Academia.eduAcademia.edu
All Topics
Engineering
Signal Processing
Adaptive Imaging

Adaptive Imaging

description11 papers
group0 followers
lightbulbAbout this topic
Adaptive imaging is a technique in imaging science that involves the dynamic adjustment of imaging parameters and algorithms in response to varying conditions or target characteristics, enhancing image quality and information retrieval. It is commonly applied in fields such as medical imaging, remote sensing, and computer vision.
lightbulbAbout this topic
Adaptive imaging is a technique in imaging science that involves the dynamic adjustment of imaging parameters and algorithms in response to varying conditions or target characteristics, enhancing image quality and information retrieval. It is commonly applied in fields such as medical imaging, remote sensing, and computer vision.

Key research themes

1. How can adaptive optics improve visual system assessment and neurophysiological imaging beyond conventional retinal imaging?

This research theme explores the use of adaptive optics (AO) in vision science to bypass ocular optical aberrations, enabling aberration-free retinal imaging and precise psychophysical measurements. It focuses on how AO systems enhance understanding of the visual system's organization, improve psychophysical assessment under manipulated optics, and facilitate neurophysiological investigations that were previously unfeasible with conventional optics.

Vision science and adaptive optics, the state of the field
by Melanie Campbell

2023, Vision research

Key finding: The paper highlights that AO allows projection of aberration-free images onto the retina enabling measurements of neural spatial limits beyond optical degradation. AO systems enable the study of visual function with native... Read more
articleView PaperdownloadDownload
Space-variant Shack–Hartmann wavefront sensing based on affine transformation estimation
by Louis Tao

2023, Applied Optics

Key finding: This study addresses space-variant aberrations in deep tissue imaging by extending Shack-Hartmann wavefront sensing to a four-dimensional representation factoring spatial and Fourier domain variability via affine... Read more
articleView PaperdownloadDownload
Dual Conjugate Adaptive Optics Prototype for Wide Field High Resolution Retinal Imaging
by Jörgen Thaung

2023

Key finding: Demonstrates that multi-conjugate adaptive optics (MCAO), with multiple deformable mirrors conjugated to different ocular layers and multiple guide stars, extends the corrected field-of-view significantly beyond... Read more
articleView PaperdownloadDownload
keyboard_arrow_downShow more

2. What advanced methodologies enable effective aberration correction and high-resolution imaging in ultrasound and medical tomographic systems?

This theme investigates matrix-based and adaptive approaches to characterize and correct wavefront distortions caused by spatial heterogeneities in ultrasound and tomography. It emphasizes the use of reflection matrices, distortion matrices, and iterative algorithms to recover isoplanatic patches and optimized focusing laws, significantly improving image quality and resolution in clinically relevant settings. The methodologies explore nonconvex optimization, matrix decompositions, and local adaptive filtering as regularization in inverse imaging.

Ultrasound Matrix Imaging. II. The distortion matrix for aberration correction over multiple isoplanatic patches
by Alexandre Aubry

2024, IEEE Transactions on Medical Imaging

Key finding: Introduces a distortion matrix derived from the focused reflection (FR) matrix that decouples input and output wavefront distortions, enabling iterative aberration correction across multiple locally invariant isoplanatic... Read more
articleView PaperdownloadDownload
Accelerated Image Reconstruction for Nonlinear Diffractive Imaging
by Dehong Liu

2022

Key finding: Proposes CISOR, a nonlinear reconstruction method incorporating a novel FISTA variant with total variation regularization, providing convergence guarantees for nonconvex inverse scattering problems. CISOR effectively models... Read more
articleView PaperdownloadDownload
Measurement Reduction Methods for Processing Tomographic Images
by Alexey Chulichkov

2023, Sensors

Key finding: Applies computer-aided measuring system theory to improve tomographic image reconstruction. The method leverages structural properties of the Radon transform without relying on uncertain object smoothness priors, instead... Read more
articleView PaperdownloadDownload
Spatially Adaptive Filtering as Regularization in Inverse Imaging: Compressive Sensing, Super-Resolution, and Upsampling
by Vladimir Katkovnik

2024, Super-Resolution Imaging

Key finding: Introduces iterative spatially adaptive filtering as an explicit regularization technique in inverse problems such as compressive sensing and super-resolution. The approach iteratively refines solutions by applying adaptive... Read more
articleView PaperdownloadDownload
keyboard_arrow_downShow more

3. How can adaptive sampling and sparse acquisition techniques enable high-resolution imaging beyond traditional hardware limitations?

