University of Twente Student Theses
A Disentangled Representation Learning Approach for Deblurring Microscopy Images
Ritsema, Stan (2023) A Disentangled Representation Learning Approach for Deblurring Microscopy Images.
|
PDF
26MB |
| Abstract: | Microscopy images are vital for plant growth prediction. These images can be used by prediction models to forecast plant growth. The accuracy of such prediction models are best when the images are clear, e.g. there is no blur present in the images. However, microscope set-ups are often imperfect, which leads to microscopy images with a certain level of blur. Furthermore, some objects might not lie in the same focal plane, causing some objects to be blurry whilst others are in-focus. To remove blur from these images and improve performance of prediction models, we propose a deblurring approach based on Disentangled Representation Learning (DRL). Disentangled Representation Learning is a deep learning technique which attempts to disentangle in the latent space between multiple generative factors underlying a dataset. These generative factors each describe a separate part of the data, like size or colour. This work aims at deblurring microscopy images by disentangling image representations into two latent codes, one encoding blur and one encoding the identity of the image. This disentanglement between blur and identity is beneficial, since it allows direct altering of blur in an image without influencing the identity. The goal of our novel approach is to deblur microscopy images using this disentanglement between blur and identity in the latent space. We compare our approach with another DRL approach named Multi-Level VAE, which functions as a baseline. Furthermore, an ablation study is performed, which evaluates our contribution to earlier work on disentanglement learning. The dataset used for training and testing contains microscopy images of plant cells with different levels of blur. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science MSc (60300) |
| Link to this item: | https://purl.utwente.nl/essays/96821 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page