Dana Cobzas

Assistant Professor, Computing Science
MacEwan University
Adjunct Assistant Professor, Computing Science
University of Alberta

e-mail: cobzas [at] ualberta.ca


home | research | publications | cv | me | contact

My main research interest is centered around imaging and computer vision, with particular interest in mathematical models for medical image processing. I have experience with methods for medical image segmentation, registration, noise reduction, as well as modern techniques for Diffusion Tensor Image processing.

Besides medical imaging, I work on 3-dimensional modeling and in particular the principles and algorithms for computing 3D representations from 2D images and video. Modeling scenes from images is an interdisciplinary subject that combines recent results in geometry, reflectance and light modeling.

Research topics

Medical imaging

Computer Vision


Reading group on medical image processing

Wed noon, CSC 349

Reading group in computer vision and robotics

Thu 1:30pm, CSC 349 (not active)

Reading group on variational and level set methods

Summer 2006


BIRS 2006 Workshop Mathematical Methods in Computer Vision - with Heyden, A. (Malmo) Jagersand, M. (Alberta) Little, J. (UBC) Sturm, P. (INRIA) Triggs, W. (INRIA) Zucker, S. (Yale)


IEEE VR 2003 Recent Methods for Image-Based Modeling and Rendering- with Darius Burschka, Zachary Dodds, Gregory D. Hager, Martin Jagersand, Keith Yerex

Deep learning for medical image segmentation

Iron in MS Quantitative image analysis for detecting iron accumulation related to multiple sclerosis

Sparse classification Detecting significant anatomy that discriminates two populations using sparse classification.

Tumor growth prediction Predicting tumor invasion margin using a geodesic distance defined on the Riemannian manifold of white fibers.

Brain tumor segmentation A variational segmentation method applied to a high-dimensional feature set computed from the MRI.

Semi-automatic segmentation software A semi-automatic software for medical image segmentation.
See also project description on Neil's webpage

Muscle and fat segmentation Automated segmentation of muscle and adipose tissue on CT images for human body composition analysis in cancer patients.

FEM for fast image registration A finite element implementation for the diffusion-based non-rigid registration algorithm.

RW for deformable image registration A discrete registration method based on the random walker algorithm.

A continuous formulation of Conditional Random Fields (CRFs) A variational formulation for a discriminative model in image segmentation.

Wavelet light basis A wavelet based light representation used for representing and capturing light.

On-line tracking and modeling On-line 3D reconstruction from tracked feature points.

3D SSD tracking A 3D consistent model is imposed to all tracked SSD regions, thus supporting direct tracking of camera position.

Variational Shape and Reflectance Estimation Variational method implemented as a PDE-driven surface evolution interleaved with reflectance estimation.
Lots of demos at Neil's webpage

Modeling with dynamic texture Inaccuracies in a coarse geometry obtained using structure-from-motion is compensated by an image-based correction - dynamic texture. A more complete description of the project

Image-based model for robot localization Panoramic image mosaic augmented with depth information applied to robot localization.

Range intensity registration A novel image-based algorithm for registering camera intesity data with range data from a laser rangefinder.