Optics, Photonics and Digital Technologies for Imaging Applications IX

We present a universal method for sculpting true three-dimensional (3D) light fields by engineering caustics with a digitally encoded axicon. This approach establishes a radial-to-axial coordinate mapping, allowing independent control over each axial plane within the light volume. By assigning distinct caustic-generating functions to different radial zones on a spatial light modulator, we synthesize volumetric light structures with tunable shape, scale, orientation, orbital angular momentum, and trajectory. Experiments demonstrate dynamic volumetric scenes, localized modulation, continuous transformations, and multi-object displays. This framework unifies and extends previous caustic and axially variant beam methods, enabling transversely proportional, freely reconfigurable 3D light constructs without mechanical motion. The technique opens new directions for volumetric displays, imaging and optical manipulation.

This work presents metasurface-based optical components for compact and efficient light sources in AR/VR head-mounted displays. MicroLEDs provide high brightness, high resolution, and a miniaturised form factor but emit unpolarised, wide-angle light, leading to significant optical losses and reduced overall efficiency in waveguide-based systems. To overcome these limitations, metallic wire gratings were developed for polarisation control, while TiO₂ nanopillar metalenses were optimised for beam collimation and steering. A multi-layer configuration integrating polariser, collimator, and steering layers achieved high transmission efficiency and broad angular performance. As a compact alternative, a single dual-functional metasurface capable of simultaneous collimation and polarisation filtering was also explored. The results demonstrate an integrated approach to light management in AR/VR systems, enabling thinner, lighter, brighter, and more energy-efficient display engines.

In this work, we present a compound transmissive metasurface architecture for single-waveguide augmented reality (AR) displays based on high–refractive-index waveguide materials. The proposed design incorporates compound metagratings that enable full-color operation with enhanced angular tolerance. Both forward and inverse design methodologies are employed to improve diffraction efficiency and overall optical performance. To overcome challenges related to waveguide thickness and back reflections, we introduce a compound metagrating strategy that achieves significant waveguide thickness reduction in systems requiring a wide field of view (FoV) and high resolution. The demonstrated approach provides a compact, high-efficiency platform for next-generation AR near-eye display technologies.

Diffraction gratings, used as couplers and expanders, are the vital components of Near-Eye Displays. The performance of these gratings is often compromised by inherent manufacturing tolerances. To overcome this challenge, we propose a robust grating design framework that accounts for fabrication tolerances to ensure consistent performance rather than merely maximizing peak efficiency.

Holographic displays offer a promising solution for realistic 3D displays by reconstructing the full wavefront of a scene. However, achieving simultaneous multi-depth viewing without complex mechanical components or high-speed hardware remains a challenge. In this paper, we propose and demonstrate a compact, scalable off-axis holographic display architecture that enables multi-plane reconstruction through a single viewing path. The system utilizes cascaded beam splitters to divide the reconstructed holographic wavefront into multiple optical paths with screens at different depths. The proposed approach is validated with experimental demonstrations of dual-depth reconstructions and three-plane reconstructions. While current results exhibit some superposition noise, the proposed architecture provides a flexible and modular foundation for the development of 3D holographic displays with multi-depth viewing.

Holographic near-eye displays (HNEDs) enable realistic 3D reconstruction with full physiological cues without vergence-accommodation conflict. To make HNEDs feasible, achieving large field of view (FoV) and full-color reconstructions is essential. The literature distinguishes two main architectures: pupil HNEDs, where FoV is limited by demagnified pixel size in the 4F setup, and non-pupil HNEDs, where FoV scales with hologram size and eyepiece magnification. While large FoV is achievable in HNEDs configurations, color-reconstruction remains challenging. Full-color HNEDs can be realized using temporal multiplexing, frequency division, or complex spatial filters, but these increase system complexity and limit FoV. A recent work [Chlipala M, et al, Scientific Reports, 2025, 15.1:15221] demonstrated a wide-angle non-pupil HNED, reconstructing full-color 3D scenes via angled RGB LED illumination, though requiring precise alignment. Design simplicity is crucial for wearable implementation; therefore, this paper proposes simpler single frame on-axis RGB illumination for pupil configuration, analyzing its FoV and reconstruction quality.

Holographic near-eye displays (HNEDs) allow for the reconstruction of 3D objects with natural depth cues. As a result, they can eliminate the vergence–accommodation conflict inherent in most conventional 3D display technologies, such as stereoscopic or multi-view systems. In HNEDs, spatial light modulators (SLMs) are used to encode computer-generated holograms that reproduce the complex amplitude of the target 3D scene. However, the pixel size and the pixel count of current SLMs are limited. These constraints directly restrict both the achievable field of view (FoV) and the spatial resolution of the reconstructed image. To address this problem, optical architectures have been proposed to effectively enlarge the FoV without changing the SLM’s parameters. Two main configurations are typically distinguished: pupil-forming and non-pupil systems. In both approaches, carefully designed optical relays can provide FoVs exceeding 80°, which is sufficient for many augmented and virtual reality scenarios. Such large angles, however, must be supported by the optics. Lenses capable of handling these large FoVs tend to be bulky and heavy. This is incompatible with the compact form factor expected from wearable HNEDs. A promising solution to solve this issue is to replace the classical optics with a diffractive optical element (DOE). DOEs can provide the required angular coverage and imaging performance while remaining thin and lightweight, thus improving user comfort and industrial design. In our work, we numerically design the required lenses phase function and fabricate the DOE in a thin polymer layer on glass and fused silica wafers using grayscale lithography. These DOEs can be used as masters for mass replication techniques, such as hot embossing or nanoimprint lithography. This approach enables high-precision, large-area diffractive eyepieces tailored specifically for holographic near-eye display applications. Developed DOE is experimentally tested in a wide-angle HNED built in pupil architecture. Obtained optical reconstructions of 3D point cloud with the DOE are compared with reconstructions with the regular optics.

