Faculty of Informatics Vienna University of Technology Institute of Computer Aided Automation PRIP Home PRIP Home
Personal tools
You are here: Home Teaching Wintersemester (WS) DiplDiss Program 2011

Program 2011

Dipl/Diss Workshop Program 



Tracking with Structure

Nicole M. Artner (PhD student)

Abstract: Structural representations are still relatively unexplored for the tracking task. This is quite surprising considering that structure is an important invariant and could improve the tracking performance. This talk is about the two main topics of my work: (1) how to represent structure and (2) how to extract the structure of an object of interest out of a video. The representation of structure is realized with the help of graphs and was inspired by the work in pictorial structures (Pedro F. Felzenszwalb) and mass spring systems (used for animation in computer graphics). The spatial relationships of features describing the object are encoded in a so-called spring system and enforced during tracking by energy minimization. The structure of a scene (separate foreground from background, find rigid parts of articulated objects) is extracted by analyzing the movement of features in a video sequence and grouping them based on the similarity of their movement. The grouping process is done in an irregular graph pyramid, which results in a decomposition of the scene into its rigid parts, where each part is represented by a vertex in the top level of the pyramid.


Spatio-temporal information in dices video sequences

Fuensanta Torres García (Phd Student)

Abstract: A novel system is presented for finding and extracting spatio- temporal information in dices video sequences. In monocular videos only a part of the surface of an object is visible in each frame. However, the dices are a special case where under certain viewings angles, when they have three visible faces, it is possible to extract their 4D (3D plus time) information. This approach is a hierarchical process to extract the shape, the appearance, the position and the orientation of the dice in each frame. This hierarchical process provides advantages to handle different related tasks from visual computing.


Automated Analysis of Facial Palsy using Feature Tracking Methods

Limbeck Philip (Msc student)

Abstract: With the advent of medical analysis which is supported by Computer Vision technologies, different methods proved themselves useful to enable more accurate and reproducible diagnoses. Although computerized methods to diagnose facial palsy exist, they either lack support of accurate measurement or require intervention down to the level of manually locating the required marker points. This thesis contributes by introducing a systematic approach to evaluate the performance of automated methods using different State-of-the-Art 2D tracking methods. Additionally, to keep an eye on the medical context, a comparison to existing manual analysis is given to demonstrate the usefulness of automated 2D tracking for this application. Features like optical flow can be used to quantify changes from one frame to the next. With support of recursive Bayes estimation techniques, markers can be tracked throughout the provided test sequence and are able to compete with manual methods.


Efficient Visualization and Interaction with Fiber Structures using the Medical Imaging Interaction Toolkit.

Ignaz Reicht (Msc student)

Tractography is a technique providing information about the anatomical connectivity of the brain using Diffusion Weighted Magnetic Resonance Imaging (DWI). Hence fiber tracking shows high potential in the field of neuro science by facilitating the study of neuronal diseases such as Alzheimer’s Disease (AD) and Mild Cognitive Impairment (MCI), computer-based applications for interacting with tractography data are needed. This work provides fundamental methods for interactive exploration of data computed by DWI based tractography algorithms in the open-source platform "The Medical Imaging & Interaction Toolkit" (MITK). Efficient visualization and interaction techniques in 2D and 3D support the development of current features as desired by neuro scientists. The ease of use in accessing and dynamically interacting with tractography results empowers analyzing of the architecture of the human brain. Fiber structures of explicitly defined areas in the brain can be compared and evaluated. Adequate fiber representation supports radiologists and neuro surgeons in planning brain surgeries to avoid damage of major nerve fibers during the operation. With respect to scientists and dev


Semi-Automatic Annotation on Image Segmentation Hierarchies

Georg Zankl (Msc student)

Abstract:For the purpose of performing evaluations of image segmentation and annotation algorithms, a set of ground truth samples is necessary. The bigger this set, the more general the conclusion drawn from the evaluation. In a more specific context, such as hierarchical semantic labeling of segmentations, a suitable ground truth database may not be available such that a database of labelings based on a set of images has to be created. The goal of this thesis is to help with the process of building a ground truth for hierarchical image annotations by combining conditional random fields with user-interaction. Thus allowing a larger database to be created in the same time frame. 


Interactive Hierarchical Image Segmentation on Irregular Graph Pyramids

Michael Gerstmayer (Msc student)

Abstract: his talk is about the idea of integrating user-interaction into an hierarchical image segmentation framework. Without external knowledge, generic algorithms mostly produce unsatisfiable results. By manually placing actions on the image (based on border deletion or its inhibition) and working with graph-representations and operations on them, it will be possible to guide the former fully-automatic algorithm and produce any segmentation. This could be useful for example in classification, annotation or tracking applications where accuracy is important. The developed tool will be presented and an outline for the thesis given.

Keywords: interactive, hierarchical segmentation, graph, irregular image pyramid