Graphics: Visual and Interactive Computing
Expressive and Non-Photorealistic Modelling and Rendering
Please also visit the Swansea volume graphics gallery. |
In recent years, the surge of applications in the entertainment and art world directed a substantial amount of research efforts towards expressive modelling and non-photorealistic (NPR) rendering. Most of these techniques were designed for surface-based graphics objects, and were integrated with traditional surface-based graphics pipelines. The past decade has witnessed significant advances in volume visualisation, which provides us with a new dimension for developing interesting artistic effects as well as an application area that can benefit from expressive and NPR modelling and rendering. The Swansea group has researched into a volume-based graphics pipeline with special reference to the place of expressive and NPR effects within its structure. Here we consciously confine the use of the adjective non-photorealistic (NPR) to the simulation of various hand painting effects, such as pen-and-ink drawing, and employ the adjective expressive to imply surreal features, such as object and viewing deformation, which may be rendered using either photorealistic or non-photorealistic methods. These effects are collectively referred to as artistic effects. At the higher level, a volume-based graphics pipeline is similar in form to its surface based analogue. In essence it is broken down into three main stages, namely modelling space, rendering space, and image space. Each of these stages facilitates the generation of different classes of artistic effects. This is determined by the amount and type of information available at each stage as well as the operations performed therein. In the modelling space expressive and NPR features can be introduced into volume models by altering existing features or by adding new features directly. We have developed a collection of filters in this space to manipulate various attributes of a volume objects. Most of filters are implemented in the form of control volumes which can be defined using mathematical, procedural or discrete specifications of scalar fields. Typical feature manipulations include distortion, scaling, making strokes and re-mapping (using look-up tables). These can be applied to the geometry, normal estimation, opacity, colour and other volume attributes. With artistic additions, expressive features such as strokes and masks can be incorporated into an existing volume object to form a composite object. One tool which has been found highly useful for the introduction of these effects is Constructive Volume Geometry (CVG). For examples, 3D strokes can be modelled using coloured scalar fields, and be used to replace the skin of a CT dataset using a CVG term involving a CVG intersection operator. The same set of strokes can also be used to create a sculpture using a CVG subtraction operator. The rendering space has the richest concentration of information that can be used for generation of various expressive and NPR effects. From the definition of the viewing system, through to the final rendering of the colour at each pixel in a synthesised image we have opportunities to introduce artistic effects by manipulating the viewing system, normals, opacity, colour, geometry, iso-surface, textures and hypertextures. Many interesting effects can be produced by utilising illumination conditions and altering the process of discrete volume sampling and opacity accumulation. It is also relatively easier to combine photorealistic and non-photorealistic effects in this space. In the image space there is a good collection of NPR algorithms for processing photographs and synthesised images in an artistic manner. Many of these algorithms rely almost entirely on the RGB information in an image itself, the resultant images often lack a hint of spatial cue, especially when dealing with images synthesised from 3D scenes. Not only can the rendering stage of our pipeline produce a synthesised image for the image space, it can also forward spatial information, such as depth, curvature, etc., in the form of images. With this spatial information, the synthesised image can be repainted with some artistic effects using a 2+D NPR rendering algorithm. For instance, in pen-and-ink painting, outline can be painted according to the relative depth information around a pixel, the orientation of strokes can be determined from curvature information. Main References
|
