Painterly Rendering for Animation

1. Painterly Rendering for Animation • Introduction speaks of focus and detail • Small brush strokes focus and provide detail • Large strokes are abstract and background • Addresses coherence in animation

2. Aims • Still frames should look like paintings • Details abstracted by shorthand brush strokes • Color breaks boundaries to create rhythm • Strokes should vary according to surface • Light should be exaggerated • Animation should maintain coherence • Images should not look like gift wrapped textures • Non-aim: Do not eliminate need for artistic vision but simply automate brush strokes

3. Basic Approach • Generate a set or particles that represent the surface • Depth sort in screen space and render front to back using painters algorithm • Stroke appearance: • Color, texture, orientation, size • Derived from information in a reference image or explicitly stored with particles

4. Generating Particles • Many possible methods, with different aims • Even coverage • More particles in interesting places • Their approach: • Break the surface into triangles • Determine number of particles based on area • Randomly distribute particles in triangle • Could generate entire set from particle system simulation

5. Brush Strokes • Image: Brush texture, with alpha • Color: Stored with particle or taken from reference image • Orientation: Stored with particle or taken from reference, shaded to encode surface normal • Size: Stored with particle or taken from reference image, but they don’t say how the reference image is created • Position: From particles position in screen space

6. Other Issues • Render separate layers and composite • Artistic direction • Frequently one layer per object, composite front to back • One light source, exaggerated colors to highlight light and shade • May or may not remove back-facing particles • If removed, fade in as they approach front facing • Some popping as depth sort order changes • Vague on fixes for this • Good: Handles large object motions well

7. Open Issues • Reducing model complexity (eg trees) • Doesn’t manage changing object sizes • Density of strokes will be wrong • Particle placement does not cover screen space equally • Fixing this may be hard for moving objects • Deformable brush strokes to follow edges

8. Real-Time Nonphotorealistic Rendering • Hardware methods for generating line drawings have problems: • Subject to aliasing • Don’t work for hand-drawn line styles (eg wavy lines) • Can’t easily extract visible segments • Paper presents fast methods for (probably) locating the important silhouettes

9. Finding Important Lines • Probabilistically finds some silhouette edges by randomly choosing edges and testing • Can improve by weighting edges by dihedral angle • Traces the edge graph to extend the silhouettes • For animations, trace from previous frame to find new silhouette edges • Use a modified Appel’s algorithm (1967) to determine visible edges • Appel’s algorithm walks along edges updating “quantitative invisibility”. QI=0 are visible

10. Rendering • Three approaches for rendering edges • Variations in line width or color, possibly depending on the depth of the segment (from QI) • Offset defined in tangent-normal basis for the curve • Textured stroke, obtained by widening each edge into a mesh and texture mapping • Surface shading using particle method • Even area distribution gives more particles near silhouettes • Place strokes along NormEye direction • Visibility from extension to Appel’s algorithm

11. Future Work • Was pretty much covered on Monday…