Focusing on computational imaging strategies, this theme addresses overcoming hardware constraints by employing advanced sampling patterns, sparse acquisitions, and adaptive interpolation methods. It investigates single-pixel imaging at resolutions approaching hardware limits via compressive and adaptive sampling, efficient acquisition of anisotropic appearance functions (BRDF) through sparse angular sampling and robust interpolation, and exploiting high dynamic range acquisition frameworks via modulo sampling to extend conventional sensor limits.

Single pixel imaging at high pixel resolutions
by Anna Pastuszczak

2022, Optics Express

Key finding: Demonstrates a framework for single-pixel imaging that harnesses full DMD resolution (1024 × 768) despite limitations in modulation frequency, by employing differential binary sampling patterns and a two-stage reconstruction... Read more
articleView PaperdownloadDownload
BRDF Slices: Accurate Adaptive Anisotropic Appearance Acquisition
by Pavel Zid

2021, Proceedings / CVPR, IEEE Computer Society Conference on Computer Vision and Pattern Recognition. IEEE Computer Society Conference on Computer Vision and Pattern Recognition

Key finding: Presents a novel adaptive angular sampling method for anisotropic BRDFs using uniformly distributed perpendicular slices through BRDF subspaces at fixed elevations. By acquiring sparse angular samples guided by this adaptive... Read more
articleView PaperdownloadDownload
HDR-TOF: HDR Time-of-Flight Imaging via Modulo Acquisition
by Gal Shtendel

2025, 2022 IEEE International Conference on Image Processing (ICIP)

Key finding: Proposes a novel HDR Time-of-Flight imaging architecture combining conventional ToF sensors with modulo ADC acquisition, enabling single-shot high dynamic range imaging without saturation. Establishes sampling density... Read more
articleView PaperdownloadDownload
keyboard_arrow_downShow more

All papers in Adaptive Imaging

Time-resolved multispectral imaging based on an adaptive single-pixel camera
by Gianluca Valentini

2026, Optics Express

Time-resolved multispectral imaging has many applications in different fields, which range from characterization of biological tissues to environmental monitoring. In particular, optical techniques, such as lidar and fluorescence lifetime... more
descriptionView Paperarrow_downwardDownload
Bio-inspired variable imaging system simplified to the essentials: modelling accommodation and gaze movement
by Jan Korvink

2026, Optics express

A combination of an aspherical hybrid diffractive-refractive lens with a flexible fluidic membrane lens allows the implementation of a light sensitive and wide-aperture optical system with variable focus. This approach is comparable to... more
descriptionView Paperarrow_downwardDownload
An adaptive framework for image and video sensing
by Morteza Shahram

2025, SPIE Proceedings

Current digital imaging devices often enable the user to capture still frames at a high spatial resolution, or a short video clip at a lower spatial resolution. With bandwidth limitations inherent to any sensor, there is clearly a... more
descriptionView Paperarrow_downwardDownload
Light sheet adaptive optics microscope for 3D live imaging
by Gordon Love

2023, Three-Dimensional and Multidimensional Microscopy: Image Acquisition and Processing Xx

