Non-Photorealistic Rendering from Stereo
Author(s): Alberto Bartesaghi, Guillermo Sapiro.
Article: Institute for Mathematics and its Applications (IMA), Vol. Preprint Series (1895), November,
2002.
[BibTeX]
Abstract:
A new paradigm for automatic non-photorealistic rendering
is introduced in this paper. Non-photorealistic rendering
(NPR) provides an alternative way to render complex
scenes by emphasizing high level or salient perceptual features.
Particularly, the pen-and-ink rendering style produces
sketchy-like drawings that can effectively communicate
shape and geometry. This is achieved by combining
drawing primitives that mimic ink patterns used by artists.
Existing NPR approaches can be categorized in two groups
depending on the type of input they use: image based and
object based. Image based NPR techniques use 2D images
to produce the renderings. Object based techniques work
directly on given 3D models and make use of the full volumetric
representation. In this paper we propose to enjoy
the best of both worlds developing an hybrid model that simultaneously
uses information from the image and object
domains. These two sources of information are provided
by a calibrated stereoscopic system. Given a pair of stereo
images and the calibration data we solve the stereo problem
in order to extract the normal and principal direction
fields, which are fundamental to guide a texture synthesis
algorithm that generates the NPR renderings. In particular,
normals guide tonal variations, while principal directions
determine the orientation of stroke-like texture patterns. We
describe a particular, fully automatic, implementation of
these ideas and present a number of examples.