Venue: Hyderabad International Convention Centre
Reception attendees
08:30 - 09:00
New York University, US
Title: Alternate Approaches to Quantization and Compression Using Higher Order Medians
Abstract: Quantization is a fundamental process that transforms continuous data into discrete representations, striking a balance between compression efficiency and signal fidelity. In this talk, I will introduce a novel quantization approach that minimizes the absolute quantization error (L1-error), leading to what we define as a Median Quantizer. I will also discuss a simplified version where the quantization interval boundaries are approximated using the median of output values, yielding the well-known Uniform Quantizer, which, while widely used, is suboptimal compared to our L1-based median quantizer.
Beyond these classical approaches, I will present a broader generalization of quantization through higher-order error minimization, including fractional values. This leads to the notion of a generalized median quantizer, which optimally balances quantization error across a range of scenarios. Through comparisons with the widely used Lloyd-Max MMSE quantizer, I will demonstrate the effectiveness of our proposed approach, particularly in scenarios involving low signal values. Using standard images, I will illustrate how this novel quantization framework can improve performance in various practical applications.
This talk will provide new insights into the role of a new median-based quantization in modern signal processing and open discussions on its implications for data compression, imaging, and beyond.
09:00 - 09:30
CREATIS, Université de Lyon, France
Title: Adaptive beamforming for ultrasound localization microscopy
Abstract: Ultrasound Localization Microscopy (ULM), a technique developed over the last ten years, provides ultrasound images of vascular microstructure with micrometric resolution and centimetric depth. ULM involves detecting ultrasound contrast agents (microbubbles) in the bloodstream and tracking their displacement during ultrafast ultrasound acquisition. To map the vascular microstructure, several steps are required following ultrasound acquisition: beamforming, detection and isolation of microbubbles, location of their centers and tracking of their displacement. One of the disadvantages of ULM is the acquisition time required to obtain a high-resolution image with a maximum number of microvessels (3-5 minutes). Reconstruction of ultrasound images using adaptive beamforming methods improves the spatial resolution of ultrasound images and reduces image noise depending on the method. We investigated the benefits of adaptive beamforming for ULM on 2D simulation and in vivo ultrasound acquisitions, as well as in 3D in vitro and in vivo ultrasound acquisitions. The interest adaptive beamforming for ULM was evaluated and compared with the classical beamforming method, the Delay-and-Sum.
09:30 - 10:00
US Naval Research Laboratory
Title: Some Vignettes in Computational Imaging: Radar and Tomographic Applications
Abstract: We describe several novel techniques that we have recently developed for incorporating deep statistical and physics-based priors into computational imaging algorithms. Our formulations deep statistical priors are with respect to the compound Gaussian (CG) distribution whereby we derive iterative algorithms for solving linear inverse problems, particularly of the type that appear in tomographic imaging and compressive sensing. Thereafter we describe algorithm unfolded versions of these algorithms and derive a detailed computation theory underlying these algorithms together with generalization bounds for CG unfolded deep neural networks. We conclude with a description of novel methods that we developed for incorporating physics-based prior knowledge of scene structure into construction of computational imaging methods for solving nonlinear linear inverse problems in radar imaging.
10:00 - 10:30
Title: Space-Time Reconstruction for Dynamic Fourier Ptychography and other Synthetic Aperture Imaging Problems
Abstract: Synthetic aperture methods such as Fourier Ptychography are an appealing approach for quantitative phase imaging, for example in microscopy. Unfortunately, quantitative phase imaging of living biological specimens is challenging due to their continuous movement and complex behavior. Here, we introduce space-time Fourier ptychography (ST-FP), which combines a fast Fourier ptychography (FP) model based on compressive sensing with space-time motion priors for joint reconstruction of quantitative phase, intensity, and motion fields across consecutive frames. Using the same input data as compressive sensing FP, ST-FP increases the space-bandwidth-time product of the reconstructed complex image sequence while leveraging redundant temporal information to achieve robust reconstruction performance. I will also talk about generalizations of this framework to other imaging modalities.
10:30 - 11:00
University of Siegen, Germany
Title: A Single-shot Fourier-sensing Camera
Abstract: In many real-world scenarios, the challenge is to retrieve the location and amplitude of sharp functions that may model, e.g., the response of a target to an illumination field, the effect of an interface between different media, a fluorescence response, or the location of a finite set of active sources. The former can be modelled by means of Dirac delta functions. Consequently, acquiring them directly translates into exorbitant sampling rate requirements and data volumes. In contrast, sampling them in the frequency domain constitutes an optimal strategy due to the maximal incoherence between the Fourier and the trivial bases. This brings in the need for sensors able to implement Fourier kernels as sensing functions. Furthermore, in many applications imaging is desired, what requires large arrays of miniaturized detectors.
In this talk we present computational hardware developments aimed to satisfy these needs. More specifically, an array of macro-pixels is introduced that operates synchronously with a pulsed illumination system. Each macro-pixel acquires data in Fourier domain at several demodulation frequencies simultaneously, thereby yielding a single-shot Fourier-sensing camera. Each subpixel of a macro-pixel is controlled by a periodic demodulation control signal of different frequency. Induced resonance at each desired frequency is advantageously exploited to annihilate harmonic distortion. As a result, highly-sinusoidal sensing functions can be realized.
The camera has been evaluated in the context of ToF imaging, including resolving the resolving multipath interference problem. Realistic simulation results show the ability of separating multiple targets using the multi-frequency measurements. Both parametric spectral estimation methods and simpler approaches, such as a super-resolved inverse Fourier transform followed by peak detection, have been considered for signal retrieval.
