Samples of photo pairings
Jul. 8th, 2009 12:37 amIn an effort to determine the feasibility of my world-spanning photo-pairing idea, I've started playing around with the photos I have sorted and tagged so far, doing searches by tag cloud similarity and looking for good candidate matches. So far I only have marked photos from Singapore, India, the Czech Republic, and part of Vietnam, which together constitute about 12% of my trip photos.
I have found that the pairs are much more aesthetically pleasing to look at when the images have similar structure as well as content. I may want to look in to using methods of extracting quantitative compositional data from the images in order to help people locate good pairs. Here are some examples of what I've found so far.
Riverfronts: Singapore vs Saigon

Religious statues: Ellora, India vs Prague, Czech Rep

Temples: Vietnam vs India vs Prague

Indians in Singapore vs Indians in India.

More on Flickr here:
I have found that I can get similarly good matches around a theme by doing two two-word queries on Flickr consisting of "location1 theme" and "location2 theme". So I could create a lightweight version of the idea that just used a stock list of locations and a stock list of common items to find interesting candidate matches. This would save me from having to sort and tag all my photos by a tight deadline.
I have found that the pairs are much more aesthetically pleasing to look at when the images have similar structure as well as content. I may want to look in to using methods of extracting quantitative compositional data from the images in order to help people locate good pairs. Here are some examples of what I've found so far.
Riverfronts: Singapore vs Saigon

Religious statues: Ellora, India vs Prague, Czech Rep

Temples: Vietnam vs India vs Prague

Indians in Singapore vs Indians in India.

More on Flickr here:
I have found that I can get similarly good matches around a theme by doing two two-word queries on Flickr consisting of "location1 theme" and "location2 theme". So I could create a lightweight version of the idea that just used a stock list of locations and a stock list of common items to find interesting candidate matches. This would save me from having to sort and tag all my photos by a tight deadline.