Špela Ivekovič

About my research...

My current research interests are in Novel-View Synthesis (of people), Multiple-View Geometry, Image-Based Rendering Techniques, Human Body Modelling and Pose Estimation/Tracking, Swarm Intelligence (PSO).

Statistical Models of Appearance

Statistical models of appearance (à la T. Cootes and C. Taylor) were the theme of my undergraduate research project under the guidance of Prof. Aleš Leonardis. I implemented two different multiresolution approaches to medical image segmentation with statistical models of appearance and evaluated their performance. The work was presented at Scale-Space 2003. abstract  paper

Immersive Videoconferencing and VIRTUE

The work on immersive videoconferencing in my current research group started with the Framework V IST VIRTUE Immersive Telepresence project. The VIRTUE project successfully implemented a videoconferencing station where the feeling of immersiveness was achieved in a completely non-intrusive way. The users were not required to wear any additional head-gear such as stereo glasses or head tracking devices. The crucial element for 3D immersivenes - the motion parallax - was achieved with the use of novel-view synthesis combined with gaze tracking.

Novel-View Synthesis

Novel-View Synthesis is an Image-Based Rendering technique which uses images taken from known viewpoints to generate new images as seen from previously unknown viewpoints. The information about the disparities between the existing views is used to correctly estimate the position of image pixels in the newly synthesised view. Disparity information coming out of traditional correspondence search algorithms is typically incomplete and critically influences the quality of the novel view synthesis. I investigated how much this could be improved by post-processing the available disparities with bicubic spline surfaces and completing the missing regions. abstract  paper

Disparity Maps

Disparity map is defined in the context of Stereo Vision. Given two images, left and right, taken from two different viewpoints, the disparity map is a 2-D array of vectors describing the displacement of every pixel in the right image with respect to the corresponding pixel in the left image (left-to-right disparity map) or vice-versa (right-to-left disparity map). Two pixels correspond if they are both an image of the same 3D scene point.

Subdivision Surfaces

Subdivision Surfaces are a popular modelling technique in Computer Graphics. Subdivision is defined as an infinite refinement process, a limit of which defines a smooth curve or surface. The popularity of the subdivision surfaces is due to the fact that they are fairly easy to implement, can be used to model surfaces of an arbitrary topological type, allow for local control and inherently support multiresolutional approach to surface representation. I have investigated modelling with subdivision surfaces for the purpose of noisy stereo data completion. abstract paper

Particle Swarm Optimisation

Particle Swarm Optimisation (PSO) is an evolutionary optimisation technique exploring the search space by drawing inspiration from fish schooling and bird flocking. As a global optimisation technique it is suitable for a variety of problems. I tested its performance on human body pose estimation from multi-view images. abstract   paper

Articulated 3-D Modelling in Disparity Space

When working with multi-view stereo data for the purpose of novel-view synthesis, the matching pairs of image points are conventionally reconstructed in 3-D and manipulated there, before being projected back into the image plane of the novel view. Stereo data can equally well be processed in disparity space, where it originates. By doing so, several computational steps are avoided. If care is taken to ensure that the homoscedastic nature of the disparity data is preserved, a potentially more accurate data manipulation is possible. I tested this idea on articulated upper human body models for the purpose of novel-view synthesis in videoconferencing-type scenarios. abstract   paper

Multiple-View Geometry

I co-authored a chapter on "Fundamentals of Multiple View Geometry" in 3D Videocommunication: Algorithms, concepts and real-time systems in human centred communication, Wiley 2005, edited by O. Schreer, P. Kauff and T. Sikora.