Thesis Defense Monday, 20 July at 10:30 am in Johnson 103: A Novel Image Segmentation Method based on Energy Minimization with an Application to Cardiac PET Image Segmentation

Open MSc thesis defense by M. Gong Cheng. All are welcome.

Image segmentation is a challenging research area in computer vision, particularly in positron emission tomography (PET) due to unpredictable object shapes and inconsistent image quality resulting from noise and sampling artifacts.

In this thesis, a novel image segmentation method has been developed. It is based on the combination of geometric interactive image segmentation, classification and Gaussian Mixture Models. It has been successfully applied to the mouse myocardium PET image segmentation and has been validated as an effective and accurate method. Unlike general interactive image segmentation methods, this hybrid segmentation method requires less input information from users which makes it more user-friendly and efficient. This method only requires users to provide a bounding box which encloses part of the target (foreground), and then it can generate an accurate segmentation result.

The first step of our project consists of using a classification based pre-processing method which allows obtaining an improved image that is the input to the segmentation procedure. Our segmentation method is in effect a Gaussian Mixture Model (GMM) based Graph Cut interactive image segmentation. The general idea is to create a GMM model for the input image provided by the pre-processing technique. The GMM consists of K Gaussian distribution components defined initially by using some clustering techniques. Consequently, each pixel is reallocated to a group defined by some mixture component according to its posterior probabilities. Then, an iterative Graph Cut based image segmentation is implemented to generate the segmentation results. More accurate segmentation results are generated by implementing a refining technique based on the level set based deformable models.

Moreover, this method is compared to another segmentation technique based on the HSI color model developed and implemented in this thesis.

Thesis Defense Friday, 17 July at 10:30 am in Johnson 103: Computation Tree Logic Is Equivalent to Failure Trace Testing

Open MSc thesis defense by A. F. M. Nokib Uddin. All are welcome.

The two major systems of formal verification are model checking and algebraic techniques such as model-based testing. Model checking is based on some form of temporal logic such as linear temporal logic (LTL) or computation tree logic (CTL). CTL in particular is capable of expressing most interesting properties of processes such as liveness and safety. Algebraic techniques are based on some operational semantics of processes (such as traces and failures) and its associated preorders. The most fine-grained, practical preorder is based on failure traces. The particular algebraic technique for formal verification based on failure traces is failure trace testing.

It was shown earlier that CTL and failure trace testing are equivalent; that is, for any failure trace test there exists a CTL formula equivalent to it, and the other way around. Both conversions are constructive and algorithmic. The proof of the conversion from failure trace tests to CTL formulae (and implicitly the associated algorithm) is however incorrect.

We now provide a correct proof for the existence of a conversion from failure trace tests to CTL formulae. We also offer intuitive support for our proof by providing worked examples related to the examples used earlier to support the conversion the other way around, thus going with our conversion but also our examples full circle. Our revised proof continues to be constructive and so the conversion continues to be algorithmic.

Our corrected proof completes an algorithmic solution that allows the free mix of logic and algebraic specifications for any system, thus offering increased flexibility and convenience in system specification for formal verification.

Hack Sherbrooke (24-26 April)

The second edition of Hack Sherbrooke was held from Friday, 24 April to Sunday, 26 April on our campus.

The event groups participants (this year 70 of them) from various post-secondary educational institutions in the region, as well as young professionals from IT companies. They are given less than 48 hrs (30 to be precise) to design and develop an App using the Open Source Data provided by the City of Sherbrooke.

This year 12 teams (4-6 members each) took part in the competition, grouping participants from Bishop's, UdeS, Cegep de Sherbrooke, Champlain, as well as one professional group. Four of these teams were composed of Bishop's students and alumni: Raphael Arseneau, Eric Boisvert, Russell Butler, Fabio Cozzuto, Armstrong Foundjem, Geoffrey Guest, Joshua Larouche, Tegan Maharaj, Mario Marchal-Goulet, Patrick Momo, Alec Neely, Andy Newball, Fidelis Njikem, Kofi Oduro, Jordan Slaman, Ashan Wickramasinha, and Shixiao Zhang.

The competition awarded 5 prizes… and Bishop's teams won 3 out of those 5 prizes!

  • First prize ($2500) went to BU students Tegan Maharaj and Jordan Slaman, accompanied by Geoffrey Guest (Bishop's CS 2013) and Kyle Gill (self-taught developer).
  • “Ville Intelligente” prize ($1000) went to same team as above.
  • The Jury Award ($500) went to BU students Fabio Cozzuto and Joshua Larouche, accompanied by Russell Butler (Bishop's CS 2014), Mario Marchal-Goulet (Bishop's CS 2014), and Carl Simard (Cegep de Sherbrooke).

Hack Sherbrooke was organized by SocieTIC, in collaboration with the City of Sherbrooke and the CS department at Bishop's.

Undergraduate Capstone Open Source Projects

If you are interested in getting real experience building a substantial software system as part of a distributed team, you'll be interested in UCOSP!

UCOSP is a senior undergraduate course, which has been running since September 2008. In this course, teams of students from several schools work together on an open-source software project. Each student registers in the appropriate course at his or her home institution (in our case this course would be either CS 404 or CS 408) and works in tandem with their peers from across the country. During one intensive weekend early in the course, students travel to meet face-to-face and work together.

This course exposes students to the tools, working practices, and issues that are now routine in global software development. Just as importantly, it enables them to get to know their peers from across the country. UCOSP is sponsored by Canadian CS Department Chairs and several industrial partners.

If you are interested in being involved, please contact Dr. Stefan D Bruda, who is the faculty partner for UCOSP at Bishop's.