(or Organized Memories)
Lihi Zelnik-Manor and Pietro Perona
Abstract
A single picture cannot always capture the full scene. It has thus become common among artists and amateur photographers to take multiple pictures of the same scene and compose them into image joiners (a term coined by artist David Hockney). Typically, these joiners are constructed by manually positioning and layering the images on a (digital) 2D canvas using software tools like Photoshop. This results in a fragmanted composition of images, where some images occlude others and boundaries across pictures are fully visible. In spite (or maybe due to) their fragmanted look, such representations are attracting a lot of attention and numerous examples are posted online in webpages such as Flickr (taged with joiners, Hockney, photocollage, etc.)
We are interested in automating the construction of such image joiners for two main objectives. The first is providing photgraphers with a good starting point from which they can more easily finalize their joiners. The second is providing a new way for exploring image collections now organized in correspondence with the underlying scene rather than by file names.
Paper
L. Zelnik-Manor and P. Perona, "Automating Joiners", To appear in NPAR 2007. (PDF)
Slides
Slides of a presentation given at Dick Lyon's photography course
Some Examples of our automatically generated joiners
Some Examples of semi-automatic joiners
Between 2 and 4 point correspondences had to be provided manually. The rest (i.e., alignment and layering of images) was done automatically.
Homage to David Hockney's LA Visitors series
Between 4 and 6 point correspondences had to be provided manually. The rest (i.e., alignment and layering of images) was done automatically.
Organized Memories
A more intuitive way for browsing image collections Preliminary demo
Related Links
A similar approach has been developed in parallel by Yoshikuni Nomura, Li Zhang and Shree Nayar
Our previous work on "squaring the circle in panoramas".
Back to Lihi's homepage
A single picture cannot always capture the full scene. It has thus become common among artists and amateur photographers to take multiple pictures of the same scene and compose them into image joiners (a term coined by artist David Hockney). Typically, these joiners are constructed by manually positioning and layering the images on a (digital) 2D canvas using software tools like Photoshop. This results in a fragmanted composition of images, where some images occlude others and boundaries across pictures are fully visible. In spite (or maybe due to) their fragmanted look, such representations are attracting a lot of attention and numerous examples are posted online in webpages such as Flickr (taged with joiners, Hockney, photocollage, etc.)
We are interested in automating the construction of such image joiners for two main objectives. The first is providing photgraphers with a good starting point from which they can more easily finalize their joiners. The second is providing a new way for exploring image collections now organized in correspondence with the underlying scene rather than by file names.
Paper
L. Zelnik-Manor and P. Perona, "Automating Joiners", To appear in NPAR 2007. (PDF)
Slides
Slides of a presentation given at Dick Lyon's photography course
Some Examples of our automatically generated joiners
| Cacti Garden
|
Arch |
Pond |
Anza-Borrego |
Long Beach Airport |
Uccello |
Tractor |
Truck |
|
Southewest |
Mickael McMillen |
Stav Zelnik |
Some Examples of semi-automatic joiners
Between 2 and 4 point correspondences had to be provided manually. The rest (i.e., alignment and layering of images) was done automatically.
Family |
Bill Freeman |
Impossible Bridge |
Homage to David Hockney's LA Visitors series
Between 4 and 6 point correspondences had to be provided manually. The rest (i.e., alignment and layering of images) was done automatically.
Francois Fleuret |
Marco Ponti |
Patrick Hughes |
Organized Memories
A more intuitive way for browsing image collections Preliminary demo
Related Links
A similar approach has been developed in parallel by Yoshikuni Nomura, Li Zhang and Shree Nayar
Our previous work on "squaring the circle in panoramas".
Back to Lihi's homepage