Master theses

1. Simulering av MR-bildartefakter orsakade av inhomogeniteter i det statiska magnetfältet
   Student: Carl Krusell
   Supervisor: Jan Weis, Per Foyer
   Subject supervisor: Joakim Lindblad
   Publisher: CBA Master Thesis No. 100 / UPTEC F07 113
   Abstract: Magnetic Resonance Tomography (MRT) and spectroscopy are today one of the most important imaging techniques within the subject of radiology and is used daily in medical services. MRT is based upon theory of nuclear magnetic resonance and to map an object with good resolution and correct information a strong and solid magnetic field is required. Inhomogeneity in the static magnetic field B0 can cause serious false representations of image intensity and geometric positions of the mapped objects, so called image artefacts. Modern magnetic cameras are equipped with systems that can deliver a homogenous static magnetic field but also the object itself can impose image artefacts due to metallic implants, dental fillings and difference in magnetic susceptibility of different organic tissues. The goal with this thesis has been to develop user-friendly software that simulates image artefacts caused by inhomogeneity and to provide a quantitative comparison between simulated images and measured images. This report contains an introduction to the theory and technique of nuclear magnetic resonance and tomography. Here is a method presented that allows simulation of image artefacts given prior knowledge of the distribution of the magnetic field in homogeneity and a reference picture to apply the transformation to. Experimental data consists of measurements of a cylindrical container filled with water and organic tissue in the form of a human brain.

2. Finding and segmenting human faces
   Student: Qing Gu
   Supervisor: Bo Nordin
   Subject supervisor: Ewert Bengtsson
   Publisher: CBA Master Thesis No. 101 / IT 08 006
   Abstract: Human face and facial feature detection have attracted a lot of attention because of their wide applications, such as face recognition, face image database management and human-computer interaction. So it is of interest to develop a fast and robust algorithm to detect the human face and facial features. This paper is about a study of finding faces within images and segmenting the face into numbered regions which are the face-, mouth-, eyes- and hair regions respectively. In the last few years, many face detection methods have been proposed based on different specific conditions. The detection system in this master thesis project is implemented using color images with complex backgrounds under various lighting conditions. Each input image is dominated by the upper half of a single person. The algorithm presented in the paper uses a combined algorithm to detect the face. First, a skin color detection algorithm is applied to detect the four possible face regions. Second, the algorithm locates eyes within the candidate regions, and then the region with eyes becomes the face region. Finally, the algorithm locates the mouth and hair from the eyes and face regions.

3. Quantitative image analysis of Cherenkov light from nuclear fuel assemblies
   Student: Marcus Molander
   Supervisor: Ingela Nyström
   Subject supervisor: Ewert Bengtsson
   Publisher: CBA Master Thesis No. 102 / UPTEC F08 047
   Abstract: For the purpose of testing the Digital Cherenkov Viewing Device (DCVD) performance on detecting partial defects in spent nuclear fuel assemblies, digital images have been simulated. These images have been analysed in this thesis, to determine if and how well they agree with real images recorded by the DCVD. Methods for comparing the real and simulated images have been implemented in MATLAB. Both quantitative and qualitative methods for the analysis are tested and evaluated, as well as a method to identify possible fuel rod partial defects in an image. The agreement between the real and simulated images are overall good. However, the comparative methods have identified a number of differences between the images, which could be improved to make the simulated images appear more similar to the real DCVD images, most notably, the difference in sharpness. Positive results are also presented for the possibility to detect partial defects, especially in the real images. Finally, some suggestions of further analysis are presented.

4. Automatic liquid chromatography rack recognition
   Student: Johannes Löwén
   Supervisor: Ola Strandberg, Biotage AB, Uppsala
   Subject supervisor: Joakim Lindblad
   Publisher: CBA Master Thesis No. 103
   Abstract: In a liquid flash chromatography system, an instrument is loaded with sample, solvents and vials to hold the resulting fractions. The system is then programmed to separate the solution into its components. The result is collected in vials that are placed in racks. Racks and vials come in a multitude of sizes and dimensions and cannot be controlled by the vendor. Multiple racks may be loaded into the system. The purpose of the project is to investigate options for automatic recognition of racks and determining presence of vials within the racks. A prototype has been constructed to demonstrate how an imaging system can be integrated to the Isolera Flash Purification System. Images are acquired from a web camera. The image analysis main objective is to identify the elliptical contours of the glass vials. The contours are extracted by detecting edges and ridges. The ellipse radius and position determines the rack type. A method of modifying the Hough transform is suggested that takes into account the local image orientation to find the ellipse centres. Certain design constraints such as size and cost are discussed to sketch how the solution could be industrialized. The work has been carried out at Biotage AB and CBA in Uppsala.

5. Improving face recognition using textual annotation
   Student: Martin Norling
   Supervisor: Pablo Belin, PicSearch AB
   Subject supervisor: Ewert Bengtsson
   Publisher: CBA Master Thesis No. 104 / UPTEC X 08 035
   Abstract: In this Master thesis project, I have designed an application combining a traditional search engine and one of the most well known face recognition algorithms to date. The purpose of this design is to attempt to improve search relevancy for queries related to celebrities and other well known persons. The application includes both face detection and face recognition. The classical eigenfaces algorithm is used in the face recognition. This algorithm turns out to perform quite badly as the algorithm requires faces to be aligned to its references which is often not the case with arbitrary images of the internet. The face detection performs decently and can be used to separate face images from non-face images in simple cases.

6. 3DIS4U:
   Design and implementation of a distributed visualization system with a stereographic display
   Student: Martin Eriksson
   Supervisor: Anders Hast, Ingela Nystrom
   Subject supervisor: Ewert Bengtsson
   Publisher: CBA Master Thesis No. 105 / IT 08 041
   Abstract: Stereoscopic displays are used in research as an aid for visualizations, but often in a special room only to be used by a small selected audience. How should such a system be setup to make it available to a larger group of users? We solve this by setting up the system in a regular lecture room, a familiar environment, and by modifying software to make the transition from monoscopic to stereoscopic displays as smooth as possible. To improve the usability further, we connect the stereoscopic installation to a high-performance computing (HPC) cluster. As a result, we offer our users to distribute their visualizations and by that the ability to use larger datasets. There are two goals for this project: (1) to setup a stereoscopic display in a class room environment; (2) to enable distributed visualization at our graphics lab and evaluate further development in this field. The first goal is accomplished by setting up the hardware and thereafter focus on making the system more usable. Three different ways are presented, one by using the Visualization Toolkit (VTK), another by developing a small C++ library for converting existing visualizations to the stereoscopic display. The final option is non-invasive stereoscopic visualization with the Chromium library. The second goal is realized by installing and configuring ParaView, a visualization application for distributed visualizations on a cluster connected to the stereoscopic display. Exploration of alternative ways of performing visualization on the Graphics Processing Unit (GPU) is also concluded. The result of this master thesis is primarily a lecture room that in a matter of a few minutes is turned into a visualization studio with a stereoscopic display for up to 30 simultaneous viewers. The result is also an extended version of VTK for our stereoscopic display, a C++ library to help users to port their program for stereoscopic visualization and examples on how to use Chromium for non-invasive stereoscopic rendering. Furthermore, we make ParaView available to HPC users by installing and configuring it on one of UPPMAX clusters.