The resolution of images in digital holography is fundamentally limited by the pixelization of spatial light modulators (SLMs). When applied directly to holographic lithography working in the Fraunhofer regime, the Nyquist theorem restricts the number of controllable samples in the focal plane to four times the SLM pixel count. We demonstrate that incorporating photoresist exposure and development into the propagation model introduces nonlinearities, effectively expanding spatial bandwidth beyond the Nyquist limit. We support this resolution gain using time-averaged batches of holograms to compensate for the noise introduced by the mismatch between the hologram and focal-plane degrees of freedom. Simulation results confirm that this approach can encode dense metasurface arrays with individually controlled meta-atoms, far exceeding conventional limits imposed by intensity-only models. This technique offers a pathway to rapid, maskless, and cost-effective prototyping of advanced photonic and electronic architectures.

We present a novel approach to computer-generated holography (CGH) using the Generalized Van Cittert-Zernike Schell propagator (GVS), incoherent light emitted from the screen of a mobile phone, and a spatial light modulator (SLM). In this approach, we use the stochastic gradient descent algorithm to jointly optimize an RGB image displayed on the mobile phone screen and a phase pattern displayed in the SLM. We use the GVS propagator to model the diffraction of partially spatially coherent light passing through the optical setup. Through numerical simulations and optical experiments, we show that our method is orders of magnitude faster and highly accurate compared to the results presented in the state of the art.

Optical Coherence Tomography (OCT) is a non-destructive imaging technique capable of performing subsurface imaging and reconstructing a full 3D volumetric image of the sample being measured by analysing interference patterns of backscattered light at multiple wavelengths. First developed for biomedical imaging, it has found many other uses, such as materials inspection in manufacturing. Traditional OCT systems use spherical lenses as the imaging optic, which restricts illumination to a limited depth, motivating interest in the use of axicons to extend the depth of focus. This work compares OCT performance using a conventional refractive axicon versus a planar metasurface axicon via simulation. The signal formation from a refractive axicon and a metasurface axicon differs, and the understanding these differences is essential in order optimise future extended-range OCT system designs.

This work presents the application of Complex Leader-Follower Full-Field Swept-Source Optical Coherence Tomography (CLF-FF-SS-OCT) for the examination of an Agate sample, a microcrystalline quartz with impurities and internal structures. A commercial tunable laser centered at 1580 nm with 120 nm range was employed as swept source due to the superior penetration depth in SiO2-based materials. The OCT system provided a high contrast 3D visualization of the sample’s inner structure, revealing several microlayers and achieving a penetration depth close to 1 mm. These results demonstrate the potential of CLF-OCT for non-destructive analysis in geological and archeological studies.

This work explores a new approach to holographic tomography (HT) that reconstructs three-dimensional refractive index (RI) distributions or phase using axial structured illumination instead of conventional multi-angle illumination. While iterative optimization methods have improved HT reconstruction quality, most systems still rely on complex multi-angle setups. Through simulations and experiments, this study demonstrates that cross-talk–free, slice-by-slice 3D RI reconstruction is achievable using on-axis structured illumination combined with sparsity constraints, provided the slice spacing follows an effective depth-of-focus design rule. The results establish the feasibility of a simplified Axial Structured Illumination Tomography (ASIT) system, requiring only minimal modification of standard digital holographic microscopes.

A polarization-multiplexed full-range OCT system is proposed for artifact-free imaging of multilayer optical samples. Two orthogonally polarized reference beams with a fixed phase delay are generated using a birefringent crystal, and an intensity-encoded reconstruction algorithm is employed to suppress autocorrelation and mirror artifacts without mechanical modulation. Experimental results on multilayer quartz and LCD samples demonstrate up to 45 dB artifact suppression and accurate full-depth imaging comparable to a commercial line-confocal sensor. The proposed method provides a compact and robust solution for non-destructive inspection of display components.

Quantum-enhanced imaging is an emerging area of research with relevance to a wide variety of application areas, including transport, gaming, environmental research, and security and defence. This subject encompasses a range of techniques and utilizes a number of developing quantum technologies. Time-resolved single-photon imaging approaches have been used to reconstruct high-resolution three-dimensional images, including challenging scenarios such as imaging through atmospheric obscurants and clutter. Critically, this approach has been extended to imaging in turbid underwater conditions. In the past, image reconstruction often proved to be time-consuming due to the inherent computational complexity, however advances in algorithms and hardware have allowed examples of “real-time” reconstruction of moving targets. Single-photon imaging has been used in demonstrations of moving target identification and human activity recognition with the aid of artificial intelligence approaches. Alternative single-photon imaging approaches, such as ghost imaging, will also be discussed.

I will present the PIXEurope Pilot Line, a recently started 400MEuro initiative under the Chips JU, that aims at developing and transferring advanced photonic integrated circuit (PIC) technologies and processes. Through Open Access, PIXEurope will support end-users in increasing the readiness level of their products. I will also present some technologies, leveraging PICs, that were developed at ICFO and transferred to spin-offs, now commercializing quantum random number generators, cryptography systems and phase imagers. Biography: Valerio Pruneri is an ICREA Professor and Corning Inc. chair, leading the Optoelectronics group at ICFO. He is also the Director of PIXEurope Pilot Line. Previously he worked for Avanex, Corning, Pirelli, and the University of Southampton. With his groups in academia and industry, he has developed technologies for the photonic, photonic integration, and quantum. He is inventor in more than 70 granted or pending patent families, leading to numerous industrial collaborations and the creation of four spin offs, Quside, Sixsenso, Luxquanta and Shinephi.

