Image Denoising in High Dimensional Space
Introduction
Non-Local Mean (NLM) Filter is commonly used for filtering Cryo-Electron Tomographic data which is contaminated by mixture of Gaussian and Poisson noise distributions at the projections level, while the noise distribution in the whole tomogram (after 3D reconstruction) is a little bit complex. The NLM is computationally expensive, especially for 3D data. In this project, we are going to investigate the efficiency of dimensionality reduction on sparse representation of the data to find the distance between blocks in lower dimensional space, which can be used for image denoising. The following diagram shows you the overall idea of denoising the Cryo-ET data:
Project Phases
| Phase | Period | Conducted by | Materials |
|---|---|---|---|
| Intro. to CryoET A Review of Image Denoising Algorithms | 3-10 Nov | El-Fiky | Slides Journal Paper 2005 FNLM Code |
| Denoising via Learned Dictionaries | 10-17 Nov | El-Fiky | Paper:CVPR 2006 |
| Principal components for non-local means image denoising | TBA | El-Fiky | Paper: ICIP 2008 |
| Dimensionality Reduction and Manifold Learning | TBA | El-Fiky | Slides Paper |
| Bonus: Sparse Subspace Denoising for Image Manifolds | TBA | El-Fiky | Paper: CVPR 2013 |
Programming Language
- MATLAB
What you are going to learn:
- Clustering and Dictionary Learning
- Figure out the Linear/Non-linear Dimensionality Reduction techniques
Useful Links
- [[][CryoET Dataset]]
- Spams-Toolbox
- Dimensionality Reduction-Toolbox
- Github