual teeth. Missing surface regions were reconstructed using third-degree NURBS patches in Rhinoceros® ver. 8 commercial software. Method reconstruction accuracy was quantitatively evaluated on 14 M by applying the method to manually created voids on intact tooth models. The average root-mean-square error (RMSE) and Hausdorff distance were 0.041 ± 0.015 mm and 0.049 ± 0.023 mm, respectively. The resulting models exhibited high surface continuity and anatomical fidelity, making them suitable for Computer Aided Engineering (CAE) analysis and prosthetic customization. This approach enables reproducible, high-precision dental modeling and represents a promising step toward fully automated 3D reconstruction workflows in clinical practice.

Keywords: CBCT segmentation , Digital prosthodontics , Molar morphology subdivision , NURBS modeling

Car seat comfort represents a fundamental dimension of automotive ergonomics, yet its assessment and optimization remain predominantly reliant on subjective feedback and iterative physical prototyping, limiting scalability and responsiveness in the design process. This study presents a novel data-driven framework that synergistically integrates statistical analysis and deep learning architectures enhanced with attention mechanisms to predict perceived driver discomfort from pressure distribution patterns. A comprehensive dataset acquired from 192 participants was used to extract biomechanical descriptors, such as contact area and peak pressure, across 9 anatomical regions, which were subsequently correlated with subjective discomfort ratings collected on a 10-point Likert scale. The proposed hybrid deep neural network achieved a modelling accuracy of R2 = 0.82 for overall discomfort, with peak pressure emerging as the most influential factor (Spearman's r = 0.78). Moreover, Finite Element Analysis (FEA) was employed to validate geometry refinements suggested by the AI model, resulting in an 18% reduction in peak seat pressure without compromising structural integrity. The findings underscore the transformative potential of integrating artificial intelligence with CAD–CAE methodologies to enable rapid, user-centered, and simulation-driven ergonomic design early in the vehicle development cycle, minimizing the need for physical prototypes and enhancing occupant comfort.

Keywords: AI-Driven Comfort Models , Deep Learning with Attention Mechanisms , Ergonomic Seat Design Optimization , Finite Element Validation of Human-Centric Design , Pressure Mapping and Biomechanical Analysis

Nowadays, to counteract the sedentary behavior in children, playground parks with their pieces of playground equipment can have a significant impact. In this context, the paper proposes a methodology based on wearable surface electromyography and inertial sensors data to derive functional data sheets for each piece of playground equipment and two novel synthetic indices for the assessment of physical fitness and motor coordination. These indices are built from: (i) muscle activation of nine forearm and leg muscles; (ii) acceleration data from two inertial sensors located on the forearm and lower back. The synthetic indices are based on the assessment of the muscular force rate and motor control in terms of smoothness. The validation of the proposed methodology is carried out through field tests, on seven pieces of playground equipment. The results confirm that: (i) the functional data sheet offers the possibility to obtain quick and intuitive visualization of the pieces of playground equipment in terms of required muscle activation; (ii) the proposed methodology represents a reliable and valuable tool to estimate physical fitness and motor coordination. This approach can be implemented in different playground parks even to guide the choice of the equipment to ensure specific levels of physical fitness and motor coordination.

Keywords: motor coordination , physical fitness , playground equipment , wearable sensors

Statistical Shape Model (SSM) for infant cranial analysis is challenging due to open sutures, which complicate the process of establishing anatomical correspondences between models. This study proposes a Wrap4D-based methodology for accurate preparation of newborn's cranial samples used in SSM. The approach integrates skin and skull segmentation with a guided wrapping technique that supports suture filling through anatomically-guided surface adaptation, while ensuring high fidelity in shape alignment. SSM algorithm uses the prepared dataset of 75 normal infant skulls from Meyer Children's Hospital IRCCS for the creation of a template, which serves as the reference model for defining a standard newborn's cranial shape. The methodology offers a rapid, reproducible and anatomically accurate solution for modelling suture closure. Preliminary results led to a Generalization of 1.245 mm, a Specificity of 2.659 mm and a Compactness of 98% with 20 modes of variation. Finally, quantitative comparison with traditional wrapping techniques reveals Wrap4D superior anatomical preservation in the region of suture.