Fresnel diffraction plays a central role in simulating light propagation for holography and other coherent imaging systems. However, computing coherent light transport is computationally demanding because every object point can contribute to every hologram pixel, requiring specialized algorithms. Furthermore, the highly oscillatory nature of wavefronts forces dense sampling to maintain accuracy. We present an overview of phase-space decomposition techniques to accelerate numerical diffraction calculations. Moreover, we introduce a mipmap-inspired phase-space representation that significantly reduces scene sampling requirements for computer-generated holography. Our methods provide an analytical framework, implementation strategy, and performance analysis demonstrating substantial improvements over conventional ray to wavefront conversion techniques.

Digital holography records both amplitude and phase, allowing subtle scene changes to appear directly in the measured interference patterns. When objects move, these patterns evolve in characteristic ways that encode the underlying 3D motion. We introduce a method for estimating non-rigid motion from holographic data using the Chirped Cross Ambiguity Function (CCAF), which is well suited to the chirped structures produced by Fresnel diffraction. By applying a Short-Time Fourier Transform, we form a 4D spatio-angular representation of the hologram and derive a generalized optical-flow formulation. This framework enables precise estimation of instantaneous translational and rotational motion for dynamic holography applications.

Highly immersive 3D holographic displays require extremely large pixel counts, creating a data transmission bottleneck that necessitates the development of efficient compression frameworks. We show that digital holograms can be compressed to below 1 bit per thousand pixels by encoding only the magnitudes of numerically reconstructed, lower-resolution images at multiple depths. Due to their image-like nature, standard codecs can be used to store such reconstructions at very low bitrates. Our gradient-descent decoder then retrieves a hologram whose propagated intensities best match this focal stack. Experiments show accurate depth and parallax reconstruction while outperforming existing hologram compression methods.

In the field of digital holography, super-resolution techniques play a critical role in enhancing the quality of object reconstruction from low-resolution holograms, thereby addressing limitations imposed by optical systems and sensor resolutions. However, conventional image quality assessment metrics such as Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM) often prove inadequate, especially in sensitive application areas such as medical imaging, due to their limited alignment with human visual perception and their insensitivity to frequency-domain characteristics. To overcome these shortcomings, frequency-domain-based evaluation methods have gained increasing attention. In this study, two metrics grounded in Fourier analysis, the Harmonic Radius Index (HRI) and the Magnitude Knowledge Accuracy (MKA), are investigated and compared for their effectiveness in evaluating super-resolved holographic reconstructions.

We introduce a broadband lensless holographic microscopy platform that repurposes a standard visible-range board-level camera for low phootn budget coherent bioimaging from the deep UV (≈240 nm) to the near IR (≈1100 nm). Using a simple lensless geometry and wavelength-tailored illumination, the same low-cost sensor provides label-free, high-throughput imaging over a centimetre-scale field of view. In the deep UV, intrinsic absorption of nucleic acids, proteins and retinoids enables chemically specific imaging of extracellular vesicles, yeast and liver tissue (e.g., screening for retinoid-rich Ito cells). In the near IR, reduced scattering allows imaging through highly turbid samples, including clearing-free mouse brain slices (up to 250 µm thick), and mouse liver tissue slices unravelling anatomical cellular structures, otherwise invisible in the visible range illumination regime. This work highlights a compact, scalable route to broadband Gabor holographic bioimaging with a regular camera.

Cardiac diseases are characterized by deviations from normal spatiotemporal contraction wavefront patterns. Optogenetics offers ways to investigate and mitigate these states. System integration of optogenetic control strategies requires real-time control of cardiac contractions but is often hindered by the data processing load of dense sensor arrays. To circumvent this hurdle, we extract wavefront properties label-free from spatially sparsely sampled optical signals. This allows a fast determination of the prevalent wavefront state. Besides fast measurements (~kHz), real-time control of cardiac activity requires high temporal resolution in stimulation, achieved here using ferroelectric spatial light modulators. Experiments were conducted on in vitro human induced-pluripotent-stem-cell-derived cardiomyocytes expressing the opsin f-Chrimson. We demonstrate all-optical closed-loop control of cardiac contraction wavefront shape and direction. This capability establishes a foundation for non-invasive, feedback-controlled studies of cardiac wave dynamics and may provide new insights into the mechanisms underlying excitation wavefront disturbances and mitigation strategies.

Optical diffraction tomography (ODT) has emerged as a label-free 3D microscopy technique for measuring the refractive index (RI) distribution of biological samples. Non-interferometric ODT techniques have received increasing attention for their system simplicity and speckle-free imaging quality. Here, we introduce Fourier ptychographic diffraction tomography (FPDT) as a representative non-interferometric ODT modality capable of reconstructing 3D RI distributions using only multiple intensity measurements. By incorporating a spatiotemporal collaborative illumination strategy, our FPDT system achieves high-resolution, dynamic 3D live-cell imaging at a volumetric rate of 5 Hz, enabling monitoring and analysis of cellular dynamics. Furthermore, when combined with a proposed dynamic support-domain constraint, the method effectively suppresses missing-cone artifacts and enhances axial resolution. Finally, we overcome a key limitation associated with the matched illumination condition by introducing a defocused modulation strategy, which enables accurate 3D RI reconstruction under high–numerical-aperture microscope systems.

