Information Technology Reference
In-Depth Information
In the sections below, the work on non-photorealistic rendering fits into the first
three stages of the meta-mountain ascent given above, while the work on emotional
modelling fits into stage 4. In the section on scene construction, we describe work
towards stage 5 above, and the work on collage generation has been done with stage
6 in mind. Finally, the work on paint dances fits best into stage 7 of the above meta-
mountain ascent.
1.4.1 Non-photorealistic Rendering
Starting with the notion of an artist simply making marks on a canvas, we imple-
mented abilities for the software to simulate natural media such as pens, pencils,
pastels, paints, brushes, papers and canvases. These tools allow the system to cre-
ate the basis of an artwork, for example, applying paint strokes on a canvas, or
making pencil marks on paper. To employ these simulations in useful ways, we im-
plemented standard machine vision techniques to enable the software to identify
regions of colour in a digital image, i.e., image segmentation. This led to a graphics
pipeline whereby an image is first segmented into a set of paint regions, and then
each region is filled and/or outlined a series of times in possibly differing styles.
To enhance this pipeline, we enabled layers of different segmentations to be de-
fined for possibly different areas of the original digital image (for instance, the user
could define a region of the image as containing a person's eyes, and specify that it
is segmented into more regions than the other areas, then painted differently). We
were careful to ensure that each stage of the pipeline can be user-controlled by a
fairly large number of parameters. For instance, image segmenting is controlled by
12 parameters, including: the number of segments required, the smallest segment
area allowed, the amount of abstraction of the segment regions, whether to allow
segments to have holes, etc. In addition, it is possible to map the colours in the seg-
mentation to a set of colours from another palette, for example, art deco colours.
The four different segmentations of a flower in Fig.
1.2
give some indication of the
range of segmentations possible via different parameterisations.
Image segmenting and the simulation of natural media are all standard non-
photorealistic rendering techniques, as described in textbooks such as (Strothotte
and Schlechtweg
2002
). We differed from the standard approach in one main re-
spect, namely that we didn't implement different methods for different media types.
For instance, the simulation of paints is usually treated differently to the simula-
tion of pencils or pastels, etc. Instead, we saw each media type as applying varying
amounts of pigment to a fixing medium such as paper or canvas. For instance, pencil
strokes could be seen as paint strokes carried out with a very thin brush and a less
than usual probability of the pigment sticking to the canvas (which gives the grainy
look required). As the individual strokes are only ever used to fill in colour regions,
we combined the parameterisation for the individual strokes with the parameteri-
sation of the way in which the strokes were employed. There are 45 parameters
controlling the way in which colour regions are rendered, and these include: aspects
