Lumolabs: K-x vs. K-7 and the dynamic range champion of 2009

falconeye · 10-30-2009, 06:54 PM

Originally posted by jamesm007

The K-x does use RAW NR starting at ISO3200 as the K-7 this NR is twice as strong as used on the K20D at ISO3200 and probably much stronger than the K-7. This of course does not answer why your ISO1600 pic from the K-7 came out the winner. Try the test at ISO3200 and see what happens. The K-x does measure high DR but its also using much stronger NR.

Is this a comment directed at me?

What ISO1600 pic do you mean?

Where do you know the K-x raw NR from?

My test reveals the exact magnitude (of raw NR at ISO 3200 and higher, for K-x and K-7). Are there other sources for it?

However, I wouldn't say the K-x NR is stronger than K-7. Because of a flatter SNR slope, noise at ISO 1600 and 3200 turn out to be about the same, for K-x, where K-7 has stronger noise at 3200 than at 1600. But this is due to the steeper curve, not the stronger NR. The raw NR gains about 2.5 dB, the same for both cameras.

This gain doesn't seem to apply to the dynamic range though. Because shadows at low ISO are left alone.

Originally posted by georgweb

I can only imagine how much work this was

It's like a mountain climb. Once you bite, you must continue to the summit. My respect for DxO has increased, actually

Originally posted by georgweb

understand you are doing this with tungsten light and shutter speeds down to 1/15 or 1/10. This would comply with my needs (stage and social stuff without flash) and I just want to ask you whether your high ISO testing includes those settings and is therefore represented adequately in your charts.

I did the test as published at 2900°K tungsten, f/4.5 and 1/15s for ISO 100. I decreased shutter for higher ISO. Just today I retested. One variation was to keep shutter at 1/15 and decrease aperture down to f/22. Both SNR curves lie on top of each other. So, I could confirm that short shutter speeds below 1/10s don't differ in noise (I explained in my blog why this is to be expected). So yes, my results should apply to your shooting style.

Originally posted by jamesm007

The trick is the RAW converter its needs to be a neutral
[...]
GBG also writes his own raw converters he has some serious sensor knowledge.

Yes, GBG deserves our respect. BTW, he did write a RAW file reader but no raw converter. He can manipulate dot data but not create an RGB linear image. He didn't do the demosaicing which is the hard part.

A raw converter cannot be neutral. A raw converter creates 3N color channels from N raw channels. This implies creating cross-correlation between neighboring color channels. By varying amount, I do agree. But never zero. It isn't possible mathematically. At 1/4 the resolution, yes. But not at full resolution.

This is why inspecting the standard deviation of rwa dots BEFORE any raw conversion is interesting. I didn't do this though. DxO may do exactly this, don't know though. But interestingly, they don't measure chroma noise ...

rawr · 10-30-2009, 07:23 PM

jamesm007, I know about Raw Therapee. But where you see coarse NR at 3200 in the K-x, I can only see very fine grained texture with very little NR.

So we amateurs could go around and around on this, based on our perceptions of what is going on within little 200x200 pixel image blocks on screen under 'Detail' view in RT.

Let's just wait and watch as falconeye, DXOLabs and others with a more scientific approach (and the time/resources to do actual lab measurements) tackle these interesting technical issues.

You may also have to accept at some point that, although the K20D is a great camera, it doesn't represent the acme of camera technology and APS-C sensor design. Others - Nikon, Pentax, Sony, Canon - could have legitimately improved on it.

tr13 · 10-30-2009, 07:54 PM

I am not a big fan of such threads.
As I found that this comparisons have no real point.
I prefer camera with better viewfinder, more buttons, more high res screen and most importantly - best handling, to one that have good high ISO performance.
Worse part here is that we see same outdated 12Mp sensor with same noise reduction functions. In reality I belive that ALL of cameras using it have noise reduction applied from base ISO. And Samsung sensor have much worse noise reduction library, this is main caouse of higher noise.

jamesm007 · 10-30-2009, 10:58 PM

Originally posted by rawr

jamesm007, I know about Raw Therapee. But where you see coarse NR at 3200 in the K-x, I can only see very fine grained texture with very little NR.

So we amateurs could go around and around on this, based on our perceptions of what is going on within little 200x200 pixel image blocks on screen under 'Detail' view in RT.

Let's just wait and watch as falconeye, DXOLabs and others with a more scientific approach (and the time/resources to do actual lab measurements) tackle these interesting technical issues.

You may also have to accept at some point that, although the K20D is a great camera, it doesn't represent the acme of camera technology and APS-C sensor design. Others - Nikon, Pentax, Sony, Canon - could have legitimately improved on it.

