CosmicAI: generating sky backgrounds through content-based search and flexible composition

  • Authors:
  • Takanobu Mitani;Issei Fujishiro

  • Affiliations:
  • Keio University;Keio University

  • Venue:
  • ACM SIGGRAPH 2012 Posters
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

In film and game production, sky images are frequently changed to fit a given situation. SkyFinder [Tao et al. 2009] and A System for Editing Sky Images Using an Image Database (Ono2011 for short hereafter) [Ono et al. 2011] allow the users to generate sky images efficiently without three dimensional computer graphics software such as Vue and Terragen. SkyFinder users specify related attributes (category, layout, horizon height, sun position, and richness) so as to easily search sky images downloaded from Flickr.com. Ono2011 lets the users to take two steps. The first step is to generate a background with atmospherically-distributed sky color through an intuitive user interface. The second step is to compose clouds using a collection of real photographs downloaded from Flicker.com. The detailed user actions are to specify related features (sky color, cloud shape, amount of edges, and sun position); to select clouds in the search results by a paint-like operation; and to compose them into the original background. It looks like the quality of sky images generated by these two systems becomes unstable, because many public photographs are unsuitable for composition (See Fig. 2(a)).