|
- Date: Tuesday December 11th, 2012
- Time: 11:00
- Location: Room 01.09.014, FMI-Building, Garching, Boltzmannstr. 3
|
Image-based Characterization of Thrombus Formation in Time-lapse DIC Microscopy - Segmentation under Low Contrast and Highly Dynamic Imaging Conditions
Abstract:
The detection of salient events in video sequences displaying low contrast and high dynamic conditions is becoming of increased interest for the community of computer vision. This thesis introduces novel methods for the analysis of such scenes, and in particular for automatic detection of thrombus formation in low contrast in vivo microscopic image sequences. This application is of crucial interest for identifying genetic variations which lead to an increased risk of developing cardiovascular diseases (CADs). Our key idea is to perform segmentation of the thrombus by distinguishing different motion patterns in image time-series rather than by performing a standard image segmentation task in each frame. We model motion patterns by energies based on the concept of Dynamic Textures (DT) and regularize segmentation by two prior energies on the topological relationship between the thrombus and the aortic vessel and on the shape of the aortic region.