Keywords: Newborn Skull , Statistical Shape Model , Suture closure , Wrapping

Point clouds have become a widely used data format in computer vision, driven by the increasing availability of 3D scanning devices in applications such as robotics and autonomous driving. However, challenges in data acquisition, such as occlusion, reflection, and limited sensor resolution, often result in incomplete scans. Consequently, point cloud completion, the task of reconstructing the complete geometry of a point cloud object from partial data, has emerged as a key area of research. Numerous deep learning-based methods have been proposed to tackle this problem. Although some surveys on shape completion exist, the rapid growth of research in recent years calls for an updated review. In this work, we present a comprehensive survey of recent advancements in deep neural networks for point cloud completion, covering literature from 2024 up to December 2025. After establishing a reference context by analyzing the major foundational methods up to the year 2023, we review the most recent methods with an analytical perspective based on how they have improved each specific component of the network. Furthermore, we present a comparative analysis of the performance of the reviewed methods on the main benchmark datasets, and finally, we discuss open challenges, offering insights and suggestions for future research directions.

Keywords: 3D Object Reconstruction , Computer Vision , Deep Learning , Point Cloud Completion , Review , Shape Completion

Capturing accurate texture maps from physical materials remains a challenge in digital prototyping and projection-based spatial augmented reality (P-SAR). This paper presents an open-source material scanning system based on photometric stereo, designed for affordability, simplicity, and efficient operation. The system combines a consumer-grade digital camera, multifaceted reflector (MR16) LED lighting, and Arduino-controlled automation to acquire material data up to A4 size within 15 s. Accurate colour reproduction is achieved through a hybrid calibration workflow that integrates camera profiling with a 3D lookup table. The resulting images are processed in a streamlined Substance 3D Designer pipeline to generate albedo and normal maps compatible with physically based rendering (PBR). To evaluate performance under realistic conditions, two fabric samples were scanned and qualitatively compared with professionally digitised references. Albedo maps were assessed based on dominant colour accuracy using CIEDE2000 (ΔE00), while normal maps were evaluated through visual rendering comparisons and directional distribution analysis. Scanning and processing times were also measured to verify workflow efficiency. Results demonstrate that the proposed system produces perceptually consistent textures suitable for real-time rendering applications while offering a low-cost and customisable solution for material digitisation.

Keywords: Custom scanner , Material digitisation , PBR textures , Photometric stereo

This study aims to develop an intraoperative navigation system for in-vivo procedures leveraging optical flow techniques. Traditional methods for organ tracking often fail in intraoperative scenarios due to domain complexity and, in the case of machine learning methods, by lack of annotated data. To address this challenge, we present a novel organ tracking system based on optical flow and evaluate the performance of four state-of-the-art neural networks for optical flow to determine the most suitable one to use in our system. The proposed system combines semantic segmentation and optical flow estimation. Segmentation networks identify the organ, surgical tools, and background in each frame. Then, deep learning-based optical flow networks estimate its motion across frames. The resulting motion is used to compute the organ's rotation and translation and update the 3D virtual model accordingly. The performances of the four neural networks for optical flow are tested on two video sequences of a robot-assisted partial nephrectomy, where the tracked organ of interest is the kidney. Among the tested networks, and for the concerns of this specific domain of use, RAFT and GMFlow achieved the most promising results in terms of IoU accuracy.

