Facebook already has the capability to predict tags, but now a new algorithm developed by a team at the University of Toronto could give it the ability to determine relationships between people.
The technology is called relational social image search and it works by using the frequency of tags within photos combined with how they appear in proximity to others in photos to determine relationships. Parham Arabi, a professor in electrical and computer engineering, developed and patented the system. Wired explains in more detail how it works.
"If you are tagged in Facebook or Flickr pictures with your mother and are close together, and you are also tagged in separate pictures with your father, the algorithm can determine that there is a relationship between those two and assess how strong that relationship may be. Imagine another set of photos where you feature with both your parents, but only your mother is tagged. If you search for your father in the batch of photos, these untagged images will also be returned because of the high likelihood that he features in the pictures."
Arabi has been working on this technology since 2005, but previously he was focused on technology that understood images based on content recognition. He switched it up to develop the current algorithm.
"Instead, we decided to focus on a very basic tag information that is often available but almost always ignored, which is the location of tags in images. By running the location of tags in images through our mathematical model, we obtain a relativity graph which helps us both understand social relationships and also to search images better," he told Wired.
Read the full story
here.
Original Source: Wired