DPChallenge: A Digital Photography Contest You are not logged in. (log in or register) 
 

DPChallenge Forums >> Hardware and Software >> Zero Noise Photography
Pages:  
Showing posts 1 - 25 of 52, (reverse)
AuthorThread
 08/15/2007 04:34:36 PM · #1
Here is an interesting article on using HDR techniques to eliminate noise in an image.

Zero Noise Photography
 08/15/2007 04:47:54 PM · #2
took a quick look , sounds very interesting, i will give it a shot tonight:)

thanks!
 08/15/2007 05:07:21 PM · #3
Read through it, it's a good idea to combine the 'expose to the right' technique with HDR.
 08/15/2007 05:26:31 PM · #4
Originally posted by gloda:
Read through it, it's a good idea to combine the 'expose to the right' technique with HDR.

Seems to be saying that this is distinct from HDR though. I'd like to know if it would be possible to mimic this using HDR software.
 08/15/2007 05:57:48 PM · #5
Well, it's sure not "zero" noise, but it does attack one of the most bothersome aspects of digital imaging - shadow noise. It is very similar to HDR, but the end goals are a little different. Instead of compressing DR, there is essentially no "tone mapping." The overexposed shot is simply pulled down in exposure in conversion, resulting in a *very* smooth shadow rendition compared with a single capture.
It's a splendid idea where shadow rendition is very important and where a static scene is being photographed.

Edit:
The reason I say it's not "zero noise" is that there are noise sources that are not addressed at all by this technique, such as readout noise.

Message edited by author 2007-08-15 17:59:12.
 08/15/2007 06:41:53 PM · #6
What's readout noise?
 08/15/2007 07:37:39 PM · #7
Originally posted by gloda:
What's readout noise?

Ditto. Huh?
 08/15/2007 07:54:05 PM · #8
Originally posted by gloda:
What's readout noise?


Readout noise is the noise inherent in the camera circuitry that reads data from the sensor. You can't get rid of this in the way he suggests, but his methodology does help minimize its visibility. For informational purposes, there are several sources of noise in images from digital cameras:
- Random noise due to photon arrival statistics
- Fixed pattern noise, which is a function of the on-sensor electronics. Think of it as "amplifier gain error."
- Random thermal noise. We're counting electrons, and that means we have to keep the electrons trapped in very small cells. Thermal energy causes some to escape. Thermal noise also plays into amplifier noise.
More and more, noise at high ISO is dominated by photon arrival statistics. This "noise" is really a feature of the physical scene we're imaging and unless we can collect more photons, we can't lower this noise source. Collecting more photons can be done in the following ways:
- Increase pixel size (duh!). Double the size, you quadruple the area and reduce noise by 4x
- Increase quantum efficiency (there's a limit to this). We're closer to the limit than you might think. If we eliminated Bayer filters, we'd me near 30%, and some monochrome cameras are as high as 65%, which is less than a stop away from "perfect".
- Gather photons longer. This is what he's doing to reduce shadow noise. Take a longer exposure, and use it to reduce noise in shadows. Works, as long as you have a second, shorter exposure to retain highlights.
 08/15/2007 08:31:47 PM · #9
wow kirbic.....

<<<>>>>

you got left and right brain stuff between your ears. Smart and Artistic all in one head?
 08/15/2007 09:33:10 PM · #10
Originally posted by Sheryll:
wow kirbic.....

<<<>>>>

you got left and right brain stuff between your ears. Smart and Artistic all in one head?

That's why we call him, "Oh captain, my captain."
 08/16/2007 11:24:10 AM · #11
Cool stuff, thanks!
 08/16/2007 12:52:55 PM · #12
Originally posted by kirbic:

- Increase pixel size (duh!). Double the size, you quadruple the area and reduce noise by 4x

Do you mean the size of each pixel on the sensor?
08/16/2007 01:01:05 PM · #13
Originally posted by gloda:
Originally posted by kirbic:

- Increase pixel size (duh!). Double the size, you quadruple the area and reduce noise by 4x

Do you mean the size of each pixel on the sensor?

