Google Research has teamed with Stanford University to improve computer image recognition capabilities.
The software being developed will allow computers to recognize objects in an image, determine context and produce a full description of the image.
For example this image:
Produces the description: Two pizzas sitting on top of a stove top oven
The technology still requires human interaction to "instruct" the computer by providing human captioned photos. Accuracy increases with each captioned image.
The most immediate impact would probably be in regards to image searches. Having a program that can determine image contents would greatly improve image search results. The search engine would not have to rely on surrounding text or the contents of an images <alt> tags.
This also holds promise for anyone that needs to produce image descriptions for large numbers of images, including photographers. The caption for every image could be automatically generated using a program instead of it having to be manually applied.
There are potential implications beyond those immediate uses. Security cameras, automated drones or cars, facial recognition software are items that could benefit from this ability to determine the items contained in an image along with context.
No comments:
Post a Comment