Optical microscopy is still the main research tool for many biological studies. Indeed with the advent of genetic manipulation and specifically, the use of fluorescent protein expressing in animals and plants it has actually seen a... more
Figure 2.1: Diffraction at an aperture in a plane screen
Figure 4.7: Sensitivity to blur, in presence of noise, assessed in Figures 4.6(a: to 4.6(e).
Figure 3.5: Magnetic continuous membrane based on this technology is used.
metric behaviour with respect to image content. Figure 4.4: metric dispersion with image content, as per eq. (4.1), as a function of PSF RMS radius. For comparison the Airy disc diameter is 2.44 pixels, and the RMS radius is 0.64 pixels.
Figure 2.6: Comparison of the aberration free and aberrated MTF.
Table 4.1: Metrics used in the simulation and experimental part of this chapter.
(c) Sharpness metric variation with (d) Derivative of the plots given in the level of defocus, on Image 4.1(c) Figure 4.2(c) (a) Sharpness metric variation with (b) Derivative of the plots given in the level of defocus, on Image 4.1(a) Figure 4.2(a)
Figure 2.2: Wavefront corrector in a 2f configuration. a normal incidence monochromatic plane wave after reflection on the DM is
so Eq. 2.17 can be approximated by the following expression:
Figure 3.2: effect on Phase of refractive index mismatch at a planar interface for ny = 1.5 et ng = 1.3 at 200 microns depth for NA =0.2, 0.8 and 1.3. Colour bars represents the phase aberration in microns.
Figure 4.5: Blur and Noise effect on the image histogram. The blurring of the image has been done with a defocused PSF of 28 pixels RMS, and the noise added to the image is uniform within the grey level range [0,0.7]. The initial image was normalised to 1. All histogram plots have a vertical axis ranging from 0 to 38200 (images are 511x511 pixels) and the horizontal axis representing the grey level, is between 0 and 1. A number of bin of 100 is used.
Figure 3.9: Shack-Hartmann wavefront sensor, sensing a plane wave (top) and an aberrated wave (bottom) ture, ® is the phase.
Figure 2.10: Example of the effect of multiple decomposition on a 1D spectrum. The first decomposition (top row) has divided the spectrum in two and each spatial frequency content is now localised on each sub image I; and I7.The sec- ond decomposition is applied to the approximated image so the resulting range of low frequencies is subsequently divided in two, leading to three distinctive ranges of spatial frequencies and their associated images. Figure 2.10 illustrates this principle, using a 1D example.
Figure 6.3: Aberration created at the centre of the cylindrical tube expressed as the first 15 coefficients (piston, tip and tilt excluded), for parameters ¢ = 0.25mm, R = 0.6mm, n, = 1.3334, ng = 1.47, d = 0.5mm and NA = 0.61. Defocus (2,0), and astigmatism (2,2) are the main contributors.
the mirror membrane and creates a localised bending.
the optimised images depending on the metric used for optimisation. Figure 4.14: Relative improvement of the metric value for the 5 different metrics. For all metrics, 10 optimisations have been performed and the average metric value, with its standard deviation has been plotted. All plots are normalised to the value before optimisation. The horizontal axis refers to the metric which has been used for optimisation, and the colour shows the metric used for the final measurement on the optimised image.
Figure 3.1: Refractive index mismatch at a planar interface (n2 < n1). planar interface [86, 21, 37].
Figure 4.10: Comparison of different optimisati on schemes based on the sim plex algorithm using the standard deviation metric. The blue and greet plots represent optimisations using mirror actuators as variables. In this con figuration, we used a 52 actuators mirror.,The red plot shows an optimisa tion considering the Zernike modes as the variables. The blue plot is us ing random initial voltage within the range 4 t+5% of the maximum stroke The red and green plots are using a random set of Zernike modes including (2,0),(4,0),(2, 2),(2, —2),(2, 2),(3, —1),(3, 1) each of these modes having a moda coefficient amplitude randomly chosen with a uniform distribution over the in terval [—0.5A, +0.5A] (wave RMS).
Figure 7.5: (a) A portion of the light sheet is used to illuminate an isolated feature in the heart section, creating an artificial guide star. (b) The feature is used by the wavefront sensor for feedback of the DM. Figure 7.5. Direct wavefront sensing will afford real time 3D correction of the
mirror’s surface as depicted in 3.4(b).
Figure 3.3: Sample holder techniques used in light sheet microscopy. Picture explained in the text.
‘igure 5.2: Illustration of the use of the back scattering light as a laser guid tar for wavefront sensing. In (1), the beam (which is not visible) located at th entre of the red square is focused on a weakly scattered part of the sample. h 2), the wavefront sensor camera, for this position, is shown. Only a few spot: orresponding to the specular reflection on the slide and the sample are present n (3), the probe beam is now focused on a highly scattered tissue, leading, it 4), to a full covering of the WFS aperture. As no pinhole is used, the spo hape inhomogeneity is noticeable.
(d) Sobel metric variation with im- (ce) Wavelet metric variation with age content image content
(b) Solution for the correction of the light sheet tilt and position - Top view : A mirror with a controlled tilted is placed in the conjugate plane to the back aperture for axial adjustment of the light sheet position along the imaging axis. - Bottom view : Another mirror with a controlled tilt is placed in the conjugate plane to the sample, for adjustment of the light sheet tilt.
