Annonce


Internship offer for 2nd-year MSc or 3rd-year engineering school students: 5 or 6 months. Possibility of PhD continuation, initially based in Toulouse, France, followed by a research attachment of up to 2 years at A*STAR, Singapore.

16 Octobre 2026

Catégorie : Postes Stagiaires ;

Inverse Problems Learning for Image Restoration and Ultrasound Application

Keywords:     inverse problems, image restoration, unsupervised learning, ultrasound imaging.

1        Project Description

Ultrasound imaging is widely used in medical diagnosis, but ultrasound images are often affected by noise, limited spatial resolution, and acquisition-related degradations. Image restoration can therefore be formulated as an inverse problem, where the objective is to recover a high-quality image from degraded measurements. Classical model-based approaches provide physical interpretability but rely on accurate forward models and appropriate regularization, while deep learning approaches can learn powerful image priors but often require large amounts of training data and ground-truth images.

Recent approaches based on learned proximal operators and algorithm unfolding have shown that mathematical optimization and deep learning can be effectively combined for image restoration [1], [2]. More recently, diffusion models have provided powerful generative priors for solving a wide range of inverse problems. These developments open promising perspectives for image restoration in challenging applications where ground-truth data are difficult to obtain.

This internship will investigate learning-based and diffusion-based approaches for ultrasound image restoration. The objective is to explore how physical knowledge of the imaging process can be combined with learned image priors to solve ultrasound inverse problems. The work will build on recent devel- opments in image restoration and inverse problems within the IRIT MINDS team, including spatially variant ultrasound image restoration [3], unsupervised plug-and-play methods for image super-resolution [4], and diffusion-based approaches for ultrasound image restoration [5]. An M1 internship will also build on previous work carried out at IRIT on related inverse imaging problems [6].

2        Project Objectives

The scope of this project includes the following objectives:

  1. Review the literature on inverse problems, proximal/unrolled neural networks, and diffusion-based image restoration, for example, [1], [2], [3], [6].
  2. Study and implement existing these approaches for image restoration.
  3. Investigate and develop their adaptation to ultrasound image restoration, taking into account the underlying imaging model.
  4. Develop and evaluate an approach for ultrasound image restoration using simulated and real ultra- sound data.

3        Required skills

  • 2nd year of Msc and/or 3rd year of an engineering school,
  • Strong background in signal and image processing, or applied mathematics,
  • Proficiency in machine learning (ML) techniques, with a particular focus on CNN,
  • Strong programming skills in either Matlab or Python,
  • Proficiency in the English language,
  • An interest in medical US imaging would be a plus, without requiring any prior knowledge.

4        Starting date

As soon as possible and no later than mars 31, 2027.

5        Application

Prospective candidates should submit the following documents as a SINGLE PDF file: 1) University transcripts; 2) A one-page cover letter; 3) Curriculum vitae (CV), publications if applicable;

Please send all documents to duong-hung.pham@irit.fr and denis.kouame@irit.fr.

Shortlisted candidates will be contacted for an interview.

References

  • M. Jiu and N. Pustelnik, “A deep primal-dual proximal network for image restoration,” IEEE

J. Sel. Top. Signal  Process.,  vol.  15,  no.  2, pp. 190–203, 2021. doi: 10.1109/JSTSP.2021. 3054506

  • H. T. V. Le et al., “Unfolded proximal neural networks for robust image gaussian denoising,” IEEE Trans. Image Process., vol. 33, pp. 4475– 4487, 2024. doi: 10.1109/TIP.2024.3437219
  • A. Floquet et al., “Spatially variant ultrasound image restoration with product convolution,” IEEE Trans. Ultrason. Ferroelectr. Freq. Con- trol, vol. 72, no. 9, pp. 1235–1244, 2025. doi: 10.1109/TUFFC.2025.3584533
  • X. H. Le et al., “PlaSo:  Unsupervised resid- ual plug and play for image super-resolution,” IEEE Signal Process. Lett., vol. 32, pp. 4059– 4063, 2025. doi: 10.1109/LSP.2025.3620784
  • X. H. Le et al., “GLUS: Guidance loss driven plug-and-play diffusion for ultrasound image super-resolution,” in Proc. IEEE Int. Symp. Biomed. Imaging (ISBI), London, UK, 2026, pp. 1–4. doi: 10 . 1109 / ISBI61048 . 2026 . 11515630
  • K. Bjorner, “Inverse problems learning for im- age restoration,” Master 1 internship report, May–August 2026, M.S. thesis, IRIT, 2026.

Les commentaires sont clos.