Super-resolution technology - turning grainy images to high resolution [4:17]
A small image of a face or a license plate taken from a surveillance camera becomes grainy and obscure when enlarged, making it impossible to discern.
The "super-resolution" technology of NEC makes recognition possible.
NEC Central Research Laboratories Akihiko Iketani
With its small area, the facial expression is hard to read from this image.
By using the super-resolution technology, the grainy image is restored in high resolution.
Compared with images processed by standard software, the difference is obvious.
First, a dictionary full of facial images is created for super-resolution.
The collected images are shrunk to deliberately create rough images.
Other parts are similarly registered and the dictionary is complete.
When a low resolution facial image is entered, one facial area is extracted.
A similar facial pattern is searched from the dictionary.
When it's found, the high resolution version is exported.
The nose is similarly processed.
The same image processing for the entire face recreates a super-resolution image.
None of my facial image data is registered in the dictionary.
What looks like my face here is actually a mix of other people's facial parts.
Strictly speaking, this isn't really me.
But the point is this technology can accurately reproduce a facial image.
Lots of brand new applications based on super-resolution technology will be available in the near future.
When a fugitive's surveillance image is released on the news, it's usually too grainy to make out the face.
By using this technology, facial features can be reproduced in high resolution.
We're hoping to combine this with facial recognition technology.
NEC's image recognition technology is constantly evolving in various applications for business and security services.
(January 25, 2013)