both indexations. Figure 2.3: first 15 Zernike modes with the dual indexation The Figure 2.3, gives a representation of the first 15 Zernike modes, with
Figure 2.9: Illustration of a one level 2D wavelet transform using the Haar wavelet (low pass and high pass filter represented on the bottom left). The 2D discrete wavelet transform is the result of a convolution with a low and high pass filter and a downsampling by a factor of 2. To begin with, both filters are applied vertically, then horizontally resulting in 4 subimages.
Figure 3.11: phase retrieval optimisation example. From the starting phase and uniform amplitude, the PSF is computed and then compared to the measured one. At each iteration, the algorithm computes a new phase and amplitude based on the intensity PSF. The merit function decreases accordingly, until it reaches a plateau indicating that the computed intensity PSF is the same as the measured one (within a given tolerance)
(a) Solution using an incoming off-axis beam. The blue beam is the incoming beams. The red and green are the reflected beams respec- tively on the front and rear surfaces of the crystalline, and are deflected from the pupil entrance of the wavefront sensor. The grey beam is the diffuse scattered beam created at the retina.
(a) Intensity squared metric varia- (b) Standard deviation metric vari- tion with image content ation with image content
Figure 2.5: Simulated lateral and axial normalised PSF of an aberration-free and aberrated system. of diameter 2mm with a focal length of 128mm, and the wavelength is 500nm. In the computer generated example, the system is limited by a circular aperture
(a) Sobel filter applied to the in-focus image (b) ) Sobel filter applied to the defocused im- 1.17. Normalised metric = 1 age of Figure 2.7(b). The normalised metric value is 0.0026.
(a) Optimised wavefront:1.43 » PTV (b) 12 first Zernike coefficients for the 3 type of con- and 0.314 \ RMS - (A = 0.633 pm) figurations
(a) wavefront sensor spots and configura- (b) Calculation of the x and y spot centroid tion on the Zernike mode (2,2)
Table 6.1: Modal amplitude standard deviation measured in Figure 6.21 for the 5 selected metrics. and compare how the error varies depending on the choice of the metrics.
the metric, the average and the standard deviation were computed. Figure 4.12: Relative improvement of the metric value for the 5 different metrics when optimising on Figure 4.9 . For all metrics, 10 optimisations have been performed and the average metric value, with its standard deviation has been plotted. All plots are normalised to the value before optimisation. The relative improvements are 2.8%, 9%, 24.8%, 32.8% and 72% respectively for the intensity squared, the standard deviation, the Sobel, the Fourier, and the wavelet based metrics.
Figure 4.16: The profile before (blue line) and after (red line) optimisation performed on Figure 4.13 with the wavelet metric. The location of the profile is given on Figure 4.13 by the green line.
Figure 6.9: Optical Configuration showing the AO SPIM. The illumination light is shown in blue - to the left, and the imaging light is shown in green, to the right. The calibration beam is in red. The symbols are explained in detail in the text.
Figure 6.18: Zernike mode amplitude at different depths in the FEP tube. is low. The error on the Lorentzian fit, is, on average 0.013 waves RMS.
in order to increase the speed of the system. Figure 4.8: A wavefront sensorless AO transmission microscope with a bright- field illumination.
(b) Alteration of the phase by a 850 wm pinhole. The RMS difference is between the transmitted phase through the pinhole and the phase propa- gated without any pinhole. All plots are terminated when the local peak intensity in the exit pupil ,is less than 5% of the intensity in the presence of no aberration.
Figure 7.3: Implementation of HiLo background rejection on a non-scanning light sheet microscope. The cylindrical lens produces a line which is relayed onto the back aperture of the illumination microscope via a 4f relay. At the focal distance of each lens, a reflective SLM is placed and used in amplitude modulation only.
convergence and a statistical analysis was then performed. Figure 4.9: Typical optimisation with ”before” image on the left and ” after” image on the right. The white scale bar corresponds to 20 xm
Figure 6.20: Heart synchronisation module in black. The DM is represented in blue, and is running in parallel. The description is given in the text.
Figure 3.8: Direct wavefront sensing configurations
Figure 7.1: Top left view : The light sheet is in the plane of the drawing and is scanned from left to right. The sample is observed from the imaging direction - Top right view : The rolling shutter is synchronised with the scan of defocus applied to the light sheet so the narrowest part of the sheet is imaged onto the camera and the other part rejected. Bottom left view : section of the light sheet, showing the position of the beam waist. - Bottom right view : image on the camera of the fish section observed without (left) and with (right) scanning of the light sheet along the excitation axis. sample (region on the right of the sample on Figure 7.1).
Figure 5.4: Closed-loop AO transmission microscope The optical set-up is shown in Figure 5.4.
(b) Illustration of the double path effect in the case of an odd (top) and even (bottom) aber- ration
