Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
11-04-2010, 09:44 PM   #1
Junior Member




Join Date: Jan 2008
Location: Portland, OR
Photos: Albums
Posts: 25
Are 14 bits enough to capture the full K-5 dynamic range?

Here is a boring snapshot of my scanner (chosen just because it contains a smooth gradient) that I used as an extreme test of the K-5's capabilities. It would have been properly exposed for ISO 6400, but the picture was taken at ISO 100, 6 stops underexposed. It was then over-exposed 8 stops in Lightroom to correct for the under-exposure (and then some). The 8 stops were achieved by combining a gradient filter (+4 stops) with the regular exposure slider (+4 stops also).



The histogram at the top right is rather sparse and I think the color patterns visible in the center image (100% crop) are due to that. So it looks like the K-5 really needs 14 bits per channel and might even benefit from more in extreme cases. Of course, this example is rather extreme and I don't expect to encounter banding in any regular photos.

Based on earlier experiments with my K-7 I don't think it could generate such banding due to its higher noise levels, even though it only has 12 bits per pixel.

11-04-2010, 10:11 PM   #2
Senior Member
65535's Avatar

Join Date: Dec 2008
Location: Canada
Posts: 187
please post the same pic from the k7 for comparisons' sake
11-04-2010, 10:13 PM   #3
Veteran Member
Smeggypants's Avatar

Join Date: Sep 2010
Posts: 1,536
Yes that pic means nothing at the moment.
11-04-2010, 10:57 PM - 2 Likes   #4
Veteran Member




Join Date: Mar 2009
Location: Perth Australia
Photos: Albums
Posts: 1,514
I assume this screen shot is on a monitor, capable of showing all the bits?

11-04-2010, 11:22 PM   #5
Veteran Member




Join Date: Dec 2007
Posts: 923
I have had badly exposed K-7 photos before, mainly due to accidentally taking the shot before the external flash was ready.
With the K-7, recovering even 3 stops underexposed at ISO 100 would have looked worse than that. I don't have a sample right now because it usually goes into the Trash bin....

If that's a 6 stops underexposed K-5 shot, its noisy and ugly, of course, but I would say its pretty impressive, nonetheless
11-04-2010, 11:27 PM   #6
Zav
Pentaxian
Zav's Avatar

Join Date: Feb 2009
Photos: Gallery
Posts: 1,371
QuoteOriginally posted by WerTicus Quote
I assume this screen shot is on a monitor, capable of showing all the bits?
That's the point.
11-05-2010, 04:34 AM   #7
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,871
QuoteOriginally posted by Olivier Quote
The histogram at the top right is rather sparse and I think the color patterns visible in the center image (100% crop) are due to that. So it looks like the K-5 really needs 14 bits per channel and might even benefit from more in extreme cases.
The K-5's 14th bit is rather random for all ISO levels. The 13th bit is a bit noisy. This is the current state of wisdom before formal lab tests are published.

As soon as part of the noise is captured, banding doesn't occur. Noise acts like the dither algorithm in printing.

The histogram may be as sparse as you like, it doesn't matter at all. As soon as you include "the noisy bit" then it doesn't help to improve bit depth beyond that point. Even duotone black & white images can look great with the correct dithering.


Also, in your LR snapshot, you should reduce the black level. You're currently destroying some image information by clipping the blacks.

11-05-2010, 06:55 AM   #8
Junior Member




Join Date: Jan 2008
Location: Portland, OR
Photos: Albums
Posts: 25
Original Poster
This is obviously a screenshot from a monitor and the Lightroom settings are such that we should hopefully see all the bits.

Here is the same picture, tweaked some more in an attempt to more clearly show what I mean:



I played with the luminance sliders and saturation to more clearly show the patches of different colors and set the black point to 0 as per Falconeye's recommendation.

These colored patches don't look like the usual sensor noise. It looks like banding due to insufficient number of bits that is not entirely masked by noise dithering. However, it might also be due to some other reason, maybe as a result of Lightroom's image processing and noise reduction algorithms?
11-05-2010, 08:40 AM   #9
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,871
QuoteOriginally posted by Olivier Quote
These colored patches don't look like the usual sensor noise. It looks like banding due to insufficient number of bits that is not entirely masked by noise dithering. However, it might also be due to some other reason, maybe as a result of Lightroom's image processing and noise reduction algorithms?
It's certainly not banding. Anything from fixed pattern noise to an uneven surface of the scanner plastic. At these low luminosities, one has to be very careful with interpretation.
11-05-2010, 08:56 AM   #10
Veteran Member
Eruditass's Avatar

Join Date: Jul 2009
Photos: Gallery
Posts: 2,207
falconeye, can you explain how the K-5 breaks the 14 stop barrier (14.1 according to DxO Mark) with 14 bit RAW files? I was under the impression there would be a hard 14.0 limit.
11-05-2010, 09:34 AM   #11
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,871
QuoteOriginally posted by Eruditass Quote
falconeye, can you explain how the K-5 breaks the 14 stop barrier (14.1 according to DxO Mark) with 14 bit RAW files? I was under the impression there would be a hard 14.0 limit.
The dynamic range of a single pixel for 1:1 SNR is 13.61 EV according to them.

The quantization noise with a 14 Bit ADC is 1:1 (or better -- would have to think about) for the 14th bit (at -14.0 EV), leaving out color and sensor response calibration (a DxO mystery). So, DxO measured DR is below the 14 Bit "threshold".