This novel sensor concept constitutes an invaluable asset in fields beyond ToF 3D imaging, such as transient imaging, ultrafast photography of sparse events, high-energy particle detection, and non-line-of-sight imaging.
11:00 - 11:30
11:30 - 12:00
Tampere University, Finland
Title: Computational Imaging with EDOF and diffractive- and meta-masks
Abstract: Computational imaging systems integrating encoded diffractive optical elements (DOEs) and meta-optical elements (MOEs) are increasingly replacing traditional imaging systems with refractive lenses across a wide range of applications. This shift is driven by their ability to precisely manipulate light, and combine optical and computational modules into a single, co-designed hardware and software system. This integration offers substantial benefits over conventional optics, including compound optical systems, providing superior imaging accuracy and quality, compact form factors, cost-effective manufacturing, and enhanced capabilities such as extended depth of field (EDOF) and achromatic broadband imaging.
This talk presents recent advances in lensless and hybrid computational imaging systems design, describe physical experimental setups, and show how hardware-in-the-loop (HIL) methodology can enable the enhanced imaging.
12:00 - 12:30
Title: Distributed Apertures for Long Wavelengths
Abstract: We consider long wavelength distributed apertures for sensing and communications. Low frequency yields significantly reduced scattering and enhanced penetration of obstacles and structures and can be deployed on teams of small autonomous platforms. We describe distributed collaborative sensing and beamforming for air and ground platforms in complex terrestrial environments. Exploiting long wavelengths enables methods that overcome position uncertainty and can learn to handle mobility, achieving coherent processing over distributed apertures.
12:30 - 13:00
Title: Phase retrieval using designed Hadamard complementary coded apertures
Abstract: Phase retrieval has applications in various fields, from microscopy to astronomy. Currently, computational imaging systems exploit encoding using random complementary coded apertures to reconstruct the amplitude and phase from a distorted beam without prior knowledge, thus solving an ill-posed problem using the phaseLift algorithm. However, conventional approaches recover the phase and amplitude using 20 coded apertures, which poses challenges in performing real-time acquisition and estimation of varying optical fields. To overcome this issue, we recently introduced eight Hadarmard complementary coded apertures that enhance the quality of amplitude and phase reconstruction. We test our approach in a simulation scenario using 23 images of the Kodak dataset, phaseLift as a reconstruction algorithm, and Fresnel as a light propagator in the near field. Extensive simulations prove that our Hadamard complementary coded aperture outperforms conventional methods of random complementary coded apertures in reducing the number of masks. Furthermore, experimental results using our optical test-bed demonstrate that our approach recovers the phase with significantly enhanced image quality.
13:00 - 14:00
14:00 - 14:30
Title: Cognitive Radar. From Adaptation to Inverse Reinforcment Learning
Abstract: This tutorial comprises two parts. The first part in an depth presentation of convex optimization methods for adaptive radar. It covers, constrained covariance estimation, optimization for practical waveform design and also large random spiked covariance matrix based models for clutter. The second part of the tutorial discusses inverse reinforcement learning for cognitive radar. It addresses the following questions: How to identify if the radar is cognitive? How to estimate the radar’s utility function and therefore predict its future actions? The aim is to detect the presence of a cognitive radar from its decisions. We will discuss the interaction between adversarial signal processing and inverse reinforcement learning in cognitive radar using revealed preferences from micro-economics stochastic optimization, and inverse Bayesian filtering.
14:30 - 15:00
Rutherford Appleton Laboratory, UK
Title: AI-Driven Advances in Computational Imaging: Enhancing Synthetic Apertures for Scientific Discovery
Abstract: Synthetic aperture techniques have transformed computational imaging by enabling high-resolution reconstructions from sparse or limited data. Meanwhile, artificial intelligence (AI), had been instrumental in transforming how large-scale experimental facilities, like light sources, acquire, process, and interpret experimental data. This talk explores how AI and machine learning enhance synthetic aperture imaging by improving signal reconstruction, solving inverse problems, and accelerating simulations. Drawing on examples from large-scale experimental facilities, the talk will discuss how AI enables new possibilities in high-throughput imaging, adaptive optics, and real-time experimental feedback. The talk will also highlight key challenges and opportunities in integrating AI with computational imaging pipelines to accelerate scientific discovery.
15:00 - 15:30
CREATIS, France
Title: Synthetic Apertures in Medical Ultrasound Imaging: Principles and Applications
Abstract: Medical ultrasound imaging is a technique for acquiring images of the internal body without any invasive procedures, making it a valuable tool for medical diagnosis and treatment. The traditional pulse-echo modality involves individual transducer elements emitting acoustic waves into a medium and recording the returning echoes with the same transducer element, forming the image line by line.
Beamforming is the primary reconstruction technique responsible for generating a spatial map of the pressure field within the region of interest based on the recorded echoes. However, the beamformed image obtained from the conventional modality often faces limitations, including low resolution, poor contrast, and a low signal-to-noise ratio, which can degrade image quality and make ultrasound image interpretation more challenging.
The Synthetic Transmit Aperture (STA) is a technique that synthetically expands the transducer array, mitigating these limitations. The key idea of STA is to sequentially transmit signals from individual or small groups of transducer elements while receiving echoes on all elements, forming a complete image each time. This process significantly improves axial and azimuthal resolution, contrast, and, most notably, the signal-to-noise ratio (SNR).
This talk will cover the principles of medical ultrasound imaging, the operation of the transducer in transmission and reception, and conventional beamforming techniques, including time-domain and Fourier-domain approaches. It will also highlight how synthetic aperture techniques can improve the quality of the beamformed image.
15:30 - 16:00
16:00 - 17:30
Panel Discussion: The Future of Imaging