Probabilistic Color-by-Numbers: Suggesting Pattern Colorizations
jueves noviembre 14, 2013 - 07:30:00
Intel Headquarters, SC9 Auditorium
3601 Juliette Ln
Santa Clara,CA 95054
Suggesting Pattern Colorizations Using Factor Graphs
Sharon Lin, Ph.D. candidate, Stanford University
Daniel Ritchie, Ph.D. candidate, Stanford University
Matthew Fisher, Ph.D & PostDoc, Stanford University
Colored pattern images appear everywhere; they are used as background and header images in web design, as decorations in games, in clothing and upholstery, and in personal arts and crafts. Choosing colors for a pattern is one way to personalize and enliven the design composition. However, it can be difficult for a beginning artist or enthusiast to effectively explore the large space of possible colorings. In this talk, we look at automatically generating coloring suggestions to help the user through the creative coloring process.
We present a probabilistic factor graph model for automatically coloring patterns. The model is trained on example patterns to statistically capture their stylistic properties. It incorporates terms for enforcing both color compatibility and spatial arrangements of colors that are consistent with the training examples. Using Markov Chain Monte Carlo, the model can be sampled to generate a diverse set of new colorings for a target pattern. This general probabilistic framework also allows users to guide the generated suggestions via conditional inference or additional soft constraints. We demonstrate results on a variety of coloring tasks, and show a demo vector-art-editing application that incorporates the suggestion engine. Similar techniques may be useful for other domains of graphic art, such as typography and layout.
Sharon Lin is a Ph.D. candidate in Computer Science at Stanford University, advised by Pat Hanrahan. Her thesis work focuses on automatically assisting color choice for graphic art by learning from examples. Previously, she has worked on extracting representative color themes from images, and using image search to choose more semantically-meaningful colors for data visualization. She earned a B.S. in Computer Science at the University of Washington.
Daniel Ritchie is a Ph.D. candidate in Computer Science at Stanford University, where he is advised by Pat Hanrahan. His current research focuses on using probabilistic computation and probabilistic programming languages to generate art, build better design tools, and otherwise improve the creative process. Previously, he earned a B.A. in Computer Science at the University of California at Berkeley and an M.S. in Computer Science at Stanford.
Matthew Fisher received his B.S. in Computer Science at from Caltech, his Ph.D. from Stanford University under a Hertz Fellowship, and is currently a postdoc in Pat Hanrahan\'s lab. He has published on a variety of graphics topics including geometric design, tessellation for micropolygon rendering, and using probabilistic graphical models for example-driven design. His thesis develops approaches for understanding and automatically synthesizing 3D environments from examples.
Video is available here.