Description
Date depot: 1 janvier 1900
Titre: Compressed Sensing Image Acquisition for Dynamic Digital Holography Microscopy
Directeur de thèse:
Jean-Christophe OLIVO-MARIN (Analyse d'images biologiques)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
{{Overview}}
This project proposes to study the applications of the mathematical framework of Compressed Sensing (CS) to digital holography microscopy (DHM). It will elaborate dynamic CS DHM imaging protocols and setups based on Fourier-domain measurements with random array detectors and dedicated parameterization of image reconstruction processes. The project will consider the design and implementation of dedicated dynamic compressed sensing acquisition protocols for DHM microscopy imaging.
{{Digital Holography Microscopy and CS}}
Holographic microscopy is a rapidly growing image-forming method. In particular, its ability to record a diffracted optical field in amplitude and phase on a sensor array is the basis of the successful lensless and diffractive imaging schemes, which are amongst the most promising methods for achieving heavily increased pixel density images with respect to standard image-forming approaches. Optically-acquired holograms have proved to be an efficient and physically realizable sensing modality that can exploit the benefits of compressive sensing, as they are naturally compressed measurements in a reciprocal space.
{{Previous work on the project}}
This project is the continuation of several years of collaboration between the BioImage Analysis (BIA) group at Institut Pasteur, the Institut Langevin and the Image Processing and Interpretation group at Telecom ParisTech, who specialize respectively in biological image analysis, optical instrumentation and signal and image processing. We are among the very first groups worldwide having proposed and applied the concepts of compressive sensing in biological imaging and demonstrated it in a real biological context. This collaboration has led to two PhD thesis (Marcio Marim in 2010 and Yoann Le Montagner in 2013), which have set the ground of CS-based microscopy imaging, proposing novel and dedicated sensing paradigms, reconstruction algorithms, parameterization strategies, and denoising methods.
More recently, we also contributed several original algorithms for CS video acquisitions (Montagner, Angelini et al. 2012), phase retrieval (Montagner, Angelini et al. 2013) and denoising parameterization (Montagner, Angelini et al. 2014). We also published a large set of experiments documenting the choice of CS optimal sampling patterns (Montagner, Angelini et al. 2011b), sampling bounds (Montagner, Marim et al. 2011) and reconstruction algorithms (Montagner, Angelini et al. 2011a).
{{Overview of the planned work}}
With this project, the goal is to implement CS DHM imaging on real optical setups and video acquisitions, as well as continue algorithmic developments and learning of dedicated dictionaries to handle high level of Poisson-Gaussian noise when working under low-light conditions, high data throughput for video imaging and 3D tomography in DHM.
The project will be articulated in three tasks:
1. Modeling and hardware implementation of CS microscopy acquisition with a specific CMOS-random camera, involving realistic simulations of the hardware limitations and temporal versatility, and then programming of the camera for real experimentations with optimized parameterization of the sensing protocol.
2. Design of CS-enhanced holographic microscopy for 3D tomography, low-lights conditions and high-throughput. Three distinct CS optimizations will be investigated: (1) resolving power, pixel density, axial resolution, and signal-to-noise ratio in low-light conditions (involving some learning and modeling of the noise and scattering components); (2) achieving very high-speed imaging, beyond current levels; (3) identifying sparsifying basis tuned to 3D scattering patterns encountered in 3D DH tomography.
3. Previously developed dynamic CS imaging protocols will be implemented, identifying the nature of the temporal DHM information that can be used as updates of a series of images being observed and recorded. This task will be evaluated in the context of in-vivo imaging on small animals.
Doctorant.e: Meiniel William