Self-supervised learning for phase retrieval

2025-09-30

AI Summary

Phase retrieval in scientific imaging is a nonlinear inverse problem for which supervised learning methods are hindered by the scarcity of fully sampled ground-truth data, while conventional unsupervised approaches struggle with accurate nonlinear forward modeling. This work introduces, for the first time, self-supervised learning to nonlinear phase retrieval. We propose an end-to-end reconstruction framework leveraging image translation invariance: given only a single intensity measurement, a deep neural network explicitly models the nonlinear forward process, and a translation-equivariant loss function is designed to eliminate reliance on ground-truth labels. Our method overcomes the longstanding limitation of prior self-supervised imaging techniques—restricted to linear systems—and achieves robust, generalizable performance on real experimental data. Quantitative and qualitative evaluations demonstrate reconstruction fidelity approaching that of fully supervised methods. This establishes a new paradigm for physics-based imaging tasks where ground-truth annotations are inherently unavailable.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Image and Video Retrieval
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Security and Privacy: Data transparency and provenance

Abstract

In recent years, deep neural networks have emerged as a solution for inverse imaging problems. These networks are generally trained using pairs of images: one degraded and the other of high quality, the latter being called 'ground truth'. However, in medical and scientific imaging, the lack of fully sampled data limits supervised learning. Recent advances have made it possible to reconstruct images from measurement data alone, eliminating the need for references. However, these methods remain limited to linear problems, excluding non-linear problems such as phase retrieval. We propose a self-supervised method that overcomes this limitation in the case of phase retrieval by using the natural invariance of images to translations.

Problem

• Solving nonlinear phase retrieval without ground truth data
• Overcoming limitations of supervised learning in imaging
• Using image translation invariance for self-supervised reconstruction

Innovation

• Self-supervised method for phase retrieval
• Uses natural translation invariance of images
• Eliminates need for ground truth references