next up previous contents
Next: Graduate education Up: Undergraduate education Previous: UU courses   Contents

Master theses

  1. Seamless Automatic Projector Calibration of Large Immersive Displays using Gray Code
    Student: Carl Andersson
    Supervisor: Mats Elfving, Sjöland & Thyselius AB, Stockholm
    Reviewer: Anders Hast
    Publisher: UPTEC F 13 032
    Abstract: Calibrating multiple projectors to create a distortion free environment is required in many fields e.g. simulators and the calibration may be done in a series different ways.

    This report will cover an automatic single camera projector calibration algorithm.The algorithm handles multiple projectors and can handle projectors covering bigger field of view than a camera by supporting image stitching. A proof of concept blending algorithm is also presented. The algorithm includes a new developed interpolation method building on spline surfaces and an orientation calculation algorithm that calculates the orientation difference between two camera views.

    Using the algorithm to calibrate, gives pixel accuracy of less than 1 camera pixel after interpolation and the relation between two views are calculated accurately. The images created using the algorithm is distortion free and close to seamless.

    The algorithm is limited to a controlled projector environment and calibrates the projectors for a single viewpoint. Furthermore, the camera needs to be calibrated positioned in the sweet spot although it can be arbitrary rotated.

  2. A trous Wavelet Transform and Trilateral Filter Algorithm forIimage De-noising
    Student: Niklas Borwell, Bo Leek
    Supervisor: Andreas Nilsson, Fredrik Olofsson
    Reviewer: Cris Luengo
    Publisher: UPTEC F 13 011
    Abstract: Confidential

  3. Volume Measurement of Wood Disks
    Student: Anders Dånmark
    Supervisor: Anders Brun
    Reviewer: Gunilla Borgefors
    Publisher: UPTEC F 13 046
    Abstract: At the Dept. of Forest Products at Swedish University of Agricultural Sciences different metrics for wood are used. The volume of wood disks' is measured using archimedes principle. There are concerns of how accurate this measurement is and a different measuringsystem is wanted.

    This thesis has investigated the possibility of measuring the disks' volumes with image analysis. The recovery error should be less than 1% of the actual volume. In general, there are two methods for recovering an object using image analysis, active and passive methods. Compairing active and passive methods, active methods usually require simple algorithms but more expensive equipment compared to passive methods. Different methods for measuring objects' volumes have been evaluated and the choosen method was ``shape from silhouette''. Shape from silhouette is a passive method, only using the silhouette of anobject from multiple views to recover the objects volume. Passive methods have one drawback, they can only recover the visual hull of an object and the wood disks can be slightly concave. Due to the questionable accuracy of the current measurement method it was still deemed as possible to achieve at least equal performance.

    When the volume measuring algorithm was developed it was first tested in two simulations using on a sphere to determine its performance with different voxel sizes and different number of images. The algorithm performed well and an error of less than 1 % was achieved with a sphere. A third simulation was performed using a simulated wood disk, which is a much more complex object, and 5 % accuracy was achieved. Finally, an experiment on real images was performed. This experiment did, however, fail due to the low quality imaging setup.

    The conclusion of this thesis is that itis not possible to achieve less than 1 % accuracy of the recovered volume using the shape from silhouette technique.

  4. Object Recognition Using Digitally Generated Images as Training Data
    Student: Anton Ericson
    Supervisor: Stefan Seipel
    Reviewer: Anders Hast
    Publisher: UPTEC F 13 010
    Abstract: Object recognition is a much studied computer vision problem, where the task is to find a given object in an image. This Master Thesis aims at doing a MATLAB implementation of an object recognition algorithm that finds three kinds of objects in images: electrical outlets, light switches and wall mounted air-conditioning controls. Visually, these three objects are quite similar and the aim is to be able to locate these objects in an image, as well as being able to distinguish them from one another. The object recognition was accomplished using Histogram of Oriented Gradients (HOG). During the training phase, the program was trained with images of the objects to be located, as well as reference images which did not contain the objects. A Support Vector Machine (SVM) was used in the classification phase. The performance was measured for two different setups, one where the training data consisted of photos and one where the training data consisted of digitally generated images created using a 3D modeling software, in addition to the photos. The results show that using digitally generated images as training images didn't improve the accuracy in this case. The reason for this is probably that there is too little intraclass variability in the gradients in digitally generated images, they're too synthetic in a sense, which makes them poor at reflecting reality for this specific approach. The result might have been different if a higher number of digitally generated images had been used.

