Algorithms for appearance modeling
Simon Perreault
Patrick Hébert (Supervisor)
Approach: This project is divided into two independent parts.

The first proposes a new median filter algorithm, which is the corner stone of several image processing situations. However, its use has often been limited due to its algorithmic complexity of O(r) according to the window radius. As the resolution of images improves and their filter windows increase proportionally, the need for an efficient median filter algorithm has become more pressing. In this research project, a new simple, but much more rapid algorithm, with a complexity of time execution of O(1), is described and analyzed. Its performance is measured and compared to that of algorithms currently being used. Extensions towards data of higher dimensionality or precision, as well as a circular window approximation, are presented.

In the second part, a modeling system for capturing the appearance of objects is presented. This system consists in a cable robot moving a digital camera around an object while taking a large number of photos. The subsequent computer processing must achieve three main tasks: segmentation, 3D postioning of the camera and reconstruction of the visual hull of the object. In this research project, the algorithms involved are analyzed, criticized and improvements are proposed. Efficient algorithms for the segmentation of the object in the images and the computation of the visual hull are presented. Finally the complete system performance is assessed according to pre-established design objectives and recommendations are made for possible improvements.

Calendar: May 2006 - September 2007
Last modification: Oct 30 2007 10:45AM by perreaul


©2002-. Computer Vision and Systems Laboratory. All rights reserved