However, both 1:1 SNR (0 dB) and spatial Nyquist frequency are arbitrary points of reference. DxO uses the 0 dB point indeed but the spatial frequency of an 8 MP image. DR increases when you downscale an image as does the image's effective bit depth. You get signal in the 15th and 16th bit by downscaling. Every reduction to 50% adds one bit to the true signal. So, Photoshop with its 15 Bit TIFF engine starts to destroy information if you downscale an ISO 80 K-5 image to 2.3 MPixel or below within Photoshop.

Last edited by falconeye; 11-05-2010 at 09:44 AM.
11-05-2010, 10:05 AM   #12
Veteran Member
Ivan Glisin's Avatar

Join Date: Nov 2006
Location: Belgrade
Posts: 656
QuoteOriginally posted by Olivier Quote
It would have been properly exposed for ISO 6400, but the picture was taken at ISO 100, 6 stops underexposed. It was then over-exposed 8 stops in Lightroom to correct for the under-exposure (and then some).
QuoteOriginally posted by Olivier Quote
The histogram at the top right is rather sparse and I think the color patterns visible in the center image (100% crop) are due to that.
If I am not mistaken, no matter how many bits you have you can always severely underexpose then stretch data and get sparse histogram as a result. Each additional bit added would require one stop of additional underexposure to reproduce the histogram you got. So even with 16 bits shooting 10 stops underexposed, then push-processing by 10 stops should produce the same result as in your example.

Simply put, at ISO 100 you have instructed your camera to sample full expected well capacity range. Since you underexposed you've basically wasted lots of bits sampling mostly darkness. (My guess is that only lower 5 or 6 bits were used for storing useful tonal information, having remaining higher 9 or 8 bits (respectively) containing only zeroes.)

Now, the story should be completely different at ISO 6400 where camera expects different well capacity range and will sample accordingly, "slicing" electron readouts assuming smaller steps.

Try to compare ISO 6400 shot and underexposed ISO 100 stretched to ISO 6400 equivalent and see what happens. I am not expecting to see sparse histogram with spikes on an ISO 6400 native image.
11-05-2010, 10:14 AM   #13
Veteran Member
Ivan Glisin's Avatar

Join Date: Nov 2006
Location: Belgrade
Posts: 656
QuoteOriginally posted by Ivan Glisin Quote
Try to compare ISO 6400 shot and underexposed ISO 100 stretched to ISO 6400 equivalent and see what happens. I am not expecting to see sparse histogram with spikes on an ISO 6400 native image.
One more thought: K-5 has a base ISO range of ISO 80 to ISO 12800 if I remember well. ISO 25600 and ISO 51200 are extended range, I presume software processed. That means extended ISO images should start showing sparse histograms. Can someone check this? (Just remember: no exposure correction in PP.)

(I'd like to check this, but there is no extended ISO range on K10D with presumably different processing compared to base ISO range.)
11-05-2010, 10:30 AM   #14
Junior Member




Join Date: Jan 2008
Location: Portland, OR
Photos: Albums
Posts: 25
Original Poster
QuoteOriginally posted by Ivan Glisin Quote
Now, the story should be completely different at ISO 6400 where camera expects different well capacity range and will sample accordingly, "slicing" electron readouts assuming smaller steps.
Yes, I completely agree.

The point of this experiment was to stretch the histogram as far as possible by under-exposing an ISO 100 shot and then boosting values in Lightroom. I was wondering if the noise would be low enough to see banding due to 14-bit quantization. There is certainly some type of banding (especially visible in the ICE sticker in the lower right), but Falconeye seems to think it is due to some other effect.
11-05-2010, 10:41 AM   #15
Veteran Member
dgaies's Avatar

Join Date: Oct 2009
Location: Maryland / Washington DC
Posts: 3,917
QuoteOriginally posted by Ivan Glisin Quote
One more thought: K-5 has a base ISO range of ISO 80 to ISO 12800 if I remember well. ISO 25600 and ISO 51200 are extended range, I presume software processed. That means extended ISO images should start showing sparse histograms. Can someone check this? (Just remember: no exposure correction in PP.)

(I'd like to check this, but there is no extended ISO range on K10D with presumably different processing compared to base ISO range.)
I don't believe this is accurate; the histograms for 25600 and 51200 aren't compromised in this way.
Reply

Bookmarks
  • Submit Thread to Facebook Facebook
  • Submit Thread to Twitter Twitter
  • Submit Thread to Digg Digg
Tags - Make this thread easier to find by adding keywords to it!
bits, camera, dslr, gradient, iso, k-5, k-5 ii, k-5 iis, k5, pentax k-5
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
Improving K7 dynamic range Tony3d Pentax DSLR Discussion 23 10-12-2010 03:27 PM
Dynamic Range on the K-x Shahmatt Pentax DSLR Discussion 33 04-22-2010 03:51 PM
Dynamic Range Rene` Pentax DSLR Discussion 6 09-25-2008 07:46 AM
K10D dynamic range drwho9437 Pentax DSLR Discussion 15 07-15-2008 02:47 PM
Dynamic Range Vlad Photographic Technique 5 06-01-2008 10:40 PM



All times are GMT -7. The time now is 04:48 AM. | See also: NikonForums.com, CanonForums.com part of our network of photo forums!
  • Red (Default)
  • Green
  • Gray
  • Dark
  • Dark Yellow
  • Dark Blue
  • Old Red
  • Old Green
  • Old Gray
  • Dial-Up Style
Hello! It's great to see you back on the forum! Have you considered joining the community?
register
Creating a FREE ACCOUNT takes under a minute, removes ads, and lets you post! [Dismiss]
Top