During chip design, one of the failing points of greater Megapixels is that those pixels all must fit in the same same area as before making each pixel smaller.
 08/16/2007 01:26:37 PM · #14
Originally posted by kirbic:
It's a splendid idea where ... a static scene is being photographed.


Maybe this is a dumb hypothetical but I'm trying to imagine a way for this to work in a single shot and be used for high contrast / high DR portraits (and am hoping Kirbic can set me straight ;)
Could a camera be fashioned so that the image is reflected on two image sensors, one at the "correct" exposure and the other at the 4 stop overexposure done via increasing the ISO to achieve better results (admittedly not as good as two shots of different length exposures).

The increase in noise reduction in the shadows of the higher iso image would have to outweigh the increased noise by virtue of shooting at a higher ISO.
Not likely for many cameras I assume but maybe for or approaching for the MKIII sensor that has great high iso noise characteristics?
Also, as tech gets better the high ISO noise quality would go down so maybe this would work if not today?

Or maybe I'm totally missing the point as better high iso quality would solve the poor shadow rendition of digital images?
The way I read this is the poor shadow rendition is inherent to the digital chip itself (at least as implemented by all chips today except the Fuji high DR one) and an less noisy high ISO sensor does not solve this abrupt clipping.

This shortcoming of digital is why I still prefer film for high contrast situations such as late day light coming through blinds on someone's face.

(was that coherent at all? :P )
 08/16/2007 01:43:19 PM · #15
Originally posted by rswank:
Originally posted by kirbic:
It's a splendid idea where ... a static scene is being photographed.

Could a camera be fashioned so that the image is reflected on two image sensors, one at the "correct" exposure and the other at the 4 stop overexposure done via increasing the ISO to achieve better results

If you had two sensor you'd need 2 shutters working at 2 different speeds.

I've been thinking about this myself in the area of Image Bracketing but using one sensor and one press of the shutter. I'd think that it would be possible to calculate the shutter times. When one presses the shutter it takes an image and pulls the pixel data at each shutter interval and saves it to an image file.

Just my thought.
08/16/2007 03:01:52 PM · #16
Originally posted by rswank:
Maybe this is a dumb hypothetical but I'm trying to imagine a way for this to work in a single shot and be used for high contrast / high DR portraits (and am hoping Kirbic can set me straight ;)


You can already do this in a way by using a RAW exposure and bumping the exposure compensation to +4EV. It won't be the same as physically overexposing the image because in doing this you are introducing more of a software aspect into the mix. Using two different ISO settings wouldn't help since they would contain two different amounts of noise. what eug said would be the best way to do it. Have a setting on the camera that auto brackets the to the appropriate settings and combines them either in camera or later on in post.

Noise is inherent to a digital chip since you are converting analog information into digital information. There is always going to be SOME amount of data loss/corruption in converting anything.

The dynamic range of a typical digital sensor is less than that of film, but has greater possibilities in that you can easily and effectively get very high dynamic range images by introducing software processing.

A less noisy sensor is going to only do just that, less noise. It won't increase the DR or shadow rendition. It will, however, reduce the noise in those shadows.

As technology progresses, you'll see sensors that feature a higher dynamic range and better shadow performance.

Originally posted by rswank:
(was that coherent at all? :P )


No, not really. :) But I hope I at least helped somewhat.
 08/16/2007 03:58:44 PM · #17
Originally posted by _eug:
Originally posted by rswank:
Originally posted by kirbic:
It's a splendid idea where ... a static scene is being photographed.

Could a camera be fashioned so that the image is reflected on two image sensors, one at the "correct" exposure and the other at the 4 stop overexposure done via increasing the ISO to achieve better results

If you had two sensor you'd need 2 shutters working at 2 different speeds.