  5. An iPad-based Drawing Processor
    Students: David Eriksson, Kristian Ionescu
    Supervisor: Måns Ridzén, LeanStruct AB, Uppsala
    Reviewer: Anders Hast
    Publisher: UPTEC IT 13 032
    Abstract: In today's society, it becomes more common with tablets, that makes us interact with computers in a whole new way. For example these are used to read and write e-mails, surf the web and playing games. Another manner of use of these tablets is to read and edit PDF documents. PDF handling is often meant to be used in books and inregular text documents, but it could also be used in the management of drawings.

    An industry that would benefit greatly from the use of tablets for this purpose is the con- struction industry. By creating an application that not only serves as a standard PDF reader, but also can handle drawings by making annotations, synchronize them toa cloud service and mail these drawings to others. This would make the management of drawings more effective and this would also revolutionize this industry.

    This thesis presents the planning, implementation, results and also the challenges that were faced during the development of such a prototype. By using object-oriented analysis and design principles, extensive use of test cases and implementation in Objective-C interesting results have emerged.

    This report mainly turns to readers with interest or has a background in computer science.
    Comment: Bachelor Thesis in Swedish. Title: En iPad-baserad ritningsbehandlare.

  6. Automatic Detection of Honeybees in a Hive
    Student: Mihai Iulian Florea
    Supervisor: Cris Luengo
    Reviewer: Anders Brun
    Publisher: UPTEC IT 13 060
    Abstract: The complex social structure of the honey bee hive has been the subject of inquiry since the dawn of science. Studying bee interaction patterns could not only advance sociology but find applications in epidemiology as well. Data on bee society remains scarce to this day as no study has managed to comprehensively catalogue all interactions among bees within a single hive. This work aims at developing methodologies for fully automatic tracking of bees and their interactions in infrared video footage.

    H.264 video encoding was investigated as a means of reducing digital video storage requirements. It has been shown that two orders of magnitude compression ratios are attainable while preserving almost all information relevant to tracking.

    The video images contained bees with custom tags mounted on their thoraxes walking on a hive frame. The hive cells have strong features that impede bee detection. Various means of background removal were studied, with the median overone hour found to be the most effective for both bee limb and tag detection. K-means clustering of local textures shows promise as an edge filtering stage for limb detection.

    Several tag detection systems were tested: a Laplacian of Gaussian local maxima based system, the same improved with either support vector machines or multilayer perceptrons, and the Viola-Jones object detection framework. In particular, this work includes a comprehensive description of the Viola-Jones boosted cascade with a level of detail not currently found in literature. The Viola-Jones system proved to outperform all others in terms of accuracy. All systems have been found to run inreal-time on year 2013 consumer grade computing hardware. A two orders of magnitude file size reduction was not found to noticeably reduce the accuracy of any tested system.

  7. Detecting Background and Foreground from Video in Real-Time with a Moving Camera
    Student: Jesper Friberg
    Supervisor: Simon Mika, Imint AB
    Reviewer: Cris Luengo
    Publisher: UPTEC IT 13 009
    Abstract: Finding the true movement in video taken by a moving camera is a complex problem, an even more complex problem accrue when this also is to be done in real time and on a low performance computer. Simple algorithms for static camera movement detection was implemented and then improved to cope with moving cameras. Results show that finding movement within a moving image at real time can be done with reasonable outcome and that post-processing can improve the quality of that outcome. This makes it able to detecting movement from moving cameras at real time on rugged laptops, controlling for instance an unmanned aircraft vehicle.

  8. Algorithms for Representation of 3D Regions in Radiotherapy Planning Software
    Student: Jonny Gunnarsson
    Supervisor: Anders Edin, Elekta Instrument AB, Uppsala
    Reviewer: Carolina Wählby
    Publisher: UPTEC IT 13 005
    Abstract: This thesis reviews the fast marching method as a technique for computing the distance transform on GPU in the context of a radiotherapy planning software. The method has some interesting characteristics that, given the right circumstances, allow the distance transform to be computed for fewer voxels than commonly used alternatives. This can result in beneficial effects both with regards to memory consumption and computation speed. A prototype is implemented to evaluate the features of the fast marching method including its suitability for execution on GPU. The implementation uses NVidia's Thrust library in order to assess it as a means of achieving performance portability, i.e. producing code that can be efficiently executed both on GPU and CPU.

