Conceptual question / Computer Science / Forums

Forums

4hv.org :: Forums :: Computer Science

« Previous topic | Next topic »

Conceptual question

1 2 next

Move Thread

LAN_403

scrooch

Fri Jun 15 2012, 01:54AM

Registered Member #908 Joined: Wed Jul 18 2007, 05:53AM
Location:
Posts: 49

I''' get straight down to the gist of it:

Whats stopping someone from replacing the work done by the focusing lens of a camera by a transfer function?

Might be an utterly stupid question but there goes haha

Sulaiman

Fri Jun 15 2012, 05:34AM

Registered Member #162 Joined: Mon Feb 13 2006, 10:25AM
Location: United Kingdom
Posts: 3140

How would a transfer function capture an image?

Lens/ccd technology is well developed, fairly cheap and instant.

scrooch

Fri Jun 15 2012, 06:12AM

Registered Member #908 Joined: Wed Jul 18 2007, 05:53AM
Location:
Posts: 49

Sulaiman wrote ...

How would a transfer function capture an image?

Lens/ccd technology is well developed, fairly cheap and instant.

The CCD would be there ofcourse. The raw information from the CCD would go straight to the algorithm. The algorithm will mimic the "focusing" which was earlier done by a lens.

So in summary normal setup; just without the lens. the lens being replaced by the algorithm.

Dr. Slack

Fri Jun 15 2012, 07:31AM

Registered Member #72 Joined: Thu Feb 09 2006, 08:29AM
Location: UK St. Albans
Posts: 1659

So why can't you deconvolve the blur of a big pinhole?

1) knowing blurring function to sufficient accuracy
2) signal to noise ratio of the sensor
3) the sensor reads intensity, not signed amplitude

And the most important of these is 3. Although in any deconvolution exercise, noise enahncement due to zeros or small numbers in the deconvolution function, and inaccuracy of the estimate, can damage the result, that's all moot because of number 3.

The image sensor can't capture negative amplitudes. It reads only intensity. If you set a function up in MATLAB or Python/Scipy, blur it by convolving with the Fresnel function for a big pin hole, you'll see that some of the resulting amplitudes will be negative. However the imaging medium, whether film or CCD, will measure the absolute value. Having rectified the image, it is now impossible to deconvolve it to what it was. You might do a job of cleaning it up a bit, but in taking the absolute value, enough information is gone forever.

If you tried something like iterative backprojection, ie what scene could have caused this image, then you are still left with a massively underdetermined problem which can only admit to very approximate solutions.

Now, if you were recording a hologram, and beating the recording with a laser, you could reconstruct the original image, because the sign as well as magnitude is encoded.

Pinky's Brain

Fri Jun 15 2012, 08:54AM

Registered Member #2901 Joined: Thu Jun 03 2010, 01:25PM
Location:
Posts: 837

Dr. Slack wrote ...
Now, if you were recording a hologram, and beating the recording with a laser, you could reconstruct the original image, because the sign as well as magnitude is encoded.

Actually generally they don't record both (that's why off axis holography is generally used, lets you separate the error from the original signal on replay).

Steve Conner

Fri Jun 15 2012, 04:40PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

The Lytro light field camera claims to allow you to refocus its pictures after they are taken. Not sure how well it works in practice.

scrooch

Sat Jun 16 2012, 03:17AM

Registered Member #908 Joined: Wed Jul 18 2007, 05:53AM
Location:
Posts: 49

Thanks for the responses guys, very interesting stuff. I guess the same applies for trying to mimic a zoom lens.

Though on a slightly different tangent I recall a concept camera developed by canon having a ridiculous level of zoom (Maybe they used a massive sensor with over sampling). I wonder what they did

Steve Conner wrote ...

The Lytro light field camera claims to allow you to refocus its pictures after they are taken. Not sure how well it works in practice.

The lytro uses a wholly new concept altogether, there are some examples on their website, it does work well, but im not aware of any real use for it besides the novelty aspect.

Dr. Slack

Wed Jun 20 2012, 06:52AM

Registered Member #72 Joined: Thu Feb 09 2006, 08:29AM
Location: UK St. Albans
Posts: 1659

This article in New Scientist this week suggests another way of improving the image, subtracting the intensity of two images, where there's a known difference in scene between the two.

They do say that, even enhanced by the technique, it's only an improvement, it's still not as good as a real pinhole

Bored Chemist

Wed Jun 20 2012, 07:37PM

Registered Member #193 Joined: Fri Feb 17 2006, 07:04AM
Location: sheffield
Posts: 1022

Broadly the equivalent of Dr Slack's third point.

hboy007

Fri Jun 22 2012, 10:32PM

Registered Member #1667 Joined: Sat Aug 30 2008, 09:57PM
Location:
Posts: 374

forgive me for not verifying it has already been pointed out here somewhere:

coded aperture lens systems avoid the blind deconvolution problem and allow a crude depth map reconstruction of a single 2d image. I recently found something interesting about the algorithmic improvements of full-res reconstructions of light field images. Coded aperture pictures tend to suffer from strange artifacts.

1 2 next

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob