CUSTOMIZING PAINTERLY RENDERING STYLES USING STROKE PROCESSES - - PowerPoint PPT Presentation

Mingtian Zhao, Song-Chun Zhu University of California, Los Angeles Lotus Hill Institute http://bit.ly/stroke-processes

CUSTOMIZING PAINTERLY RENDERING STYLES USING STROKE PROCESSES

Stroke-Based Painterly Rendering

• Brush Modeling
• Mostly Objective
• Existing Methods: Procedural, Example-Based
• [Strassmann 86; Cockshott et al. 92; Hertzmann 98, 02; Baxter et al. 01, 04]
• Stroke Placement
• Highly Subjective: Styles, Feelings, etc.
• Existing Methods: Greedy, Global Energy Optimization
• [Hertzmann 98, 01; Collomosse et al. 02; Hays and Essa 04; Zeng et al. 09]
• We study the latter in this paper.

The Problem: Language Gap

• Artists’ Language
• Vibrant Colors
• Gestural Strokes
• Sense of Illumination/Motion
• Computer Scientists’ Language
• Color Vector (RGB, between 0 and 255)
• Stroke Length/Width (in pixels)
• Difficult to map artists’ descriptions to rendering parameters

Characteristics

• Observation
• Local contrast is important (the “tempo”)
• The Bridge: Spatial Statistics
• Forestry and Plant Ecology
• Epidemiology
• Seismology
• …
• Our Approach: Stroke Processes
• Marked Point Process for Stroke Layout
• Reaction-Diffusion for Stroke Attributes

Stroke Processes

• Stroke Element
• Position
• Orientation
• Size
• Color
• Stroke Neighborhood Graph
• Second-Order Features
• Computing Tools
• Perceptual Characteristics/Dimensions
• Quantitative Evaluations
• Rendering Parameters

Perceptual Dimensions

• Density

Low Density High Density

Perceptual Dimensions

• Density
• Non-Uniformity

Low Non-Uniformity High Non-Uniformity

Perceptual Dimensions

• Density
• Non-Uniformity
• Local Isotropy

Low Local Isotropy High Local Isotropy Histograms of Orientation Differences

Perceptual Dimensions

• Density
• Non-Uniformity
• Local Isotropy
• Coarseness

Low Coarseness High Coarseness

Perceptual Dimensions

• Density
• Non-Uniformity
• Local Isotropy
• Coarseness
• Size Contrast

Low Size Contrast High Size Contrast Histograms of Size Differences

Perceptual Dimensions

• Density
• Non-Uniformity
• Local Isotropy
• Coarseness
• Size Contrast
• Lightness Contrast
• Chroma Contrast
• Hue Contrast

Low Hue Contrast High Hue Contrast Histograms of Hue Differences

Example: Lightness Contrast (Fig.1)

Software Interface

Layout Process

• Non-stationary Hard-core Poisson
• Segmentation [Zeng et al. 09]
• Salience Map
• Steerable Filtering
• Stroke Density Map
• Histogram Matching
• Spatial Sampling
• Rejection Sampling

Stroke Neighborhood Graph

• Initialization
• Orientation Field [Zeng et al. 09]
• Local 2D Cartesian Coordinates
• Computing Nearest Neighbors
• One in each of the four quadrants
• Anisotropic

Attribute Processes

• Ensemble Statistics ---- Gibbs Energy ---- Diffusion Process
• Orientation
• Size
• Color
• Hue: Periodic, similar to orientation
• Lightness and Chroma: Aperiodic, similar to size

Experiments (Fig.8)

• (a) neutral
• (b) high size contrast (leaves) and
• low local isotropy (background)
• (c) high hue contrast

(a) (b) (c)

Experiments (Fig.9)

• (a) neutral
• (b) low density and high coarseness (wall)
• (c) high size contrast and high hue contrast
• (wall, shelf, can in the middle)

(a) (b) (c)

Experiments (Fig.10)

• (a) neutral
• (b) high lightness contrast
• (c) low local isotropy

(a) (b) (c)

Summary

• Customize painterly rendering styles via eight intuitive parameters
• Emphasize local contrasts
• Interactive software with real-time feedback
• Can simulate styles difficult to achieve using previous methods
• Future Work
• Parameter Space Analysis
• Neighborhood Design

http://bit.ly/stroke-processes

THANK YOU