Master theses

1. Improved Path Opening by Preselection of Paths
   Student: Teo Asplund
   Supervisor: Cris Luengo
   Reviewer: Robin Strand
   Publisher: UPTEC IT 15014
   Abstract: Enhancing long, thin, sinuous structures in images is a common problem in image analysis. Mathematical morphology is often used to solve this problem. One approach is known as the path opening. The goal of this project was to investigate whether a preselection of a limited number of paths, based on the upper skeleton of the image, could be used to find an approximate, faster path opening. In this thesis, a new, graph-based algorithm, that is the result of this investigation, is presented. The new algorithm is compared with the traditional path opening and, to some extent, with the parsimonious path opening. Experiments suggest that the implemented algorithm is faster for increasing path length, and runs in linear time with respect to image size. They also suggest that the new algorithm is similar to the traditional path opening when measuring length distributions, while being orders of magnitude faster, thereby making it comparable in speed to the parsimonious path opening, while mitigating the problem of blind spots that the parsimonious path opening suffers from.

2. Real-Time Fish Type Recognition in Underwater Images for Sustainable Fishing
   Student: Fritjof Jonsson
   Supervisor: Vladimir Curic, Dept. of Cell and Molecular Biology, Computational and Systems Biology
   Reviewer: Ida-Maria Sintorn
   Publisher: UPTEC IT 14019
   Abstract: It has been investigated if it is possible to selectivly catch farmed salmon (Salmo salar L., 1758) and sea trout (Salmo trutta L., 1758) without disturbing the wild fish. A image analysis software that can separate wild from farmed salmon and salmon from sea trout has been developed. This is interesting since the advent of hydro power stations has obstructed the natural migration of these species to their natal river streams. Even though ladders have been built, fewer fish find their way back up stream. This has lead to farming of salmon and sea trout to compensate for a lower population. However, this is bad for the natural genetic variation and it would be desirable if only the wild fish could enter the rivers. The software could be installed in traps at fish ladders to help with this problem. It is common to cut the adipose fin from the farmed salmon and the lack of this fin has been used as a key character to separate farmed from wild salmon. A real-time algorithm was developed which could recognize the farmed fish with high accuracy by searching for presence or absence of the adipose fin. Additionally, two morphometric measurements were compared in order to investigate if it is possible to separate salmon from sea trout using image analysis. Preliminary tests show that it was possible to separate the species by looking at the ratio between the height of the caudal fin and the height of the caudal peduncle.

3. Procedural Modeling of Rocks
   Student: Anders Söderlund
   Supervisor: Stefan Seipel
   Reviewer: Anders Hast
   Publisher: UPTEC IT 15024
   Abstract: Gaming and other virtual environments are a big part of today's society, but manual modeling of terrains used in such environments can be a lengthy and tedious process. This thesis serves to explore a few methods of procedurally generating models of rocks or boulders that could be used in such contexts. This includes geometry and shading. A couple of different methods are explored. Sphere inflation, inspired by a classic sphere modeling method, involves "inflating" a base mesh (usually a platonic solid) to grow towards the boundaries of a sphere using an iterative subdivision approach, halting at a predetermined level of iteration. The second approach, recursive subdivision of segmented edges, involves dividing a base mesh into edge segments based on a predefined segment size, subdividing a polygon with pre-segmented edges with a recursive subdivision method based on the sphere inflation subdivision scheme. The segmented edges method is followed by a corner cutting step to "carve" the base mesh into a shape approaching a rock. The segmented edge method was not successfully finished within due time, but the sphere inflation method shows promise in generating fairly believable rock models. The shading includes GLSL based fragment and vertex shaders employing a Perlin noise based procedural granite 3D texture.
   
4. Calving Events Detection and Quantification from Time-lapse Images: A Case Study of Tunabreen Glacier, Svalbard
   Student: Sigit Adinugroho
   Supervisor: Dorothée Vallot, Dept. of Earth Sciences
   Reviewer: Robin Strand
   Publisher: UPTEC IT 15038
   Abstract: A fully automated method for detecting and measuring calving regions of a glacier is an important tool to gather massive statistical data of calving events. A new framework to achieve the goals is presented in this thesis. First, time-lapse images are registered to the first image in the set. Registration process makes use of the M-Estimator Sample Consensus (MSAC) method to estimate a transformation model that relates a pair of Speeded-Up Robust Features (SURF). Then, the terminus of a glacier is separated from other objects by a semi-automatic Chan-vese level-set segmentation. After that, calving regions in a terminus are detected as a combined difference of Local Binary Patterns (LBP) of two successive images. Clustered points that form a difference image are transformed into polygons representing changed regions by applying the -shape method. Finally, the areas of changed regions are estimated by the pixel scaling method. The results highlight the performance of the method under normal conditions and reveal the impact of various weather conditions to the performance of the method.