Microlens arrays (MLAs) enhance light collection in photon detectors by redirecting incident light to active pixel areas, improving the pixel fill factor and thus the external quantum efficiency. We present recent advancements in MLA optimization and integration for SPADs, SiPMs, and even PICs, addressing diverse substrates and broad optical ranges. A key innovation is the upscaling of MLA fabrication to 200 mm wafers and support for multi-project wafers (MPW), enabling cost-efficient prototyping and industrial compatibility. These capabilities are offered through CSEM’s MLA foundry services, providing design, mastering, and wafer-level replication for next-generation detectors in applications from life sciences to high-energy physics.

Biomedical visual models seek to facilitate the diagnosis, detection or segmentation of X-ray images. However, these models are trained with a distribution of images that carry biases given the context in which they are usually collected. During the stress-test process, it is possible to quantitatively measure the performance of these models in instances outside of the population with which they were trained. To this end, it is proposed to edit biomedical images, particularly chest x-rays. Editing is carried out by diffusion probabilistic models, specifically in discrete latent representation. The algorithm for inverting diffused images during inference is modified to give information from the original image and text as a condition. Thus, it is possible to access a diverse out-of-distribution images within reach of a medical description. Ready to test the scope of state-of-the-art visual models.

Fast and high-precision metrology is critical for production processes in the expanding high-tech industry sector. We explore the use of simple and cost-effective optical sensors in combination with a novel learning-based algorithm to simultaneously improve the accuracy of the parameter extraction, cost of the metrology sensor and time required for the measurement. We report sub-nanometer measurement accuracy and precision for our algorithm in the presence of strong optical aberrations and noise, which enables the use of affordable and simple optical measurement systems and faster measurement times. To contextualize our findings, we also performed Fisher information analysis.

We present a process workflow to generate realistic images of industrial products via standard raytracing tools. The goal of this technique is to provide digital images of surface defects of industrial components, under varying illumination conditions and from different viewpoints, and to build a library of computer-generated images for training AI models in automatic defect recognition. This approach differs from other rendering tools in the fact that the surfaces’ topology is measured at the microscale level, and its geometry is replicated in a CAD model; in addition, the light-surface interaction is described with accurate and rigorous BSDFs obtained from real scattering measurements.

The process water from petrochemical industries contains particles and oil droplets that must be removed before discharge into the environment. To detect and analyze these contaminants, a particle analysis system utilizing hyperspectral imaging, fluorescence imaging, and white-light imaging has been developed. This system targets particles and droplets ranging from 5µm to 150µm in size and utilizes an ultrasound particle manipulation system to guide particles and droplets into the focal plane of the imaging systems. Traditional analyzers rely on monochrome cameras with backlight illumination, producing monochrome images that provide only size and shape data. While this allows differentiation between solid particles (which appear black) and oil droplets, detailed material analysis of the solid particles remains limited. The hyperspectral imaging system offers spectral resolution, enabling material identification of the particles. The fluorescence imaging ensures reliable oil droplet detection, while the white-light imaging captures high-quality color images for both particles and droplets. This comprehensive data is essential for optimizing water treatment processes in oil production. The results of the field tests are presented, and the strengths and limitations of each subsystem are discussed providing valuable insights for end-users, researchers, and engineers seeking to select the most suitable sensor for their specific application.

We report a practical, noncontact system for measuring deep micro-holes on PCBs (example: 300 μm diameter, 6 mm depth) by combining spectral-domain coherent interferometry with a large-travel positioning machine. An energy-centroid signal-processing method yields micron-level precision and a depth relative error below 1%. Multithreaded scanning enables coverage of ≥50 cm × 50 cm panels, and an image-analysis pipeline referenced to machining standards attains a 97% bad-hole detection rate. The platform provides a scalable solution for accurate, factory-compatible inspection of challenging through-hole geometries.

We propose local regression mixture models as an adaptive and compact continuous radiance field representation. The whole extent of the scene is covered densely by multivariate 3D Gaussian components and associated linear regressors encode the means and local gradients of the radiometric responses. The partitions of unity from these marginal kernels act as gating functions that mix local regressors into a continuous volume that is explicitly defined everywhere. Both sharp or smooth regions may be captured by the relative overlap between density kernels. Our preliminary experiment on synthetic models shows that it is possible to represent manifolds with far fewer parameters than classic or splatting methods. We compare the result of raytraced numerical integration using two variants for the model representation: volumetric density (rho) or volumetric opacity (alpha).

ViTCapsNets is a groundbreaking hybrid deep learning architecture that combines Vision Transformers and Capsule Networks for automated building damage assessment from multi-temporal satellite imagery. The system addresses critical disaster response needs by processing heterogeneous sensor data with varying resolutions and conditions. The architecture leverages Vision Transformers' global contextual understanding and Capsule Networks' geometric transformation invariance, processing images at multiple scales for comprehensive building analysis. Evaluated on the extensive xBD dataset containing 850,000+ building annotations across global disasters, ViTCapsNets achieves 81.2% accuracy with exceptional performance in identifying intact (F1=0.90) and destroyed buildings (F1=0.72). The model's satellite-trained equivariance properties enable adaptation to drone imagery with minimal additional training, crucial for localized response operations. An open-source QGIS plugin is under development to facilitate rapid damage mapping in humanitarian contexts. This represents the first successful Vision Transformer-Capsule Network hybrid for remote sensing damage detection, pioneering sensor-independent geospatial intelligence capabilities for disaster response.

Polynomial Neural Networks (PoNNs) are a novel variant of the recently introduced Kolmogorov-Arnold Networks (KANs), designed to harness the approximation properties of orthogonal polynomials for modeling complex nonlinear functions. The KAN framework represents a new path in neural architecture design by introducing learnable 1-D functions instead of traditional weights, biases, and fixed activation functions. This work presents applications of two simple neural architectures built upon Hermite and Legendre polynomials, which serve as classification heads on top of a classical convolutional-net (ConvNet) acting as an automatic feature extractor. We directly compare the performance of these polynomial heads against a standard KAN head, with all models sharing the same convolutional backbone. Finally, we present recent advances in the development of a Convolutional Polynomial Neural Network (Convolutional-PoNN). This architecture replaces each parameter in a classical convolutional kernel with a learnable 1-D polynomial. All architectures are trained and evaluated on curated medical datasets from the MedMNIST collection.

