Researchers Achieve Breakthrough In Image-Recognition Software
Scientists have been developing techniques to make computers capable of recognizing objects in an image. The field of image recognition has advanced in last few years but it still needs a lot of work. With the advent of Artificial Intelligence — the field of making computers act like humans — researchers have achieved milestones in image recognition.
Recently, two groups of scientists, working independently, announced that they have made their image-recognition software capable of recognizing content of photos and videos with far greater accuracy as compared to current software — thanks to artificial intelligence.
Researchers from Google and Stanford University unveiled this new software on Monday. They described that it is able to identify entire scenes, sometimes even mimicking human levels of understanding. The scenes can be anything, for example a bunch of crows sitting across the fountain.
After the software has identified the scenes, it then describes the scene in natural language. The researchers said that the descriptions are surprisingly accurate.
Here are a few more examples of sentence descriptions from images described by the experimental software:
This advancement can help a lot in making search over Internet highly relevant. Currently the search algorithms rely on annotated texts on image or videos.
For example, if a YouTube video has a title “Stairway To Heaven,” then the algorithm will look at the title and present it to the user who searches for this video. It has no clue what actually is inside the video. I could upload a video of a dog and name it “Stairway To Heaven” and YouTube would not be able to differentiate between the real and fake video.
This research will make the search algorithm look at the actual content inside the video, learn the scenes, and bring out highly accurate search results. It will then also be able to identify that “Stairway To Heaven,” uploaded by me, actually contains a dog which is not a stairway to heaven.
This technology could also be deployed in security cameras. It will then be able to identify suspicious behavior and raise alarm. Current tech used in cars also uses a similar technique to identify that accident is about to happen and will pull breaks even if driver fails to do so.
Google worked on a similar project called Google Brain, which identified cats in YouTube videos. Earlier in the summer at Microsoft Research Summit, Microsoft revealed Project Adam that accurately recognized the breed of dogs in the photos. But both of these projects identified single object in the photos.
On the other hand, this new research is able to identify whole scene containing multiple objects.