NEC develops super-resolution technologies for fine magnification of surveillance camera images

Dec 16, 2011

NEC Corporation announced today the development of super-resolution technologies for fine magnification of surveillance camera images, including persons' faces and license plates.

Until now, existing technologies required numerous still images from a video in order to improve a subject's resolution and enable its . However, these images blur when magnified by more than 2 or 3 times (4 to 9 times the number of pixels). Therefore, significant demand remains for technologies that improve resolution and enable images to be magnified with greater clarity.

NEC's new technologies create a super-resolution image from just one shot of a subject, such as a person's face or a car's license plate, by utilizing a database (library) of categorized images. This enables images to maintain fine details even when magnified by more than 4 times (more than 16 times the number of pixels), making it possible to distinguish small and distant subjects, which was difficult with conventional technologies. These new technologies are expected to be applied with surveillance cameras that cover such large areas as airports or traffic intersections.

Key features of these technologies are as follows:

1. Creates images with fine details when highly magnified

These technologies create a library of images for subjects that are used to carry out super-resolution processing of images. In order to deal with subjects with different sizes, the library stores images with a variety of resolutions, and the best resolution images are automatically selected.

2. Small size image library data

A small size image library of data that can be deployed with a variety of equipment is created from a large set of image data. Redundant are quickly eliminated from the library, which makes the library as small as possible without affecting image quality.

Looking forward, NEC aims to expand the use of these technologies into a broad range of fields, including the enhancement of satellite and , as the company actively promotes development of technologies that contribute to the safety and security of daily lives.

Explore further: Ride-sharing could cut cabs' road time by 30 percent

add to favorites email to friend print save as pdf

Related Stories

Sharp Introduces 65V-inch LCD Monitor

Feb 17, 2006

Sharp will introduce into the Japanese market the PN-655 LCD Monitor made at the Kameyama Plant, which uses a 65V-inch full-spec high-definition LCD panel (resolution of 1,920x1,080 pixels).

Recommended for you

User comments : 7

Adjust slider to filter visible comments by rank

Display comments: newest first

Pattern_chaser
5 / 5 (1) Dec 16, 2011
I wonder what is the chance of INCORRECTLY identifying a number plate, as the detail is effectively created, and not derived from the original picture?
fmfbrestel
4.3 / 5 (3) Dec 16, 2011
@pattern chaser: If your not being rhetorical -- yes it will have the ability to make mistakes.

If you are being rhetorical:
What are the odds that a human eye witness would INCORRECTLY report a number plate? error happens. Should an image greatly magnified beyond resolution be used as the sole evidence in a case? No. Could it still provide useful leads that are good enough for probable cause? probably. Everything can be misused, that doesn't mean the tech is worthless, it just means we need to understand it's limitations.
ODesign
2 / 5 (1) Dec 16, 2011
Makes sense. if you see a circle in the image and increase the resolution, there's no need to make the edge blurry or introduce stair-stepping. Amusingly, the result will be idealized enhancements. Circles become smoother, skin will be enlarged with either stereotypical idealized pores or more likely no skin pores.

The eye does this already as can be seen when a dotted line from a distance is percieved as a thinner solid line.
Jotaf
5 / 5 (1) Dec 16, 2011
Matching to images in a DB and calling it "super-resolution" is very misleading!

If your face happens to be one of the training samples, then many criminals will look just like you, because your face's large-scale features happen to be similar to those in some surveillance images.

If you're curious to what the state-of-the-art on super-resolution *really* is, check out the work by Cremers, you can download a good paper here. The nice pictures are near the end. This is recovered *from a video*, not a single image.

http://cvpr.in.tu...pr08.pdf

I hate it when companies try to pitch their products as more than what they're really are. In the area of security it's even more irresponsible.
plasticpower
2.5 / 5 (2) Dec 17, 2011
I think what they're talking about here is "vectorizing" the image to a degree. Common shapes will be replaced with vector graphics. Vector graphics are described by math, and therefore can be zoomed into infinitely.
Vendicar_Decarian
5 / 5 (2) Dec 17, 2011
You can not extract from an image, information that was not there.

You can embellish an image and make it look more realistic, but it is no different than an artist using photoshop to make an image look "better".

A system such as this has no place in a security setting, and no legitimate court would allow such information in court.

Pattern_chaser
5 / 5 (1) Dec 20, 2011
Clarification: I have two points:

1. The detail is effectively **created**, and not derived from the original picture. The resulting "super-resolution" image could be misleading, or just plain WRONG. We need to know how likely this is!

2. The results from this technology are likely to be presented as being more accurate than they are, by NEC, by lawyers, and by anyone with a vested interest.