Non Photorealistic Rendering


Non-photorealistic rendering from photographs often uses “filters” that change a small image patch into a brush stroke. But making brush strokes can never transform a photograph into art, because artists change the shape and size, the colour and the contrast of objects so that a viewer gaze is directed. This requires the artist (or computer) to visually understand the scene as a whole. In short we say
     drawing implies seeing

or its equivalent seeing is necessary for drawing. Notice that he skill of the artist or the aesthetic quality of the artwork is wholly irrelevant to this statement. One should regard artwork as salience maps that model the visual world.

Our research moves away from filters to use more sophisticated computer vision methods instead. These allow the changes of shape, perspective, distortions in time that is typically seen in real art. Moreover, reaching these goals forces new problems and contributions to computer vision as well as to computer graphics.

The list of projects below is in more-or-less chronological order, conical views, angiograms, and learning excepted. Not all of my work is represented here.

Generalised Cross Hatch Shading
This is early work using 3D models as input was amongst the first to produce artistic rather than photographic output.
Cubism from Photographs
Most “art from photo” software uses one picture, filters, and a lot of user interaction. The cubist project was the first to paint image regions, and user interaction is reduced to a few mouse clicks. It was also the first to make art using more than one photograph at once.
Artwork is unlike photographs in very many ways. One of the most important is that artists almost never use photographic like perspective (despite the Renaissance). Instead artist of all skill levels, from all over the world, and throughout history have drawn what they know. This project showed it is possible to create artistic perspectives using real cameras, and a bit of processing.RTcam.html
Video Paintbox
We solved the long-standing problem of painting over video without visually distracting “flicker”
that made cartoonified video look as if it was swimming about. We also introduced cartoon effects like squash and stretch, ghosting, and so on.Video_Paintbox.html
Shapes fit for purpose
Artists tend to sketch the overall shape of objects, filling in details later if they want to. This project is the first non-photorealistic system to use a shape classifier that automatically chooses the best overall shape of a region in an image. We used it to make art from paper cut-outs, but the potential for other sorts of rendering exists.Shapes.html
Child-like Art
Picasso - one of the most influential artists ever - wanted to draw like a child; this project shows how a computer can do exactly that

Salient Painting
Paintings emphasise objects that are of interest, when compared to other visible objects. This project investigated a simple way to define pixel salience.
An extension to RTcams that models colour cameras, as well as other non-geometric attributes. This allows for areal perspective, depth-of-field, colour aberration as well as star-burst lenses. Lenses may be drawn by users, and colour-corrected environment matting is possible. ARTcam.html