I've been thinking about this myself in the area of Image Bracketing but using one sensor and one press of the shutter. I'd think that it would be possible to calculate the shutter times. When one presses the shutter it takes an image and pulls the pixel data at each shutter interval and saves it to an image file.

Just my thought.


I'm trying to think of a way to not require two different shutter speeds but still increase the shadow rendition with a modified technique outlined in the posted link.
Specifically for subjects that are not still and two different exposure times would cause a problem - person, animal, car etc.

Originally posted by SamDoe1:

Using two different ISO settings wouldn't help since they would contain two different amounts of noise.

Two different amount of noise wouldn't matter as long as the shadow noise in the lower ISO image is greater than the shadow noise in the higher ISO image, right?
That's where the combination technique of "4 F-Stops" would help the shadow definition and would make the shadow to highlight transition appear non-linear like film's response instead of hard clipping it like digital does (which is what I see as an added benefit that the linked article did not explicitly point out, no?).
That's why I brought up that less noisy future sensors would make this more plausible since the higher ISO image would be more likely to not have more noise in the shadows as in the lower ISO image.

Two image sensors with one lens/aperture with the image mirrored to both sensors at the same time.

Wouldn't work?
 08/16/2007 04:06:33 PM · #18
wooaahhh
 08/16/2007 04:14:29 PM · #19
Originally posted by rswank:
I'm trying to think of a way to not require two different shutter speeds but still increase the shadow rendition with a modified technique outlined in the posted link.
Specifically for subjects that are not still and two different exposure times would cause a problem - person, animal, car etc.


This is where all the software comes into play. RAW files can do just this.

Originally posted by SamDoe1:

Using two different ISO settings wouldn't help since they would contain two different amounts of noise.
Originally posted by rswank:
Two different amount of noise wouldn't matter as long as the shadow noise in the lower ISO image is greater than the shadow noise in the higher ISO image, right? That's where the combination technique of "4 F-Stops" would help the shadow definition and would make the shadow to highlight transition appear non-linear like film's response instead of hard clipping it like digital does (which is what I see as an added benefit that the linked article did not explicitly point out, no?).
That's why I brought up that less noisy future sensors would make this more plausible since the higher ISO image would be more likely to not have more noise in the shadows as in the lower ISO image.

Two image sensors with one lens/aperture with the image mirrored to both sensors at the same time.

Wouldn't work?


There's no way that the noise in the lower ISO image would be MORE than the noise in the higher ISO image, that's not possible. The reason the "4 f-stop" method works is because it simply records more information for the shadow regions and puts it in there instead of the noise. Noise is simply garbage information that doesn't really mean anything. In shadows, there isn't anything much to record therefore introducing noise. By overexposing, you are recording MORE light from the image and therefore getting more information out of the shadow regions to replace the random garbage you see as noise.

Having two different sensors with the same shutter speed is kinda pointless because they would, and should, expose the same image because of what I said before in that differences in ISO isn't what we're after with this method.

Originally posted by lovethelight:
wooaahhh


I know, right? This is what engineers do all day...

Now to wait for kirbic to either come and back me up or tell me I'm a fool.

Message edited by author 2007-08-16 16:25:42.
 08/16/2007 05:13:36 PM · #20
Originally posted by SamDoe1:
Originally posted by rswank:
lots of stuff you can read above if you want :P


Originally posted by SamDoe1:
There's no way that the noise in the lower ISO image would be MORE than the noise in the higher ISO image, that's not possible.

I'm not suggesting that there would be more total noise in the lower ISO image, just more noise to actual image data in the shadow area vs. the shadows in higher ISO image.

Originally posted by SamDoe1:
In shadows, there isn't anything much to record therefore introducing noise. By overexposing, you are recording MORE light from the image and therefore getting more information out of the shadow regions to replace the random garbage you see as noise.

