Saliency algorithm: Puzzle-like Collage

Researcher:
Prof. Ayellet Tal | Electrical and Computer Engineering

Categories:

Information and Computer Science

The Technology

Automatic creation of digital albums, searching for visual photos and even creating a puzzle-like collage require nicely cut and matched images. Several methods of automating creation exist, and many implementing software packages are on the market. These involve the extraction of interesting, or regions from a series of images using various visual algorithms. These regions are then assembled in various fashions or transitions that are smoothed by graph cuts and alpha blending, which result in nicer transitions between images. Background regions from individual images, however, are usually not satisfactorily eliminated. Furthermore, current methods limit the region of interest to a rectangular shape. Since items like people are not rectangular, this causes the algorithm to include non-essential background regions in the final collage. The novel technology presents an algorithm for prime image segmentation, image rendering and image re-targeting for image-based applications. The method makes several additional contributions. First, it introduces a new algorithm for saliency map computation, which competes favorably with previous work. Second, a novel Region-of-Interest (ROI) extraction algorithm is presented. It is shown to extract non-rectangular regions that coincide with the meaningful object and background boundaries. Finally, it describes a composition algorithm that places these non-rectangular pieces together.

Advantages

  • Fully automatic operation while retaining optional user interaction
  • better utilization of space (in the final result) while retaining aesthetic composition
  • High quality image segmentation, image rendering and image re-targeting
  • Highly aesthetic appearance, resembling the work of a human artist

Applications and Opportunities

  • Image rendering and image segmentation for image based applications
  • Results of visual search- visual representation of photo search results online
  • Visual summary for indexing purposes- internet products (DVD, electronic appliances, E-Books), TV Channels, Websites, etc
arrow Business Development Contacts
Dr. Arkadiy Morgenshtein
Director of Business Development, ICT