    The fast marching method is evaluated based on speed, memory consumption and accuracy. These measurements are compared to an existing method for computing the distance transform in order to put the results into context. The assessment of the Thrust library is based on the experience of implementing the prototype. It is analyzed with regards to aspects such as the perceived ease of implementing the algorithm and the efficiency of the resulting solution.

    The conclusion of this thesis is that the fast marching method may well be a suitable approach for computing the distance transform on GPU. This is based on results in best case scenarios showing twice as fast computation speeds while only using a tenth of the memory compared to the chosen benchmark method. With regards to the Thrust library, however, this thesis concludes that it is not suitable for the implementation of an algorithm of this complexity. The impression is that thedevelopment of the prototype has been severely hampered by the use of Thrust and the performance of the resulting code is poor. This is demonstrated by a part of the prototype being re-implemented using CUDA resulting in a speedup for that part of between five and thirty times, depending on the scenario.

  9. Automatic Segmentation of Skeleton in Whole-Body MR Images
    Student: Anders Hedström
    Supervisor: Robin Strand
    Reviewer: Joel Kullberg, Dept. of Radiology, Oncology and Radiation Sciences, UU
    Publisher: UPTEC IT 13 011
    Abstract: Magnetic Resonance Imaging(MRI) has developed as a widespread technique to examine various body parts and diagnose a wide range of diseases. MRI can often be superior to other imaging techniques such as Computed Tomography(CT) since it does not use ionizing radiation and can give a clearer image of soft tissue. As MRI becomes a more important part in medicine the demands on software to analyse the images and extract useful information increases. Today medical image analysis can be used to localise tumours, measure brain substance and to isolate specific organs. Although much has happened in the field in recent years there is still little published about segmentation of skeleton in MRI images, this might be because cortical bonen either contains fat nor water and thus gives a weak signal in MRI images. Skeletal segmentation could still be useful to localise other body parts, to guide further analysis of whole body images and to do attenuation correction in PET/MRI systems. This work aims to increase the knowledge about skeletal segmentation in fat and water(FWI) MR images, and the goal is to produce a method that is flexible and robust enough to work on different MR machines with patients of various body types. This work implemented and evaluated two methods for skeletal segmentation in fat and water MR images. The first method divided the body into different regions and segmented each region with a region-specific algorithm and the other method consisted of a filter that detect patterns in the proximity of bone.The evaluation used reference segmentations performed with the program SmartPaint, and overlap with the automatic method was measured. Subjects used in this work originated from two studies, one on small patients and one on larger patients, thus giving an indication of how well the methods work on a population with large variance. Results show that the filter method produce a more accurate result than the body division method. The body division method had an average dice coefficient of 0.836, over segmentation ratio of 0.225 and under segmentation ratio of 0.120. The filter method had a dice coefficient of 0.944 and over and under segmentation rates were both 0.055. Both methods needed post processing in order to get a result that minimised the over segmentation in order to achieve an acceptable result. Neither of the methods allows accurate assessment of bone volume, but an approximation might be possible with the filter method. This project has shown that it is possible to segment skeleton in whole body MRimages with a decent result without using either registration or deformable models. More advanced methods will most likely be needed to minimise the over segmentation and increase segmentation accuracy.

  10. A Starting Point for Constructing a Digital City Map for the Visually Impaired
    Student: Alexandra Helin
    Supervisor: Lars Oestricher
    Reviewer: Stefan Seipel
    Publisher: UPTEC IT 13 004
    Abstract: The physical map that has been used until now can in the modern days be replaced with digital maps available in smart phones and tablets. One disadvantage of both is that the digital maps require the user to have a fully functional sight. The problem with the research done so far is that it provides little explanation as to why developers have decided to design the maps in a specific way. This thesis has been designed to address this problem.

    In order to provide the knowledge needed, a literature study was done to construct interview questions to an employee from SG Access AB. These answers and the literature study were used to construct interview questions to members and employees from Synskadades Riksförbund (SRF). A method inspired by Cultural probes was done to improve these questions. The literature study and the answers from the interviews were then used to answer the five domain questions.

    This thesis managed to answer four of five questions, and provides the basic knowledge needed to develop a tactile city map.
    Comment: In Swedish, title: Första avstampet för att konstruera en digital stadskarta för personer med nedsatt syn

  11. Web-based Sprite Sheet Animator for Sharing Programmatically Usable Animation Metadata
    Student: Xinze Lin
    Supervisor: Anders Hast
    Reviewer: Lars Oestricher
    Publisher: UPTEC IT 13 024
    Abstract: In this project, we have developed a prototype application which is capable of creating and sharing programmatically usable sprite sheets via the web. At the same time, we also proposed a technique called Meta-pixel Enhanced Sprite Sheet which can enforce 2D game animation metadata to be always attached with its corresponding sprite sheet image. The project is dedicated to help 2D game programmers to quickly obtain programmatically usable raster graphics.

