Computational Manga
CREATED: 200811260303 Speaker: Yingge Qu ** Importance of Manga
- significant industry in Japan, 16% of GDP
- labour intensive industry, computer as a basic drawing tool ** Manga colorization
- B/W and color mangas are separately produced, different techniques used in production
- Image analogies: require example pair
- Optimization [Levin et al, 2004]: rely on graylevel continuity
- Manga presents pattern continuity instead *** Pattern continuity
- shading and texture
- screening and hatching
- similarity of patterns defined based on Gabor wavelet feature *** Pattern segmentation
- using level set propagation
- leak-proofing by adjusting relaxation factor *** Colorization
- stroke-preserving colorization by color bleeding
- pattern to shading ** Structure-aware halftoning
- background production is tedious
- halftoning already being employed but results in lost of detail
- converts a grayscale image to bitonal image
- desirable properties: tone reproduction, fine structure preservation, blue noise properties
- define an objective function using tone reproduction and fine structure preservation, incorporate human visual perception
- optimation method: start with a bitonal image, swap image pixels to improve objecive function ** Richness preserving Manga screening
- halftone result still does not look like traditional manga
- rich screening using in manga
- manually assigning screen using semi-automatic methods still tedious in complex images
- 2D screen space (tone and pattern variety)
- richness preservation: tone similarity, texture similarity, distinguish colors based on pattern
- segmentation using mean shift algorithm
- framework: 1) texture matching 2) color-to-pattern mapping 3) tone matching
- texture matching using Gabor wavelet features
- similar colors assign to similar patterns using Multidimensional Scaling (2D chromaticity to 24D pattern space), multiple solutions possible
- anchor color-to-pattern matching using results from texture mapping
- tone matching using screen density ** Future directions
- Relighting
- Reproduce clear and neat lines from photos or sketchings
- Example based generation of characters
- Special effects