Abstract
This paper presents a method for measuring the similarity in style between two pieces of vector art, independent of content. Similarity is measured by the differences between four types of features: color, shading, texture, and stroke. Feature weightings are learned from crowdsourced experiments. This perceptual similarity enables style-based search. Using this style-based search feature, we demonstrate an application that allows users to create stylistically-coherent clip art mash-ups.
Movie
Downloads & Links
- Paper [pdf, 6.27MB]
- Supplementary Part 1 [pdf, 13.7MB ]
- Supplementary Part 2 - Search evaluation [html]
- T-SNE Representation [html]
- Mash-up application [html]
- SLIDES [pptx, 83.6MB]
- ALL CLIP ART DATASET [png, 5.54GB] (Data copyrighted and used here under the limitations of the Fair Use doctrine)
- TRAIN/TEST DATASET [png, 127MB] (Data copyrighted and used here under the limitations of the Fair Use doctrine)
- FEATURES AND WEIGHTS [mat, 380MB]
- MTURK DATA [mat, 331KB]
Bibtex
Acknowledgements
We want to thank the reviewers for their insightful comments, the participants of the experiments, Carlos Bobed for his help with the experiments, Peter O'Donovan for sharing his regression code, and Daniel Osanz for some designs. This research has been funded by the European Commission, Seventh Framework Programme, through projects GOLEM (Marie Curie IAPP, grant: 251415) and VERVE (ICT, grant: 288914), the Spanish Ministry of Science and Technology (TIN2010-21543), and a generous gift from Adobe Systems. The Gobierno de Aragón additionally provided support through the TAMA project and a grant to Elena Garces.