  12. Texture Feature Analysis of Breast Lesions in Automated 3D Breast Ultrasound
    Student: Haixia Liu
    Supervisors: Tao Tan, Bram Platel, and Nico Karssemeijer, Radboud University, Nijmegen, The Netherlands
    Reviewer: Ewert Bengtsson
    Publisher: UPTEC IT 13 052
    Abstract: This thesis investigated a variety of texture features performances on classifying and detecting breast lesions in automated 3D breast ultrasound (ABUS) images with computer-aided diagnosis and detection system. Regions detected by the computer-aided detection system could be categorized into benign and malignant classes, which are supposed to have different texture features.

    After normalization and segmentation on the original 3D ultrasound breast images automatically, we implemented four texture feature extraction algorithms on the detected targets. The proposed four algorithms are based on 3-dimensional gray level co-occurrence matrix (3-D GLCM), local binary pattern (LBP), Haar-Like and regional zernike moment (RZM) separately. Three major experiments were carried out on a set of ABUS images. In experiment one, we focused on distinguishing malignant lesions (165 samples) from benign lesions (258 samples). In experiment two, we added a number of normal cases (150 samples) to the dataset, by grouping them with benign lesions against malignant lesions and by isolating them from benign and malignant lesions. In experiment three, we tested texture features ability on reducing false positives in the existing computer-aided detection system. In this step, only normal cases (5263 samples) and malignant lesions (165 samples) were examined.

    To estimate the discrimination power of different texture features, Support VectorMachine (SVM) and AdaBoost classifiers were adopted in corporation withleave-one-patient-out and 10-fold cross validation schemes respectively. The areaunder the receiver operator characteristic (ROC) curve (AUC, also known as Az)values were analyzed corresponding to each texture feature extraction method. TheAz values computed in experiment one are compared as follows: Haar-Like features performance outweighs others with the Az value of 0.86, followed by LBP (0.84),RZM(0.81) and 3-D GLCM (0.75). With respect to the results from experiment two,the Az value of grouping normal cases with benign lesions against malignant lesions isbetter than separating them from benign and malignant lesions, in general. Regardingthe outcome from experiment three, the Az value was increased from 0.79 to 0.82after adding LBP and Haralick features to the existing computer-aided detectionsystem.

    Based on the overall results, we concluded that texture features are useful on classifying benign and malignant lesions in ABUS images and they can improve the performance of the existing computer-aided detection system on detecting breast cancers.

  13. Investigating Multi Instance Classifiers for improved virus classification in TEM images
    Student: Sujan Kishor Nath
    Supervisor: Gustaf Kylberg
    Reviewer: Ida-Maria Sintorn
    Publisher: UPTEC IT 13 084
    Abstract: CBA together with the industrial partners Vironova AB (Stockholm) and Delong Instruments (Czech Republic) have a joint research project with the goal of developing a table-top TEM with incorporated software for automatic detection and identification of viruses. A method for segmenting potential virus particles in the images has been developed as has various measures of characteristic features, mainly based on texture, for distinguishing between different virus types. Different virus species generally have different sizes and shapes but their width (diameter if approximately spherical) is a rather conserved feature as is the protein structure on their surface (seen as texture patterns in the images).

    In the project they currently focus on using different texture measures calculated on a disk centered within an object for classifying the virus species. Extracted feature measures calculated for one position for (at least) 100 objects of 15 different classes of viruses exist for use in this project. The aim of this thesis is to investigate if/how feature vectors calculated in multiple positions can be used to improve the classification. Since the viruses have very different shapes, from approximately spherical to highly pleomorphic (like boiled spaghetti), the number of possible positions for extracting feature vector will be different for different virus objects. Another goal is to investigate how the distribution of measures calculated on small patches within the disk shaped feature area can be used in the classification, rather than combining them into one measure as is currently done.