I have a problem agreeing with this.
The way I understand it is that there isn't "anything much to record" it's just that less of the 12 bits are dedicated to that region which is why you expose to the right in the first place. To get more shadow data.
It's not that there isn't anything to record it's that the sensor (or raw converter in te camera?) is only allocating 128 of the available 4098 "levels" to this darkest area.
Overexposing via RAW doesn't change the fact that at the initial capture this darkest region was only allocated a much lower amount of data compared to overexposing be it via increased shutter speed, larger aperture or higher ISO.

Originally posted by SamDoe1:

Having two different sensors with the same shutter speed is kinda pointless because they would, and should, expose the same image because of what I said before in that differences in ISO isn't what we're after with this method.

They wouldn't expose the same since one image is at ISO 100 1/250 f/8 and the other is at ISO 640 1/250 f/8.
More data allocated to the "shadows" in the second image than in the first.
Same affect of shooting two shots at different times ISO 100 1/250 f/8 and ISO 100 1/30 f/8 or different apertures ISO 100 1/250 f/8 vs. ISO 100 1/250 f/2.
The difference between each pair of exposures is about 4 stops (if my math was right).

08/16/2007 08:30:53 PM · #21
Originally posted by rswank:
lots of stuff you can read above if you want :P


Ditto

Originally posted by rswank:
I'm not suggesting that there would be more total noise in the lower ISO image, just more noise to actual image data in the shadow area vs. the shadows in higher ISO image.


While that may be true, wouldn't the extra noise in the higher ISO image (which would be 4 stops and significantly higher) negate the advantage of getting actual image data for the shadows? The noise at this higher ISO would be much greater and have a different pattern than the one at lower ISO. Kind of like taking an image at ISO 100 and then doing a dark frame at ISO1600 for a subraction.

Originally posted by rswank:
I have a problem agreeing with this.
The way I understand it is that there isn't "anything much to record" it's just that less of the 12 bits are dedicated to that region which is why you expose to the right in the first place. To get more shadow data.
It's not that there isn't anything to record it's that the sensor (or raw converter in te camera?) is only allocating 128 of the available 4098 "levels" to this darkest area.
Overexposing via RAW doesn't change the fact that at the initial capture this darkest region was only allocated a much lower amount of data compared to overexposing be it via increased shutter speed, larger aperture or higher ISO.


You're right in that RAW doesn't change the initial exposure which is why seperate exposures would be a must to make this really work. I also agree in that it's not like there's nothing to record, but the sensor doesn't "see" what's there because it's beyond the dynamic range of the sensor's ability to record. This is why you push the exposure so that the dynamic range is shifted towards the higher end, making it so that the sensor can "see" what's there.

Originally posted by rswank:

They wouldn't expose the same since one image is at ISO 100 1/250 f/8 and the other is at ISO 640 1/250 f/8.
More data allocated to the "shadows" in the second image than in the first.
Same affect of shooting two shots at different times ISO 100 1/250 f/8 and ISO 100 1/30 f/8 or different apertures ISO 100 1/250 f/8 vs. ISO 100 1/250 f/2.
The difference between each pair of exposures is about 4 stops (if my math was right).


Right, but then the same different ISO issue comes into play.

I have a great idea though, why don't we actually try this and find out? Haha.

Message edited by author 2007-08-16 20:31:02.
 08/16/2007 10:14:03 PM · #22
Sameer is pretty much correct. It *is* in a sense that there is little to record in the shadows. The number of incoming photons is so few that there is "noise" associated with the variation in the number of photons that hit each pixel. Light is very much a "particle" phenomenon as far as the sensor is concerned. The sensor counts individual photons.
The only way around this type of noise is to either increase the area of each pixel (make the individual pixels larger) or to expose for longer. Period. That's why the +4 stops exposure does such a nice job with the shadow noise. By exposing 16 times as long, 16 times as many photons are collected.
 08/16/2007 10:27:40 PM · #23
Originally posted by _eug:
Originally posted by rswank:
Originally posted by kirbic:
It's a splendid idea where ... a static scene is being photographed.

