Game Development Reference
In-Depth Information
Coherence-Enhancing Filtering
on the GPU
Jan Eric Kyprianidis and Henry Kang
Directional features and flow-like structures are pleasant, harmonic, or at least
interesting to most humans [Weickert 99]. They are also a highly sought-after
property in many of the traditional art forms, such as painting and illustration.
Enhancing directional coherence in the image helps to clarify region boundaries
and features. As exemplified by Expressionism, it also helps to evoke mood or
ideas and even elicit emotional response from the viewer [Wikipedia 12]. Partic-
ular examples include the works of van Gogh and Munch, who have emphasized
these features in their paintings. This chapter presents an image and video ab-
straction technique that places emphasis on enhancing the directional coherence
of features. It builds upon the idea of combining diffusion with shock filtering for
image abstraction, but is, in a sense, contrary to that of [Kang and Lee 08], which
is outperformed in terms of speed, temporal coherence, and stability. Instead of
simplifying the shape of the image features, the aim is to preserve the shape by
using a curvature-preserving smoothing method that enhances coherence. More
specifically, smoothing is performed in the direction of the smallest change ( Fig-
ure 5.1(a) ), and sharpening in the orthogonal direction ( Figure 5.1(b) ). Instead
of modeling this process by a partial differential equation (PDE) and solving it,
approximations that operate as local filters in a neighborhood of a pixel are used.
Therefore, good abstraction results can be achieved in just a few iterations, mak-
ing it possible to process images and video at real-time rates on a GPU. It also
results in a much more stable algorithm that enables temporally-coherent video
processing. Compared to conventional abstraction approaches, such as [Win-
nemoller et al. 06, Kyprianidis et al. 10], the presented method provides a good
balance between the enhancement of directional features and the smoothing of
isotropic regions. As shown in Figure 5.2 , the technique preserves and enhances
Search Nedrilad ::

Custom Search