McGill University - School of Computer Science

Algorithms Seminar Summer 2001

Everybody is welcome!

Traditionally, computer graphics has focused on the realistic representation of an objective point of view, while the visual arts have been concerned with a personal expression of a subjective point of view. To bridge the divide, new algorithms for the artistic rendering of images need to be explored. In place of visual fidelity, the diversity of visually meaningful rendering styles is proposed as the measure of success for automated nonphotorealistic rendering techniques. The aim is to empower the artist to adapt the medium to suit the message.

The process starts with a template image which may be annotated to reflect the artist's intentions. The typical artistic image rendering framework first extracts the salient image features to drive a sampling algorithm. A mark, such as a brush stroke, is placed at each sample point and its size, orientation, shape, color, and texture are determined so as to locally reflect the template image while achieving the desired overall visual effect. Finally, the rendered image is created by compositing the individual marks.

Along with a review of artistic image rendering techniques, the seminar will present some early results of my own experiments. This method models the spatial relationships between neighboring sample points using a proximity graph, such as a Delaunay triangulation. Simple tiling, shading, and texturing procedures give rise to a wide variety of effects reminiscent of a stained glass panel, an inlay mosaic, a chalk sketch, or a sponge painting. The resulting multiresolution image representation is well suited to applications such as progressive image transmission and video transitions.