Could a camera be fashioned so that the image is reflected on two image sensors, one at the "correct" exposure and the other at the 4 stop overexposure done via increasing the ISO to achieve better results

If you had two sensor you'd need 2 shutters working at 2 different speeds.

Just my thought.


why do you need two shutter speeds, why can't you stop reading data from 'short exposure' sensor. Is there any problem in this approach.

Message edited by author 2007-08-16 22:34:30.
 08/16/2007 10:28:57 PM · #24
Originally posted by kirbic:
Sameer is pretty much correct. It *is* in a sense that there is little to record in the shadows. The number of incoming photons is so few that there is "noise" associated with the variation in the number of photons that hit each pixel. Light is very much a "particle" phenomenon as far as the sensor is concerned. The sensor counts individual photons.
The only way around this type of noise is to either increase the area of each pixel (make the individual pixels larger) or to expose for longer. Period. That's why the +4 stops exposure does such a nice job with the shadow noise. By exposing 16 times as long, 16 times as many photons are collected.


Your brain is huge! Thanks for that.
08/16/2007 10:31:25 PM · #25
Originally posted by SamDoe1:
Originally posted by rswank:
lots of stuff you can read above if you want :P


Ditto

Originally posted by rswank:
I'm not suggesting that there would be more total noise in the lower ISO image, just more noise to actual image data in the shadow area vs. the shadows in higher ISO image.


While that may be true, wouldn't the extra noise in the higher ISO image (which would be 4 stops and significantly higher) negate the advantage of getting actual image data for the shadows? The noise at this higher ISO would be much greater and have a different pattern than the one at lower ISO. Kind of like taking an image at ISO 100 and then doing a dark frame at ISO1600 for a subraction.

That is the million dollar question here.
I would think that the signal to noise ratio of the shadow region of a RAW file from a Canon MKIII would be lower than that of the same region of an image shot +4 EVs.
Hell maybe even better on my noisy D70.
And if not today I would expect it sometime in the near future.
Reason being is that for the first image we're dealing with a maximum potential of only 128 'levels' of data for the blackest shadow regions compared to 2048 'levels' of potential data for those same areas of the image.
(btw, I got this from the luminious landscape article on exposing to the right - but doh! it's also in the OP's link haha)
The other thing unclear is the affect that a different noise pattern would have on doing the image combine.
I see nothing in the OP's link that says different noise patterns would cause a problem to his algorithms. Maybe I'll email him and ask.

Originally posted by SamDoe1:

You're right in that RAW doesn't change the initial exposure which is why seperate exposures would be a must to make this really work....

Cool, we agree on this.
I thought you were suggesting that you could overexpose the original RAW file 4 stops to achieve the effect in the article.

Originally posted by rswank:

They wouldn't expose the same since one image is at ISO 100 1/250 f/8 and the other is at ISO 640 1/250 f/8.
More data allocated to the "shadows" in the second image than in the first.....

Originally posted by SamDoe1:

Right, but then the same different ISO issue comes into play.

(see million dollar question posit above ;)

Originally posted by SamDoe1:

I have a great idea though, why don't we actually try this and find out? Haha.


I'd love to!
You have a couple of APC sensors to mess around with that you don't mind ruining? :P

I'm really trying to figure out if similar results can be done with a single shutter release as doing 2 so you are not relegated to static subjects and require a tripod.
(I guess you'd also need some crazy ass camera!)

This idea also hinges on the ability to project the same image on two sensors at the same time.
Is this optically possible? I'm sure it is but what would the mechanism be, combination of prisms and mirrors I assume?

Pages:  
Current Server Time: 04/25/2024 05:37:48 AM

Please log in or register to post to the forums.


Home - Challenges - Community - League - Photos - Cameras - Lenses - Learn - Prints! - Help - Terms of Use - Privacy - Top ^
DPChallenge, and website content and design, Copyright © 2001-2024 Challenging Technologies, LLC.
All digital photo copyrights belong to the photographers and may not be used without permission.
Current Server Time: 04/25/2024 05:37:48 AM EDT.