Actually, now it is (sometimes) possible to sharpen blurry video to get more detail. Last month the EE Times had an article on the subject. A new company is offering just this service. I can't give you a direct link, but you can go to eetimes.com and put "blurry video" in the search box, and click on the top returned article,
Service reconstructs blurry video as sharp stills
EE Times
, September 18, 2006
.
The article mentioned the product service name: Sarensix. Googling on that brought me to the company web site,
http://www.pyramidvision.com/products/sarensix/index.asp.
There you can download a PDF file with more information & images,
http://www.pyramidvision.com/products/sarensix/sarensix_datasheet.pdf .
I'm sure it is not inexpensive.
I'm sure the sharpness comes from aligning & combining multiple noisy image frames of the object, such as a person's face.
If you only have one frame of the object, this technique will not work.
--
Harry Parker
Senior Systems Engineer
Dialog Imaging Systems, Inc.
----- Original Message ----
From: Johannes Schindelin <
[hidden email]>
To:
[hidden email]
Sent: Monday, October 16, 2006 5:18:57 AM
Subject: Re: Video and Image restoration case
Hi,
On Sun, 15 Oct 2006, Yili Zhao wrote:
> Recently, there was a thief case happpened in my university, and the
> observer system had recorded it. However, the thief's face was not clear
> in the recorded video.
> I want to know is it possible to restore the man's face in some degree
> (maybe restore to 100% is not possible)?
Contrary to TV shows, it is not possible to "restore" the face. It is a
simple law of Information Theory that by post-processing images, you
actually take information _out_ of the image, not put some in.
Having said that, there are a few methods to put some external belief into
the images, like smoothing (from experience you expect the face to be
non-noisy), and then sharpening again (from experience you know that a
face is not blurry). Alternatively, you could apply a band-pass filter or
despeckle the image.
But please understand that this method is prone to errors. It is actually
very easy to do some innocent operations on a face image, and come up with
a face resembling someone else.
It is much safer to rely on other clues like "where has the thief
entered?" or "did he evidently know something only few people know?". For
example, in a case I know of, the thief tried to hide from the camera, but
only few people knew there was one.
Hth,
Dscho
Locked