Active piezo-driven optical systems produced by low-precision but low-cost mesoscale fabrication, suffer from strong actuator coupling, hysteresis, creep, and fabrication imperfections that render conventional influence-matrix calibration ineffective. Purely data-driven methods, while expressive, require excessive training data and lack real-time feasibility. We introduce a lightweight closed-loop calibration framework that uses an available nominal physics-based model as an informed prior. Starting from model-predicted commands, closed-loop position feedback iteratively refines the actuator commands. By leverging the nominal model, the correction landscape becomes low-rank and well-conditioned, achieving convergence in <15 iterations with residual pointing errors below 4% RMS of full stroke—an 87% improvement over uncorrected open-loop operation—without explicit modeling of hysteresis and other nonidealities. The rapidly collected calibration data can also train a compact feed-forward neural network for subsequent high-bandwidth open-loop use. This hybrid approach combines model-based stability with data-driven flexibility, significantly relaxes fabrication tolerances, and enables precision using inexpensive manufacturing equipment.

We explore Raman spectroscopy as a versatile, non-destructive tool for material characterization, offering insights into the structural, mechanical, and electronic properties of materials. This study highlights three exemplary applications in measuring stress in silicon, detecting point defects and lattice damage in silicon carbide, and sensing high pressure. In silicon, Raman spectroscopy reveals stress-induced shifts in the phonon peak and maps stress distributions and material compositions along cracks. For silicon carbide, it detects lattice damage and point defects like vacancies and dislocations. Its sensitivity also facilitates high-pressure sensing and phase transition analysis, underscoring its role in advancing semiconductor materials science.

A prototype system for illuminating clinical endoscope field of view simultaneously by three RGB laser lines at the wavelengths 450 nm, 520 nm and 638 nm has been updated, aiming at applications with all kinds of flexible fibro-endoscopes comprising a free working channel. The illuminating laser light is delivered via a specially designed optical fiber of appropriate diameter, with output aperture angle ~ 25 degrees. The system also includes a unit for reducing laser speckle artefacts in endoscopy images and a color sensor for controlling intensity ratios of RGB laser emission lines at the distal end of endoscope. Laboratory tests of the system have confirmed its suitability for clinical implementation.

Hot or smoldering materials in recycling streams pose significant risks of fire, equipment damage, and operational downtime in material recovery facilities. This study investigates the feasibility of using low-cost infrared (IR) cameras for real-time detection of hot objects in mixed recycling streams. A compact micro-thermal camera was integrated into a laboratory-scale sorting system and evaluated under controlled conditions. Temperature thresholds and image-processing algorithms were optimized to differentiate true hot objects from ambient variations and reflective artifacts. Experimental results show that the IR camera system achieved high detection accuracy. These findings demonstrate that IR-based detection offers a robust, non-contact method for early identification of hazardous materials, enabling effective fire prevention in recycling operations.

Hyperspectral line-scan systems are widely employed in industrial sorting applications to distinguish materials based on subtle spectral characteristics, or spectral fingerprints, that are imperceptible to the human eye. Common applications include contaminant detection in food processing and the separation of different plastic types. These systems are well suited for high-throughput inspection due to their ability to capture spatial and spectral information line by line. Consequently, only a line-shaped region of the sample requires illumination. Conventional illumination configurations employ multiple halogen spotlights that produce overlapping elliptical illumination areas. While this approach ensures coverage of the region of interest, it results in non-uniform intensity distribution along the scan line and illuminates a larger area than necessary. In this work, we propose an alternative illumination concept utilizing readily available automotive H7 halogen headlight bulbs in combination with an elliptical reflector and a lens array. The performance of the proposed system is compared to that of conventional halogen spot illumination in terms of intensity uniformity and spectral performance.

We present a sequential ray tracing tool, HoloTraceX, to evaluate, prior to experimental recording, the image quality of see-through waveguides formed by a glass substrate and two coupling/decoupling holograms. The model incorporates exact geometry, refractions and reflections with Fresnel coefficients, and volume holography using Kogelnik coupled wave theory to estimate ray efficiency. Wavefront maps and image metrics are calculated along the guided propagation, including chromatic dispersion and angular tolerances. Wavelength and angle scanning allows regions of high efficiency and low aberration to be identified, optimising the design of the couplers and the exit pupil. The results demonstrate that this preliminary analysis would optimise the holographic recording geared towards the optical architecture of see-through AR systems.

We present the design of a new family of diffractive lenses based on the aperiodic Paperfolding sequence, implemented through both a binary-phase and a Kinoform phase profile. This family of optical elements exhibits a distinctive behavior: while the binary-phase lens presents an aperiodic phase distribution consistent with the generating sequence, the Kinoform version displays a periodic phase profile, resulting from the partially repetitive nature of the sequence. Both lenses can therefore be regarded as bifocal elements, with their focal relationships depending on the specific phase implementation.

