Just give it a photo and it will find what you need

Nov 19, 2014 09:39 GMT  ·  By

Image recognition has mostly been used for log on and the like, not counting the applications in criminal science and investigations. A group of researchers from Dartmouth wanted to take things further though.

Their idea revolves around a pretty simple concept. If a sensor, backed by an appropriate program, can “see” someone or the environment and recognize them for what they are, then the same should be doable for photos.

This opened up a new way to look at data stored on the Internet. There has been the option to search for pictures for a long time, but finding relevant related images doesn't always pan out.

You really need to be lucky to get good results from the “I'm feeling lucky” option on Google, at least in cases of image similarity searches.

Now, a study published in PAMI (IEEE Transactions on Pattern Analysis and Machine Intelligence) reveals that an AI has been created to make everything easier and more accurate.

The new AI uses image recognition software

Image recognition is that nifty little ability that some devices and software programs have to, well, recognize an image for what it is.

A new system has been developed by Dartmouth College, Tecnalia Research & Innovation, with help from Microsoft Research Cambridge.

This system has been tested on photos and can be applied to videos as well. A machine was able to learn an algorithm to recognize images and retrieve accurate and efficient data from it in order to search relevant matches to a query through online documents.

Pixel data in images and (eventually) video was never possible to use this way before, only text. That means that the system works both ways, giving computers the ability to recognize pixels associated with search phrases by studying the results from text-based image search engines.

The more such knowledge is accumulated, the more can be applied to other photos without tags or captions.

Thus, relevant pictures and videos that would otherwise be missed during a web search can finally show up in Google and other search engines even if the uploader didn't add the necessary tags or keywords in the description.

It should also, in time, allow one to find interesting frames in videos without loading or playing said videos, if photos aren't available or enough.

The bottom line

Loads of images on the Internet that have been overlooked by search engines for the past 30 years will finally come to light and do their part in the furthering of everyone's personal or professional learning and entertainment efforts.