Contact: Dr. J. Rué Queralt, Dr. D. Sage (EPFL Center for Imaging), Dr. V. Stergiopoulou (LCAV, Galatea), Dr. E. Soubies (IRIT Toulouse)
Synopsis: Modern fluorescent microscopes are capable of acquiring high-resolution 3D images of an entire specimen (e.g. light sheet microscopy). This generates very large, voluminous images that are difficult to process on a computer and unravel using deconvolution algorithms. This project presents an exciting opportunity to develop deconvolution algorithms with a domain decomposition strategy in order to address the current limitations in handling 3D large-scale data.
Level: Master (Master Thesis / Internship)
Description: This project will explore the integration of domain decomposition for advanced deconvolution algorithms. This method aims to enhance the deconvolution process’s efficiency, allowing it to manage large images more effectively. The technical approach will involve the finding of the optimal strategy for the 3D decomposition of the image and the point-spread function (PSF). Inside the deconvolution, a new scheme has to be introduced in order to limit the border artifacts due to the 3D decomposition. The algorithm will be deployed in parallel computing environments across multiple CPUs or GPUs, leading to a significant boost in computational speed. The project will also delve into optimizing these algorithms for performance, leveraging parallel computing environments and possibly GPU-based acceleration.
Objectives:
- Develop a domain-decomposition deconvolution with seamless borders.
- Optimize the algorithm for handling large microscopy images.
- Evaluate the performance in terms of accuracy and computational efficiency.
- Apply the algorithm to real-world microscopy data to validate its effectiveness.
Deliverables:
- A domain-decomposition deconvolution able to deal efficiently with large images.
- Implementation of the algorithm in a high-performance computing environment.
- A comprehensive set of tests and benchmarks on microscopy datasets.
- A potential publication detailing the algorithm and its applications in scientific imaging.
Prerequisites:
- Background in digital signal processing.
- Programming skills in Python.
- Knowledge of parallel computing would be beneficial.
- Understanding of computer science concepts and a keen interest in addressing scientific questions through computational methods.
Type of Work :
- Algorithm Design: 40%
- Implementation and Optimization: 40%
- Data Analysis and Testing: 15%
- Documentation and Dissemination: 5%