Infrared (IR) breast thermography is a non-invasive and complementary technique with great potential for the early diagnosis of pathologies such as breast cancer. For the correct characterization of tissue abnormalities and the accurate classification of thermograms, it is essential to extract robust statistical descriptors from Regions of Interest (ROI). This study evaluates on the performance of three key statistical moments of the ROI temperature distribution for automated classification: Maximum Temperature (T_máx), Standard Deviation (σ_ROI), and Skewness (γ_ROI). The integration of these descriptors into a Support Vector Machine (SVM) classifier demonstrates discriminatory superior to any individual descriptor, establishing a solid foundation for the development of Computer-Aided Diagnosis (CAD) systems in breast thermography. The proposed methodology was validated using the Database for Research in Mastology (DMR) for the automated classification of breast thermograms.

We propose a modification of the kinoform lens design that is capable of reproducing the effect of spherical aberration on the axial irradiance distribution under monochromatic illumination. This new design presents a phase distribution with a sawtooth profile; however, unlike conventional kinoform lenses, the phase function is periodic in αr^4+(1-α)r^2, where r is the radial coordinate in the lens plane. We demonstrate that, by adjusting the value of the parameter α, it is possible to replicate the effect of spherical aberration with a specific Zernike coefficient. This novel design could be useful not only as an alternative for compensating spherical aberrations in diffractive lenses but also for designing optical elements with an extended depth of focus.

Due to the development of optics and digital image processing, implementation of plenoptic imaging applications are gaining increasing interest. Therefore, there is a need for tools through which application and algorithm developers could test their ideas. In this paper, we will represent a simple web based tool to interactively simulate the processing of test raw plenoptic images into rendered ones and to generate image processing functions for further testing of the rendering possibilities and qualities.

In manufacturing and quality inspection, demand for three-dimensional scanning has grown rapidly. This work presents an eye-in-hand RGB-D 3D scanning system mounted on a UR10 robot for volumetric measurement of large objects beyond turntable-based scanners. An Intel RealSense D435i camera is attached to the robot end-effector, and robot kinematics are combined with depth data to reconstruct watertight 3D models from multi-view point clouds. The pipeline includes path planning for cylindrical and orbit-type trajectories, point cloud preprocessing and registration, mesh reconstruction, and automated volume estimation. Repeated measurement experiments on regular and irregular objects such as cylinder, cube, asymmetric conical frustum, as well as objects larger than the camera field of view, achieve accuracy above 95% with respect to CAD reference volumes, comparable to our previous 2-DOF rotational RGB-D scanner while extending the measurable size and shape range. The system shows promise as a metrology solution for bulky parts. Acknowlegments : This work was supported by the Korea Institute for Advancement of Technology (KIAT) grant funded by the Korea Government (MOTIE) (RS-2024-00409639, HRD Program for Industrial Innovation; RS-2025-24536353).

This paper presents a Multi-Stage AI-Driven Total Solution designed to resolve inspection bottlenecks in semiconductor High-Volume Manufacturing. By integrating In-line ADC+, Post Micro ADC, and Whole Wafer (Macro) ADC, the system strategically augments rule-based AOI to filter optical artifacts in real-time and identify process-level patterns. Deployment results demonstrate a 95% reduction in nuisance defects caused by illumination artifacts (e.g., fluorescence noise) and 100% accuracy in wafer-level pattern classification. This unified ecosystem ensures data continuity from acquisition to review, significantly enhancing throughput and yield learning without compromising the sensitivity of optical inspections.

Digital Holographic Microscopy (DHM) is an interferometric technique widely used for surface profiling. It is valued for its high phase measurement accuracy. However, DHM relies on plane-wave illumination, which limits resolution. Structured illumination (SI) can overcome this limitation, improving resolution. DHM reconstructs amplitude and phase images digitally; however, the phase is confined to a 2π. Recent research has explored multi-beam illumination and proposes Digital Holographic Profilometry (DHP) as a potential solution. In this work, the SI-DHP system, which is a combination of DHP and SI techniques, will be proposed. The developed system, equipped with a DMD modulator, generates illumination with sinusoidal fringes for SI. A special algorithm, combining the longitudinal scanning function of DHP with the SI algorithm, was developed for shape reconstruction. The article presents both simulation and experimental results that demonstrate an increase in transverse resolution, while maintaining high longitudinal resolution and an enlarged measurement range.

Fusion of medical images plays a fundamental role for disease diagnosis and interpretation because it allows the combination of structural and functional modalities to provide more relevant and comprehensive data. The fusion is the process of integrating complementary information from multiple source images into a single composite representation, thereby enhancing interpretability for subsequent analysis. In this work, we propose a novel fusion technique based on the fractional Hermite transform (FrHT) and sparse representation (SR). The method is applied to fusion of MRI and PET/SPECT images of the brain. Specifically, the fusion of Magnetic Resonance Imaging (MRI), which offers high-resolution anatomical detail, with Positron Emission Tomography (PET) or Single Photon Emission Computed Tomography (SPECT), which reveal metabolic and functional activity, enables clinicians and researchers to better localize abnormalities and improve diagnostic accuracy. The FrHT extends the classical Hermite transform providing adjustable fractional orders in the analysis and enhancing feature extraction across multiple domains. Sparce representation is a technique which employs a linear combination of a small number of atoms from an overcomplete dictionary to decompose a signal into a sparse vector. In the proposed framework, coefficients obtained from the FrHT decomposition at different orders are further analyzed using the SR model. The fusion process is performed using the resulting sparse vectors, ensuring that both structural and functional features are preserved during the fusion process. The technique was evaluated on public brain MRI and PET/SPECT images, demonstrating its effectiveness in producing fused images that retain anatomical precision while highlighting functional activity. Typical evaluation metrics for image fusion assessment were used to validate the proposed method. The results suggest that the proposed FrHT-SR approach can serve as a valuable tool in medical imaging, supporting improved visualization and clinical decision-making.

