MSc Thesis defence by Chuang Wu
2D & 3D Segmentation with Similarity Invariant Shape Prior and Smoothness Term
Segmentation is often a required step in further processing and analysis of images and image volume.
When image data is complex, prior information on segmentation goals can be very important in guiding the process. We have as our goal the segmentation of mouse and rat brain from MR scan.
In this thesis, we combine two types of shape priors, a classical smoothness term and a geometric one, which incorporates the knowledge of one or more training shapes into a dissimilarity measure which is inherently invariant by scaling, rotation and translation. The new model combines smooth and convex non-smooth terms. The entire cost function is non-convex. We use several descent methods to optimize them. We illustrate their performance on synthetic images and a real MR scan.
Supervisor: François Lauze, DIKU
External examiner: Morten Pol Engell-Nørregård, Alexandra Institute