Related Experiment Video
Updated: Sep 22, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
652
An approximate expectation-maximization for two-dimensional multi-target detection
Shay Kreymer1, Amit Singer2, Tamir Bendory1
1School of Electrical Engineering of Tel Aviv University, Tel Aviv, Israel.
IEEE Signal Processing Letters
|May 23, 2022
Summary
This study introduces a new expectation-maximization framework for multi-target detection (MTD) to reconstruct images from highly noisy data, outperforming previous methods in single-particle cryo-electron microscopy applications.
Area of Science:
- Image reconstruction
- Computational imaging
- Signal processing
Background:
- Single-particle cryo-electron microscopy (cryo-EM) requires reconstructing 3D structures from noisy 2D images.
- Existing methods struggle with high noise levels where individual targets are undetectable.
Purpose of the Study:
- To develop a robust method for estimating target images from noisy measurements containing multiple rotated and translated copies.
- To address the challenges of multi-target detection (MTD) in high noise regimes.
Main Methods:
- Developed an expectation-maximization (EM) framework to approximate and maximize the likelihood function.
- Applied the framework to the two-dimensional multi-target detection problem.
Main Results:
- Successfully demonstrated image recovery in highly noisy environments.
- The proposed EM framework significantly outperforms traditional autocorrelation analysis across various parameters.
Conclusions:
- The novel EM framework provides a powerful solution for image reconstruction in challenging, high-noise conditions.
- This approach advances capabilities in fields like single-particle cryo-EM where data quality is often compromised.