Airborne hyperspectral cameras is crucial for remote sensing applications like crop monitoring, soil assessment, and forest health evaluation. However, multi-sensor integration creates hetero-spectral registration challenges, and large-scale acquisition introduces flight strip stitching errors. This study proposes a block-affine registration and cloud-controlled stitching method. For hetero-spectral registration, block-based affine transformation with grid partitioning and local feature matching compensates for local variations and geometric distortions. For stitching, a cloud-control framework leverages existing geospatial data for feature matching, extracting control points to refine POS parameters for high-accuracy alignment. Validated in Yushu, Jilin Province, the method achieves localization accuracy within three ground sample distances (GSD), meeting hyperspectral remote sensing requirements and supporting large-scale environmental monitoring.

We present a programmable spectral-bandwidth control method for an acousto-optic tunable filter (AOTF) that combines multi-frequency RF driving with detector delayed integration. A multi-tone composite RF drive simultaneously produces several spectrally separated diffracted passbands. By time-multiplexing a small set of composite waveforms within a single exposure and tuning the integration delay, diffracted signals generated in successive intervals are accumulated into one image frame, synthesizing an effective passband with a user-defined bandwidth. We further provide parameter-selection guidelines for RF frequency, amplitude, phase, and segment dwell time to balance bandwidth coverage and spectral fidelity. The approach suppresses intermodulation-related distortions while reducing waveform switching complexity, enabling efficient bandwidth reconfiguration for multispectral imaging and rapid spectral acquisition.

Imaging and tracking objects hidden behind scattering media is challenging as multiple scattering randomizes the optical wavefront and produces seemingly random speckle patterns. While the optical memory effect enables the recovery of object information from speckle autocorrelations, many existing approaches rely on phase-retrieval algorithms, which are computationally expensive and unsuitable for dynamic scenes. In this work, we propose a numerical framework for motion detection through scattering media based on speckle intensity correlation analysis. Instead of reconstructing the object at each frame, motion is extracted from temporal differences in speckle intensity correlations, specifically I(tn) ⋆ I(t0) − I(t0) ⋆ I(t0), which suppresses the static background and isolates dynamic components. Motion detection is performed using correlation-based metrics and the structural similarity index (SSIM), combined with an adaptive exponentially weighted moving average (EWMA) threshold- ing scheme. Simulations incorporating memory-effect-induced blurring and additive Gaussian noise demonstrate robust detection of motion and object disappearance events without requiring phase retrieval. The proposed approach provides a computationally efficient alternative for motion detection in scattering environments.

Volumetric printing by reverse tomography—commonly referred to as tomographic volumetric additive manufacturing (TVAM), or simply VAM—is a layerless 3D printing approach that creates solid objects by projecting sequences of light patterns into a rotating photosensitive resin. The cumulative energy deposition leads to localized photopolymerization, enabling the fabrication of complex 3D structures within seconds. Since its first demonstration in 2019, TVAM has rapidly gained attention and has been extended to a wide range of materials, including polymers, hydrogels, ceramics, metals, and glass. Despite these advances, several fundamental challenges remain: expanding the achievable build volume, improving resolution toward that of two-photon polymerization, and extending the method to non-transparent or composite resins. Beyond these technical hurdles, TVAM also opens unique opportunities not accessible to traditional additive manufacturing techniques—for example, the ability to fabricate directly around existing structures. This talk will review the progress of the field, highlight key developments, and explore emerging strategies to address current limitations while pointing toward new application domains.

Hollow core fibres (HCFs) guide light through a gas-filled core rather than solid glass and offer unique optical properties and transformative potential. Recent breakthroughs have reduced losses to levels below those of conventional silica fibres across a broad spectrum—from the ultraviolet to the mid-infrared. While these advances were primarily driven by applications in high-capacity data transmission, the capabilities of HCFs extend far beyond communications. They enable enhanced light–matter interactions, exceptional radiation hardness and low non-linearity for sensing applications such as gas detection and precision gyroscopes. There is even the prospect of light transmission in vacuum, pushing towards ultimate waveguide performance. As fibre designs improve, controlling the internal gas composition and pressure becomes critical, driving progress in gas dynamics modelling, advanced characterisation, and post-processing techniques. This talk will explore recent progress in HCF technology, the pivotal role of gas in transmission, and the opportunities and challenges of guiding light through a hollow core.

We present an unsupervised video denoising method designed to remove structured noise, such as stripes or other random patterns common in microscopy and remote sensing, without needing clean training data. Unlike existing spatial-domain approaches, our method operates in the Fourier domain, leveraging the observation that Fourier coefficients of structured noise exhibits independence. By randomly masking and replacing Fourier coefficients across frames, the model learns to recover clean video content. Results show effective noise suppression without prior noise modeling, offering a practical solution for real-world imaging systems where clean references are unavailable.

Gridding algorithms are essential for applying the Fast Fourier Transform (FFT) to nonuniformly sampled data, as commonly encountered in MRI and CT imaging. These algorithms interpolate irregular samples onto a uniform grid before FFT computation, and their accuracy depends critically on the choice of the gridding kernel. This work introduces a new optimization framework based on vector optimization, where kernel optimality is defined through Pareto efficiency of the error shape operator Λ, a mapping that quantifies how interpolation errors vary across frequencies. Using scalarization techniques, we design kernels that approximate user-defined target error shapes, enabling customized accuracy in regions of interest. Numerical experiments show significant improvements over classical kernels such as prolate spheroidal wave functions and state-of-the-art NUFFT implementations, paving the way for adaptive and application-specific gridding strategies.

