A large part of computer graphics is focused on non-photorealistic rendering (NPR). Unlike photorealistic rendering, in this case the aim is not to reproduce realism, but to mimic artistic styles like drawing or painting. Even though NPR often favors rendering quality over rendering time, this does not mean that rendering speed is negligible. Production costs can be greatly decreased with a faster algorithm, especially when the composition requires tuning of the NPR result. The 3D graphics pipeline is really efficient compared to manual solutions, but it is difficult to reproduce classic techniques with it. The greatest issue regarding image space solutions is to ensure time coherency during animation. This means, that the features and particles used for illustration, are following the geometry.
In this thesis an algorithm for line-based techniques (drawing and painting) is introduced. The lines are defined in image space and they are distributed evenly. The line particles are describing the underlying geometry with their direction and bending, they use image space data to achieve this like luminance and curvature. Time coherency is ensured by moving the particles. This will mess up the screen space distribution, so we filter out the line particles which are too dense, and fill the empty areas with new particles. To aid fluid motion, the line primitives' states are animated, there can be no sudden direction changes or disappearances this way. The algorithm is implemented on the GPU and it runs with real time speed.