14 DAY TRIAL // Teaming up with different partners from Europe, IBM has launched for the public a new search engine that uses brand new image and video searching techniques. The algorithm scans every image uploaded on the web for color schemes, color structure, color layout, shape edges, patterns and texture, besides the classic tags or image metadata used by renowned search giants like Google, Yahoo or Bing.
The project is called SAPIR (Search in Audio-Visual Content Using Peer-to-peer Information Retrieval) and involves several research teams across Europe like IBM experts in Israel; Telenor researchers from Norway; the Max-Planck Institute in Germany; the Institute of Information and Science and Technology and the University of Padua, Italy (EURIX); Xerox researchers from France; the Masarvkova University in the Czech Republic and Telefonica experts from Spain. The project is currently developed using EU funds.
Thanks to its new search and scanning technology, it has risen above other image or video search engine researches like Muffin, CoPhIR, Quaero, Theseus and projects from LTU Technologies and Eyealike.
SAPIR aims to supply a rapid and accurate technology for users to employ whenever searching for related images or videos on the web. IBM and the researchers hope that in the future, users will be able to take a photo of (let's say) a logo and SAPIR will return other photos containing that logo, and possibly a photo containing a tag with the name of the company. Other useful implementations (the ones IBM is after) can be in medical diagnostics, product recognition (take a photo of some clothes, and SAPIR will tell you where to buy them), face recognition, location finder, GPS implementations (take a photo of where you are and SAPIR will tell you the location) and more.
A test search page can be found here, with great results in some cases, and some truly bad in others. The page queries for now only BBC videos and Flickr photos. Users can enter a search term and images are returned for that word. The user has additional search options after results have been displayed: to find similar images (will return images looking the same in colors and shape), or a “combo search” option (will use the selected image, alongside the search term – let's say searching for Barack Obama in a black suite sitting on a chair).
Also, after getting results from a search query, the user can click on the “Image” tab and enter latitude and longitude coordinates for an image using a Google Map API. The site may be down for some periods, but be sure to test it out since it's pretty accurate in some queries and might be useful for designers.
A demonstration video can be seen below.