   
   

5. Interactive Methods for Procedural Texture Generation with Noise
   Student: Boris Kachscovsky
   Supervisor: Martin Strandgren, Goo Technologies, Stockholm
   Reviewer: Anders Hast
   Publisher: UPTEC IT 15045
   Abstract: Many computer graphics applications have used procedural noise since the 1980s, but there are still very few tools which allow non programming-oriented users to make procedural textures. This paper attempts to provide a framework for building and creating procedural noise based textures, in a way that can be easily abstracted and understood by those users. A careful study is conducted of Perlin Noise, and similar interfaces and tools are examined in order to create a framework centered around composable parts and semantic abstractions. The framework is then used to build a proof-of-concept interface which exemplifies some of the conclusions drawn from the study. The proof-of-concept interface successfully creates an environment which can be used to create procedural textures, and serves as a guide for future interfaces in the field.
   
6. Exploring Eye Tracking Techniques On Smartphones
   Student: Iosif Karkanis
   Supervisor: Edith Ngai, Division of Computer Systems, Dept. of IT, UU
   Reviewer: Anders Hast
   Publisher: UPTEC IT 15052
   Abstract: Eye tracking is a major field in medical sector, especially in psychiatry for giving an insight of the patients with mental disorders. Nowadays, the existence of mobile devices with powerful hardware can provide an opportunity to investigate if it is possible to track the eye using these devices without using any additional hardware. This thesis will try to explore the possibility of tracking the center of the pupil using mobile devices. To achieve that, the eye tracking algorithms of template matching and eye detection using image gradients, are implemented in these devices. The application was also implemented as a background service and as a stand-alone activity in order to investigate the performance and usability for these two methods. Both template matching and eye detection using image gradients algorithms show promising results in terms of performance and accuracy respectively.
   
7. Visual Representation by Triangulation
   Student: Max Pihlström
   Supervisor: Anders Hast
   Reviewer: Anders Brun
   Publisher: UPTEC IT 15061
   Abstract: In this thesis the triangulation is treated as a general-purpose visual representation by investigation of various domain-specific methods such as triangulation interpolation, mesh flows, vertex neighborhood feature measures and re-triangulation for spatial transformations. Suggested new methods include an effective cost for image interpolation based on work by Sederberg et al. and a ridge-edge measure related to the Harris edge detector.

8. Designing a User Interface for the Pico Image Processor
   Student: Staffan Edström
   Supervisor: Lars Oestricher, Division of Visual Information and Interaction
   Reviewer: Stefan Seipel
   Publisher: UPTEC IT 15078
   Abstract: The Pico image editor was developed at Bell Labs in 1984. The program includes a small pseudo-mathematical image transformation language. Although being well-designed, the functionality of the program does not meet modern standards. It only works on greyscale images and it runs in a command-line interface. To make it work on modern digital images the program must be extended with color support and additional features. The task also includes finding a solution for incorporating the functionality of the program and displaying the images in a modern graphical user interface where usability and effectiveness are prioritised. By using principles of image processing theory and graphical user interface design theory an extended version of the Pico program and a graphical user interface was created. It was built as a web browser application using HTML, CSS and JavaScript. The dynamic parts of the prototype program were built using the Nodejs framework. It has extended functionality combined with an easy to use interface. The program can be used for educational purposes to demonstrate the principles of image processing and as a creative tool for generating images and creating art.

9. Improving the Fast Evaluation of the Robust Stochastic Watershed
   Student: Abdul Rahman Khankan
   Supervisor: Cris Luengo
   Reviewer: Filip Malmberg
   Publisher: UPTEC IT 15085
   Abstract: The stochastic watersheds algorithm was first proposed by Angulo and Jeulin (2007) as a marker-controlled watershed-based stochastic segmentation method using Monte Carlo simulation. This project is based on the work of Selig et al. (2015), Fast Evaluation of the Robust Stochastic Watershed, which was the extension of Malmberg and Luengo Hendriks (2014) and Malmberg et al. (2014) who introduced an exact and efficient evaluation method of the stochastic watershed. The algorithm proposed running the exact evaluation method three times after adding noise to the input image then averaging the three edge probabilities together. Their method was identical in terms of average F-measure, but it was an order of magnitude shorter. This project aimed to propose an improved version of Selig et al.'s algorithm which is better in terms of accuracy and faster in terms of processing time. The final result was an algorithm that is matching in accuracy but about 25