Logo der Universität Wien

Termine der Fakultät für Informatik

Title: Masterprüfung mit Defensio, Kukharenka Tatsiana
Location: Besprechungsraum 4.34, W29

07.11.2017

16:30 Uhr

Besprechungsraum 4.34, Währinger Str. 29, 1090 Wien

 

"Stitching microscopic images from cell samples"

The topic of my master thesis is the stitching of microscopy images of cell samples. Images to be analyzed are provided by Miltenyi Biotec, with whom a research collaboration is ongoing. Image stitching is the process of combining multiple image tiles with overlapping parts to create one composite image [37]. Nowadays, a lot of tools exist for image stitching, most of them focusing on creating landscape and architectural panoramas. They typically require an image overlap of more than 20% [59]. From an analysis of the provided cell images, we know that their overlap is only about 1-5% and images are both rotated and translated. Furthermore, the images have low contrast and contain repeated structures, making the identi cation of distinct features challenging. Hence, the application of general purpose image stitching algorithms and tools is not possible. To solve this problem, I propose a stitching method based on the work of Preibisch et al. [48], that adds extra preprocessing steps to shrink the search regions and increase local image contrast and also uses phase correlation [33] to detect pairwise overlaps between the images. The proposed algorithm uses parallelization to improve performance. In the rst part of the thesis, I give an overview of existing algorithms and methods for stitching microscopic images. I then perform stitching experiments using general purpose image stitching tools [56]. In the next chapter, I describe the phase correlation method [33] and global minimization algorithm [48]. I will continue by describing my solution work ow and provide results for the application of my tool on test data. From a software development point of view, a major challenge is achieving high runtime eciency. From an algorithmic point of view, it is necessary to attain a satisfying minimization of the global stitching error and reach better overall stitching quality. The goal of this master thesis is to provide a fully automated tool that produces high quality stitched images for downstream analysis, like detection of proteins in speci c cells. The tool should run on a personal computer without any special hardware requirements. The software is implemented in Java and uses the image processing library OpenCV.

Back
Universität Wien | Universitätsring 1 | 1010 Wien | T +43-1-4277-0