Figure 4.6: Sharpness metric variation with a uniformly distributed noise and defocus when applied to image 4.1(a). The colour code corresponds to the upper limit of the uniform distribution. For example, for one (grey plot), the noise is uniform in the grey level range [0 - 1]. The plots display the average of 20 values and the error bars correspond to the standard deviation. The error bars remain very small, with a maximum of 2% for Sobel and wavelet metrics.
(a) Aberrations in a light sheet microscope due to deviation of the light sheet - Left view : the light sheet entering the sample from the left is deviated by a surface creating an angle between the light sheet and the imaging plane. - Right view : effect of the tilt perceived on the image, the image is seen sharp on the left and the blur is increasing to the right.
Figure 4.15: Zernike modal coefficient per metrics
Figure 3.12: Simplex algorithm organisation chart. The best value corresponds here to a minimum.
Metrics based on the wavelet filter are assessed using the python package
Figure 6.1: The top view represents the SPIM when placed in the plane of the microscope. In this section, the optical power of the cylindrical lens is visible. n A The focus of the lens is placed in the back aperture of the illumination objective. t The bottom view represents the SPIM as seen from the top. The interaction of he light sheet and the sample create fluorescence which is then collected by an imaging microscope.
Figure 6.8: Variation of defocus with NA (brown), glass tube refractive index (blue), tube thickness (green) and inner radius (magenta).
Hartmann WFS for the closed loop system. Figure 5.7: Melanin dots in a hair follicle in back mouse tissue, on which the optimisation and closed loop are run. The red square corresponds to sample 1 and the green square is sample 2. The reference wavefront is recorded on the optimised image of the sample 1.
from the first surface of the cylinder. Figure 6.13: AO in SPIM with a zebrafish in a glass borosilicate pipette. (a,b,c) are images taken for a flat mirror shape, a mirror shape optimised for the system aberrations, and for a mirror shape optimised directly on the fish. The white square corresponds to the ROI on which the optimisation is performed and is 24 microns wide.
Figure 5.11: Metric value comparison while the system is optimising on sample 2 in a sensorless configuration using the simplex (shown here in blue), and in a sensored configuration (violet). After reaching the plateau, in the closed loop configuration, the stability of the system is checked by sending random aberrations onto the mirror.
(a) Live zebrafish heart section. - Left: Image taken with synchronisation but with AO off. Right: Image with synchronisation and AO on.The horizontal side of the white rectangle is 64 wm.
Figure 6.21: Right vertical axis - Variation in the estimation of the defocus and astigmatism modal amplitude over 101 measurements using modal optimisation (based on 3 images). Each point gives the difference between the measurement (defined as being the location of the Lorentian maximum which best fits the 3 samples) and the average of the 101 measurements. The different colours correspond to the different metrics. Left vertical axis - the frame to frame image motion (black solid line plot) is represented.
Figure 6.19: Example of normalised metric variation with or without synchro- nisation. (b) and (c) shows the 2 heart ventricles in their extreme positions.
Figure 6.12: Measured and simulated focus and astigmatism variation with depth at a wavelength of 550 nm.
Figure 7.4: concept of a reversible SPIM - DM refers to Dichroic mirror, and CL refers to cylindrical lens. The two identical objectives allows the visualisation of the sample along two orthogonal views and can also simultaneously produce a light sheet with the help of an illumination system placed behind the CL.
Figure 5.10: Image of sample 2 before optimisation (a), when the closed loop is ON (b) and after a sensorless optimisation (c). The green square delimits the widefield optimisation area during sensorless optimisation.
Figure 6.7: Variation of defocus in the tube (the horizontal and vertical axes represent the z and the y positions respectively as represented in Figure 6.4). The illumination beam is propagating into the tube from the bottom to the top and the imaging objective is located on the left. The top left image corresponds to the defocus coefficient created on the imaging path only. The top right image represents the defocus as seen on the imaging path but created only by the illumination. The bottom image represents the global contribution. The white line represents the location of the Z scan performed in the experimental part with the beads. The scale bar represents the Zernike coefficient amplitude of mode (2, 0) in A RMS. Simulation obtained with Zemax.
Figure 6.11: Variation of the metric during the 4 optimisation runs (different zones are described in the text).
shape for a given depth determined from the previous calibration, as before. Figure 6.16: Optimisation on zebrafish in a refractive index matching FEP mounting tube. The images show part of the pectoral fin. (a,b) are images taken for a flat mirror shape and for a mirror shape optimised directly on the fish. As discussed in the text - the improvements here are marginal (although clearer in the original data than that shown here). The white square corresponds to the ROI on which the optimisation is performed and is 19 um wide.
Figure 6.6: Variation of astigmatism with NA (brown), glass tube refractive index (blue), tube thickness (green) and inner radius (magenta).