Loose and Sketchy Animation

In narrative character animation, engaging the viewer and conveying emotion and personality are of the highest importance. Early in the development of the medium, animators learned that an effective way to achieve these goals was to "present a unified single idea with nothing complicated, extraneous, or contradictory in its makeup."[1] But this principle is seldom applied successfully in computer animation because of the demands of photorealistic rendering.

Photorealism, like pornography, leaves nothing to the imagination. It presents the viewer with a world of objects complete with volume and texture, which is far more information than the viewer needs to get the point. Furthermore, unless great effort is devoted to every detail of modeling, shading, and lighting, much of that information will actually contradict the central idea, distracting the viewer. The need to cancel out these spurious impressions places an unreasonable demand on the animator.

A hand-drawn sketch can convey a lot of information in a much simpler package, if one draws only what is necessary. The gesture of line can also convey the energy of the artist's hand, or a character's mental state. And because a sketch is clearly an artist's interpretation of the world, it encourages the viewer to complete the picture by imagining the details that are missing. This engages the viewer's mind in a way that a photorealistic image cannot.

A filter for converting a 3-D animation into a sequence of gestural sketches promises to be a powerful tool for motion testing, or for creating finished animations in a non-photorealistic style. This sketch describes one approach to implementing such a filter.

The "loose and sketchy" filter automatically draws the visible silhouette edges of a 3-D model using image processing and a stochastic, physically-based particle system. For input, it requires only a depth map of the model (fig. 1) and a few simple parameters set by the user. First, the depth map is converted into two images:

• The "template" image (fig. 2), in which each pixel represents the amount of ink needed in its immediate neighborhood. This image is obtained by calculating the magnitude of the gradient of the depth map, thresholding to give binary values, and then blurring the result.

• The force field (fig. 3), a vector field which pushes particles along the silhouette edges. Such a field can be obtained by calculating unit vectors perpendicular to the depth map's gradient.

1. A depth map.	2. The template image.	3. The force field.

Next, particles are generated, one at a time, for a fixed number of particles. Each particle's initial position is chosen at random from within the template image, with a bias toward areas that need more ink. (No particles are ever born in areas that need no ink.) Acceleration at each timestep is based on the force field, with additional coefficients for randomness and drag. The particle is rendered onto the canvas as an antialiased line segment. If a particle wanders into an area that needs no more ink, it dies and a new one is born in another random place. The particle also erases the dark pixels from the template image as it travels, so that those edges will not be drawn again. (This aspect of the technique is similar to previous approaches to pen-and-ink rendering [2, 3].)

By tweaking drag and randomness, a continuous gamut of styles can be generated ranging from a tightly controlled technical drawing (fig. 4) to a loose and gestural sketch (fig. 6). The higher the drag, the tighter the style. These parameters also affect the character of motion, which can range from near-perfect smoothness to a lively staccato. The looser styles are particularly appropriate for animated characters, giving them a bit of life even when they stand perfectly still. It is also worth noting that for animation, it is not necessary to exactly render all of the silhouette edges all of the time. An error in a single frame will usually go unnoticed, because each frame lasts only a fraction of a second.

4. High drag, zero randomness.	5. Moderate drag and randomness.	6. Low drag and high randomness.

Figures 7-11 show just a few of the other styles available by varying these and other parameters. Each of these images took only 10-60 seconds to compute.

7.	8.	9.

10.	11.

[1] Frank Thomas and Ollie Johnston, "Disney Animation: The Illusion of Life", Abbeville Press, New York, 1981.

[2] Michael P. Salisbury, Michael T. Wong, John F. Hughes and David H. Salesin, "Orientable Textures for Image-Based Pen-and-Ink Illustration", Proceedings of SIGGRAPH 1997.

[3] Greg Turk and David Banks, "Image-Guided Streamline Placement", Proceedings of SIGGRAPH 1996.

Cassidy Curtis 8:14 pm, 15 June 1998