Neuromorphic computing leverages the intrinsic responses of functional materials to mimic key information-processing behaviours in biological neural networks. In this work, we present photo-responsive materials capable of emulating synaptic plasticity, which is the basis for learning and memory. Photonic layers of carbon-based nanomaterials were engineered to exhibit dual fluorescence–phosphorescence emission and photo-activation dynamics, mimicking short-term memory, long-term memory, and synaptic potentiation, respectively. Using standard off-the-shelf optoelectronic components for the optical signal excitation and readout, including a smartphone for the optical readout, we demonstrate the materials’ ability to memorize, forget, and compute signals. Their responses were applied to the MNIST digit-classification task, achieving test accuracies of above 95% (training) and 94% (testing) after 51 training epochs. The integration of mobile-based optical readout provides a distinctive advantage in terms of accessibility and portability, offering a completely innovative route toward practical, low-power, and widely deployable implementations of optical neuromorphic computing.

This presentation introduces a lightweight, content-adaptive deep learning model for reliable face liveness detection. By generating spatial filters conditioned on image content, the method efficiently captures textural and reflectance cues distinguishing genuine from spoofed faces. Experiments across multiple public datasets show near-perfect detection accuracy with low computational overhead, making the framework ideal for mobile and embedded authentication devices. The approach contributes to advancing secure and explainable vision-based identity verification in multimedia environments.

Ischemic stroke, resulting from a blockage in cerebral blood flow, is a major global health concern due to its high rates of mortality and long-term disability. A fast and accurate diagnosis is essential for effective treatment and improved prognosis that allows to preserve the life of patients. Magnetic Resonance Imaging (MRI) is considered the standard image modality for stroke evaluation due to its sensitivity in detecting the ischemic lesions and the ability to provide detailed visualization of brain tissue abnormalities. In this work, we propose an automated system for ischemic stroke detection in MRI scans using the fractional Hermite transform (FrHT), a technique that captures localized spatial-frequency features. The methodology involves a preprocessing stage of the MRI slices, the FrHT-based features extraction, and the classification of stroke-affected regions using a supervised learning model. The system was evaluated on the publicly available ISLES dataset, which includes annotated MRI scans of stroke patients. Experimental results demonstrate the effectiveness of the proposed approach.

Understanding the intricate 3D motion of human sperm flagella is vital for uncovering mechanisms of motility, navigation, and fertilization. This study presents a computational framework for extracting spatio-temporal patterns from 3D+t imaging of sperm flagellar beating. Using feature-based descriptors, including parsed motion primitives and polar histograms, we characterize distinct motility patterns under varying physiological conditions. The proposed approach provides a quantitative means of analyzing flagellar dynamics, offering potential applications in fertility assessment, sperm diagnostics, and bio-inspired locomotion studies.

The visual inspection of non-diffusive objects such as transparent or specular objects remains a challenging task due to complex object geometries and light–matter interactions. In this work, we demonstrate an inverse light-field illumination approach for the single-shot, robust inspection of transparent objects, in which a tailored light field is derived from a single sample, compensated for variations in size, shape, position, and optical properties, and then projected onto the test object. We evaluate quantitatively how this enables reliable defect detection with maximized contrast using only a single acquired image.

This study utilized shortwave infrared (SWIR) observations from TROPOMI/Sentinel-5P (1.6–2.3 µm) to investigate atmospheric methane (XCH₄) variability across Bangladesh from 2018 to 2025. Persistent methane hotspots were detected over urban–industrial and rice-growing regions, with national XCH₄ increasing by about 10 ppb yr⁻¹ and localized trends reaching 45 ppb yr⁻¹. Strong seasonality was observed, with minima in spring and maxima in autumn, associated with industrial activity, wetland emissions, and post-monsoon hydrology. Pearson and Spearman analyses revealed a predominantly negative correlation of XCH₄ with temperature (−0.21/+0.06), along with positive correlations with pressure (+0.45/+0.20) and wind speed (+0.35/+0.34), and negative correlations with solar radiation (−0.34/−0.30) and precipitation (−0.27/−0.23). Machine learning (ML) and geographically weighted regression (GWR) models consistently identified temperature, wind speed, and pressure as dominant predictors. These results demonstrate the capability of SWIR spectroscopy for assessing greenhouse gas variability and climatic influences in tropical regions.

Hyperspectral imaging (HSI) is a powerful non-invasive technique that combines imaging and spectroscopy to capture detailed spectral information across multiple wavelengths. However, HSI systems are typically com- plex and expensive, factors that constrain their widespread adoption. This project presents the development of a multispectral imaging (MSI) prototype employing time-multiplexed LED illumination across 31 wavelengths spanning 400 - 700 nm. The system leverages deep learning algorithms, specifically convolutional neural networks (CNNs), to reconstruct hyperspectral data from multispectral images. This approach offers significant advan- tages, including low cost, high mobility, high frame rates, and efficient spectral reconstruction. The end-to-end deep learning model demonstrates the ability to learn complex patterns and achieve high accuracy in spectral re- construction, making it suitable for medical imaging, remote sensing, and other fields requiring detailed spectral analysis.

Elastography is a novel medical imaging method that maps soft tissue mechanical properties, such as stiffness, viscosity, porosity and anisotropy, based on image reconstruction of data acquired by traditional medical imaging technologies such as MRI, Ultrasound and OCT. The development and validation of elastography reconstruction methods for the complex mechanical conditions found in soft tissue, such as the anisotropic, fluid-saturated, vascularized environment of the brain, for example, creates an unmet need for elastography data acquisition methods that do not rely on expensive medical imaging technologies that often impose severe physical limitations on the imaging target. Here we present a novel elastography data acquisition approach based on optical scanning tomography (OST), where the physical limitations on the imaging target are much simpler than those in MRI, for example. In addition to describing this innovative elastography method, we will present cross validation of the method via MRI methods and external mechanical testing.