  14. The Triangulation as an Alternative Painting Medium
    Student: Max Pihlström
    Supervisor: Anders Hast
    Reviewer: Anders Hast
    Publisher: UPTEC IT 13 057
    Abstract: In as much as raster and vector graphics have complementary roles in digital imagery they both have limitations. In this paper, the two frameworks are in part bridged in the triangulation mesh where in particular the ideas of the spatial neighborhood and representation by geometrical primitives are combined. With a triangulation algorithm for preserving integrity of contour and color together with methods for introducing geometric detail and blending color, the end result is a configurable medium with qualities resembling those of physical paint, demonstrating potential as a viable alternative for graphics creation.
    Comment: Bachelor Thesis

  15. Rotation Invariant Registration of 2D Aerial Images Using Local Phase Correlation
    Student: Lu Ping
    Supervisor: Anders Hast
    Reviewer: Stefan Seipel
    Publisher: UPTEC IT 13 030
    Abstract: Aerial image registration requires a high degree of precision. In order to improve the accuracy of feature-based registration, this project proposes a novel Log-Polar Transform (LPT) based image registration. Instead of using the whole image in the conventional method, feature points are used in this project, which reduces the computational time. For rotation invariance, it is not important how the image patch is rotated. The key is focusing on the feature points. So a circular image patch is used in this project, instead of using square image patches as used in previous methods. Existing techniques for registration with Fast Fourier Transform (FFT) always do FFT first and then Log-Polar Transformation (LPT), but it is not suitable in this project. This project does LPT first and then the FFT.

    The proposed process of this project contains four steps. First, feature points are selected in both the reference image and the sensed image with corner detector (Harris or SIFT). Secondly, image patches are created using feature point positions as centers. Each point is a center point of LPT, so circular image patches are cropped choosing a feature point as center. The radius of the circle can be changed. Then the circular images are transformed to Log-Polar coordinates. Next, the LPT images are dealt with using phase correlation. Experimental results demonstrate the reliability and rotation invariance of the proposed method.

  16. Tracking Individual Bees in a Beehive
    Student: Zi Quan Yu
    Supervisor: Cris Luengo
    Reviewer: Ida-Maria Sintorn
    Publisher: UPTEC IT 13 009
    Abstract: Studying and analyzing interactions among bees requires tracking and identifying each individual among hundreds of them on a complex background. Automatic tracking and identification is challenging because of the unreliable features and appearance changes. In order to map bee's social interactions, low computational cost algorithm needs to run for a long time and process has to be done at the same time.

    We present comparison among several methods and how we stabilize the features and reduce the appearance changes. We have improved much in set-ups and made a newly designed tag. Meanwhile we have developed the prototype of this automatic algorithm to track and identify each individual bee among hundreds of bees in a beehive over time. The rate is 15 frame per second at this stage and for the global detector it takes around 21s to process one frame and for the local detector it takes around 11s to process one frame. The algorithm can correctly detect 89% of around 300 tagged bees over hundreds of frames on average, but there are still around 11% misdetections.

  17. Object Recognition Using the OpenCV Haar Cascade-Classifier on the iOS Platform
    Student: Staffan Reinius
    Supervisor: Amen Hamdan, BMW, Shanghai, China
    Reviewer: Anders Hast
    Publisher: UPTEC IT 13 007
    Abstract: Augmented reality (AR), the compiling of layered computer-generated information to real-time stream data, has recently become a buzzword in the mobile application communities, as real-time vision computing has become more and more feasible. Hardware advances have allowed numerous such utility and game applications to be deployed to mobile devices. This report presents a high-level implementation of live object recognition of automobile interiors, using Open Source Computer Vision Library (OpenCV) on the iOS platform. Two mobile devices where used for image processing: an iPhone 3GS and an iPhone 4. A handful of key-feature matching technics and one supervised learning classification approach were considered for this implementation. Speeded Up Robust Features (SURF) detection (a key-feature matching technique) and Haar classification (supervised learning approach) were implemented, and Haar classification was used in the final AR prototype. Although the object classifiers are not yet to satisfaction in terms of accuracy, a problem that could be overcome by more extensive training, the implementation performs sufficiently in terms of speed for the purpose of this AR prototype.
    Comment: Bachelor Thesis

  18. Parameter Comparison of Non-Rigid Registration of Whole-Body MR-Images by Multiple Evaluation Methods
    Student: Lei Wang
    Supervisor: Robin Strand
    Reviewer: Joel Kullberg, Dept. of Radiology, Oncology and Radiation Sciences, UU
    Abstract: Confidential

next up previous contents
Next: Graduate education Up: Undergraduate education Previous: UU courses   Contents