Google software can interpret and describe complex images as naturally and accurately as a sighted person – without human assistance.
Beyond Object Recognition
Two new artificial intelligence software programs can accurately interpret images and describe them more accurately than ever before. Developed independently by Stanford University and Google, the software programs can often describe images as accurately and meaningfully as humans.
Until now, image recognition software was limited to understanding individual objects in an image. The new software teaches itself to understand an entire scene, capturing how the objects relate to one another and describing it in a natural-sounding language. For example, “A group of young people playing a game of frisbee.”
The immediate application for the software will be to catalog all of the images on the Internet. But its potential for assisting the blind is obvious.
Applying Neural Networks in Software
To develop their software, Google and Stanford researchers used programs known as neural networks. Functioning much like neural networks in human brains, these computer programs train themselves to learn the relationships between objects and to identify patterns – even when their human creators can’t identify them.
The programs aren’t perfect and often misinterpret an image with errors that range from “minor errors” to “unrelated to the image.” However, the success is far greater than any software has achieved to date. The initial results are very promising.
To learn more about Google’s new image software, check out the article on the Google Research Blog.
- Learn about Deque’s advanced automated software.
- Do a quick check to see if your website is accessible.
- Check out our upcoming webinars.
- Contact us with your accessibility questions.