Authors:
Gabriela Csurka
and
Marco Bressan
Affiliation:
Xerox Research Centre Europe, France
Keyword(s):
Non-photorealistic rendering, Semantic segmentation, Drawings, Coloring and activity page.
Related
Ontology
Subjects/Areas/Topics:
Computer Vision, Visualization and Computer Graphics
;
Non-Photorealistic Rendering, Painting-Like Rendering, Drawing
;
Rendering
Abstract:
We present a technique to generate some popular activity sheets from arbitrary images, in particular user photographs. We focus on activity sheets that are closely linked to coloring and shape completion. We first introduce a baseline approach based on color regions that works well for cartoon-like images and uncluttered hotographs. In more complex scenes, we show how this approach can be integrated with global textural cues for increasing the level of details that can convey semantic information. A final local stage takes advantage of object recognition and scene classification techniques for selective detailing in the foreground background regions. Though the resulting approach can be deployed in a fully automatic fashion, interactivity can be a desirable feature since it allows to account for errors and, more important, increase the level of personalization. We propose three levels of interactivity, depending on the user skills. For all steps of our system and addressed activity s
heets we show representative results.
(More)