Think TV show type image enhancement is unrealistic?

[UlricT]

I remember people bitching about shows doing "impossible" image enhancement. Well, Intel has news for you.

from AT:
http://www.anandtech.com/tradeshows/showdoc.aspx?i=2367&p=2

:music: now i'm a believer :music:

 

[Viper GTS]

It's one thing to do it with a video (with multiple frames to work with), it's entirely another to do it with a single frame.

This is similar to the techniques used to get HD transfers of old film (dirty, scratched, etc.) and SD material.

Quite impressive though.

Viper GTS

 

[UlricT]

<devils advocate>

ok... but most stuff you see in these shows are from security cameras 🙂

</devils advocate>

 

[purbeast0]

thats pretty cool. the algorithm isn't even that complicated for something like that, i did stuff dealing with neighboring pixels in my image processing class. but you would definitely need to look at every single frame, which is ~ 75 in 3 second clip. and then look at every pixel in that image and THEN its neighboring pixels to some n^2 degree ...

it would take MUCH computer power to do it as they do on TV in the snap of a finger 🙂

 

[ucdnam]

Ya... in Enemy of the State, there's a scene in the store where a 3D image of a bag is created from a single security camera. They then rotate around the person and zoom in on the bag to show the shape and size of the item inside. I don't think that'll ever happen.

 

[purbeast0]

Originally posted by: ucdnam
Ya... in Enemy of the State, there's a scene in the store where a 3D image of a bag is created from a single security camera. They then rotate around the person and zoom in on the bag to show the shape and size of the item inside. I don't think that'll ever happen.

haha yea that was so over the top ...

jack black: "hypothesis coming right up ..."

2 seconds later, they have a "hypothesized" rendered shot of the other side of will smith and his bag.

 

[Rogue]

Just watched that Vegas show and they took a video camera shot, duplicated the side shot of the guy's head pressed on a counter top with a hand on it, grabbed it out of the image file, duplicated it, reversed it and mapped it to a 3-D model of the guy's head in a matter of seconds. I almost choked on my Pepsi when I saw that. It was even followed by the proverbial, "Honey, they can't really do that in real life, right?" question from my wife.

 

[mcveigh]

read the article...it says by 2015 we'll have that.....

 

[pulse8]

Originally posted by: Viper GTS
It's one thing to do it with a video (with multiple frames to work with), it's entirely another to do it with a single frame.

This is similar to the techniques used to get HD transfers of old film (dirty, scratched, etc.) and SD material.

Quite impressive though.

Viper GTS

Somehow I don't think that makes a difference. If the video is crappy all around, how is one frame going to help the other?

If you're transfering from film to tape and have dust or scratches, then you have surrounding frames where the image is clean and can pull information from that. I don't think that was done here.

 

[Ikonomi]

Wake me up when I can take a grainy picture of a building from 500 yards, rotate the building 180 degrees, and then zoom 500% on a window and enhance the resolution to crystal clarity. 😀