Speaker
Description
Background and Motivation:
In Near-field holography effective denoising and flat field correction are critical for accurate image reconstruction. Traditionally, flat field correction relies on principal component analysis (PCA), which computes the PCA components from a series of images without objects and subtracts the PCA representation from the images with objects. While effective, PCA is a linear method and may not fully capture the complex variations in holographic imaging conditions.
Recent advances in deep learning have demonstrated the ability of neural networks to extract rich, nonlinear feature spaces from data. These feature spaces can serve multi-purpose roles, including denoising and flat field correction. This project proposes a unified approach that leverages the feature space of a denoising deep learning model not only for noise removal but also as a replacement for PCA components in flat field correction. By using the same model for both tasks, the proposed method aims to enhance the accuracy and robustness of the correction process while simplifying the overall framework.
Significance and Impact:
This project has the potential to significantly advance the field of holographic imaging by introducing a deep learning-based alternative to PCA for flat field correction. The proposed method can improve image quality, facilitate more accurate scientific analyses, and pave the way for broader adoption of machine learning techniques in optical imaging. The outcomes of this research can benefit applications in biomedical imaging, material science, and other fields that rely on holographic imaging techniques.
Strengths of the Idea:
- Combining Denoising and Flat Field Correction: Leveraging the same feature space for both tasks is efficient and reduces the need for separate models or processes.
- Deep Learning Over PCA: PCA is linear, while deep learning can model nonlinear, complex variations, which are more representative of real-world imaging conditions.
- Feasibility in Six Weeks: Since the denoising model is already developed, the project focuses on optimizing and adapting its feature space, making the scope manageable within the timeline.
- Scientific Impact: A more robust flat field correction method using deep learning can significantly improve image quality, making it highly impactful for fields like biomedical imaging and materials science.
Special Qualifications
programming, machine learning
| Group | FS-PETRA / CXNS |
|---|---|
| Project Category | A6. Theory and computing |
| DESY Site | Hamburg |