Keywords: 3D Organ Tracking , Deep Learning for Surgery , Partial Nephrectomy , Real-Time Surgical Assistance , Semantic Segmentation

Point cloud registration is a central theme in computer vision, with alignment algorithms continuously improving for greater robustness. Commonly used methods evaluate Euclidean distances between point clouds and minimize an objective function, such as Root Mean Square Error (RMSE). However, these approaches are most effective when the point clouds are well-prealigned and issues such as differences in density, noise, holes, and limited overlap can compromise the results. Traditional methods, such as Iterative Closest Point (ICP), require choosing one point cloud as fixed, since Euclidean distances lack commutativity. When only one point cloud has issues, adjustments can be made, but in real scenarios, both point clouds may be affected, often necessitating preprocessing. The authors introduce a novel differential entropy-based metric, designed to serve as the objective function within an optimization framework for fine rigid pairwise 3D point cloud registration, denoted as Iterative Differential Entropy Minimization (IDEM). This metric does not depend on the choice of a fixed point cloud and, during transformations, reveals a clear minimum corresponding to the best alignment. Multiple case studies are conducted, and the results are compared with those obtained using RMSE, Chamfer distance, and Hausdorff distance. The proposed metric proves effective even with density differences, noise, holes, and partial overlap, where RMSE does not always yield optimal alignment.

Keywords: density , entropy , holes , iterative closest point (ICP) , noise , partial overlap , Point cloud , registration , root mean square error (RMSE)

This study presents a fully automated pipeline for the morphometric analysis of the dens, based on an enhanced Statistical Shape Model of the C2 vertebra. Developed in MATLAB, the pipeline integrates semantic segmentation directly into the shape model, enabling precise and reproducible identification of anatomical regions without manual intervention. Unlike deep learning approaches, the proposed framework is well-suited to applications involving limited or heterogeneous data. The model was trained on 25 C2 vertebrae from the VerSe dataset and tested on three independent specimens. Evaluation of model compactness, generalization, and specificity confirmed the robustness of the Statistical Model. Morphometric parameters are extracted in reference to a local coordinate system derived from the mean C2 shape, ensuring anatomical consistency across specimens. The morphometric results for the dens fell within the anatomical ranges reported in the literature, supporting the method's validity. While not intended for clinical use due to dataset limitations, this work lays the foundation for scalable, reproducible, and anatomically informed morphometric analysis of the dens.

Keywords: Automatic Feature Recognition , Biomedical Image Analysis , Statistical Shape Modeling

This paper proposes an innovative framework based on artificial intelligence techniques to optimize the reuse of End-of-Life (EoL) ICT components. The system integrates two complementary modules: (1) zero-shot identification and classification of mechanical components from images using Segment Anything Model (SAM) and Contrastive Language-Image Pretraining (CLIP); (2) reuse profitability assessment through a Recovery Profitability Index (RPI) and Transformer models. Experimental results demonstrate an identification accuracy of 78% for the zero-shot module and up to 91% for profitability classification, highlighting the potential of this approach in the context of circular economy and industrial sustainability.

Keywords: Circular Economy , Transformer Models , Zero-Shot Learning , ICT Components Reuse

Accurate identification of spinal landmarks is critical in postural assessment, spinal deformity monitoring, and biomechanical modeling. Conventional techniques, such as radiographic imaging and manual palpation, suffer from significant limitations, including radiation exposure and operator dependence. This study presents a novel, fully automated method for localizing spinal apophyses using high-resolution 3D surface scans and shape index analysis. The proposed approach extracts the dorsal symmetry line and analyzes perturbations in the shape index, a differential geometric descriptor computed from local principal curvatures, to infer spinous process locations. Experimental validation was conducted on 21 healthy subjects using a structured-light 3D optical scanner, with palpation-based identification serving as a clinical reference. The method demonstrated high sensitivity, with a false negative rate of only 1.58% relative to operator-identified landmarks. The approach also identified additional plausible landmarks not detected by the operator, suggesting potential for improved diagnostic accuracy. These results support the method's suitability for objective, repeatable spinal landmark detection, with implications for both clinical assessment and computational modeling. Future work will focus on validation against radiographic ground-truth and applicability in pathological cases.

Keywords: 3D Surface Topography , Automated Landmark Detection , Shape Index Analysis , Spinous Process Detection