Segmentation of images: object-oriented toolbox


Projects > Segmentation of images: object-oriented toolbox

The analysis of an image usually begin with the construction of a model of the scene. Segments and intersection points are two primitives traditionally used in the development of the model. This report presents a simple method to obtain these primitives. The algorithm of Canny uses the information of the gradient to obtain the edges with the local maximums. The edges obtained are gathered in contours with the use of a double threshold, then are approximated by line segments. The points of intersection are calculated from these segments.

A set of C++ classes in which these algorithms are implemented is also presented with some results obtained on test images. The results obtained on the very simple scenes correspond perfectly to the real characteristics, but the results on complex scenes or scenes with illumination problems or scenes with noise are less interesting.

Finally, a new C++ class to read archived files images is presented. This tool is very simple of use and can decode several types of files without particular indication by the user.

More information: PDF.

This work was carried out in the Summer of 1999 at LVSN with the support of a NSERC undergraduate student research award.