A joint project involving Nanyang Technological University in Singapore and Capital Normal University in China is working on computer-vision software that can not only recognize faces but link similar features in order to determine if the faces in question are related.
This program, which is trained to look for family resemblance, initially drew from a database of well-known adults and their children, with match and mismatch examples, in order to help it better distinguish likeness in looks.
It works to first identify if a photo is of an adult or a child. Analyzing photos on a pixel-by-pixel basis, the program then works off of its knowledge of facial similarities in the correct matches already in its database. According to a test study done with 160 adult-child pairs (of which half were correct matches and the other half not), the success rate of confirming or denying matches was 68 percent.
A team at Cornell University expanded upon this project, narrowing down six criteria that are the best indicators of genetic relation in the face. Using this information, a test of 150 celebrity adult-child matches came out with a success rate of 71 percent, which fared slightly better than a human-conducted test of 67 percent.
While it’s not yet certain how this technology could be improved to increase the success rate, there are already many ways in which people could benefit from this system. Possible uses for this technology include simple things, such as identifying related persons on social networks, to more advanced uses, such as tracking down birth parents or refugee family members. In what other areas could this be put to use?