I present what I find to Gordon B Good who with all do respect is probably the most knowledgeable sensor scientist that I know so far... . In respect to a neutral RAW converter I mean one that does not add NR as many do thats all., Until shown other-wise GBG word is finale for me including over DXO. He said he believed something was happening at low ISO with the D90 I was able with pics from my K20D help prove this to be the case NR at base ISO. Now as for the K-x I presented it to GBG and he measured with his tools, not his eyes!

"the K-x does apply raw Noise Reduction (NR) to the K-x raw data for ISO's of 3200 and up, and I measure from the French contributor's posted raw files that this raw NR is a little bit heavy handed to reduce the noise by a factor or two in standard deviation. I don't believe that any NR is applied to the raw data for ISO's of 1600 and lower. This NR applied by default (can't be turned off) by the K-x to high ISO raw data looks to be about twice as strong as that applied to the raw data of the K20D for ISO's of 1600 and above, as in the K20D only reducing the standard deviation of the noise by a about a factor of about 0.7 instead of by a factor of 0.5 for the K-x. Of course, it is this fairly heavy handed raw NR that makes the ISO 6400 and 12800 reasonably usable without further application of NR during JPEG processing"

This per Gordon B Good, not me.

Now overall the Sony sensor has better DR and high ISO DR, noise can get confusing as if you measure it at 18% the Samsung sensor will come out lower or with more detail, if you measure the noise at a lower light level the Sony will have the least noise. Its hard separating whats happening in the image engines (PRIME EXPEED) from what the sensor is giving it, this Gordon does well with his tools.

I think the Samsung K20D sensor was great and under rated when it came out but now from what I have seen the Canon 7D may have the best overall APS sensor out, depending on how much noise it does have at lower ISOs. The D300 or Sony 12mp sensor has always been great the differences were in the image engines and Pentax has a great approach making the K-x a wonderful high ISO camera in that you don't have to worry about tinted corners, applying NR, all the work is done for you. Just to make clear I know full well what camera does what and I am not some K20D super fanboy, but... I do call things how I see them and ask am I seeing right, in this one case I did see right. But what does it mean?

Last edited by jamesm007; 10-30-2009 at 11:20 PM.

ogl · 10-31-2009, 05:57 AM

I've played with RAW from K-x. It seems to me - K-x has strange yellow, red and green.
To correct the colour of each photo in convertor is rather lazy for me.
Yellow has green cast, red is with raspberry pink cast, green is a bit acid green.
The colour are not well-balanced and I can say it's rather pops colours.
With higher ISO - more acid. The jpeg from Raw and in-camera jpeg are rather similiar.
Like in Olympus.

jamesm007 · 10-31-2009, 10:13 AM

Originally posted by falconeye

Is this a comment directed at me?

What ISO1600 pic do you mean?

Where do you know the K-x raw NR from?

My test reveals the exact magnitude (of raw NR at ISO 3200 and higher, for K-x and K-7). Are there other sources for it?

However, I wouldn't say the K-x NR is stronger than K-7. Because of a flatter SNR slope, noise at ISO 1600 and 3200 turn out to be about the same, for K-x, where K-7 has stronger noise at 3200 than at 1600. But this is due to the steeper curve, not the stronger NR. The raw NR gains about 2.5 dB, the same for both cameras.

This gain doesn't seem to apply to the dynamic range though. Because shadows at low ISO are left alone.

It's like a mountain climb. Once you bite, you must continue to the summit. My respect for DxO has increased, actually

I did the test as published at 2900°K tungsten, f/4.5 and 1/15s for ISO 100. I decreased shutter for higher ISO. Just today I retested. One variation was to keep shutter at 1/15 and decrease aperture down to f/22. Both SNR curves lie on top of each other. So, I could confirm that short shutter speeds below 1/10s don't differ in noise (I explained in my blog why this is to be expected). So yes, my results should apply to your shooting style.

Yes, GBG deserves our respect. BTW, he did write a RAW file reader but no raw converter. He can manipulate dot data but not create an RGB linear image. He didn't do the demosaicing which is the hard part.

A raw converter cannot be neutral. A raw converter creates 3N color channels from N raw channels. This implies creating cross-correlation between neighboring color channels. By varying amount, I do agree. But never zero. It isn't possible mathematically. At 1/4 the resolution, yes. But not at full resolution.

This is why inspecting the standard deviation of rwa dots BEFORE any raw conversion is interesting. I didn't do this though. DxO may do exactly this, don't know though. But interestingly, they don't measure chroma noise ...

No comment was directed at you. For RAW converters I am saying one that does not apply any NR.

jamesm007 · 10-31-2009, 10:33 AM

Originally posted by tr13

I am not a big fan of such threads.
As I found that this comparisons have no real point.
I prefer camera with better viewfinder, more buttons, more high res screen and most importantly - best handling, to one that have good high ISO performance.
Worse part here is that we see same outdated 12Mp sensor with same noise reduction functions. In reality I believe that ALL of cameras using it have noise reduction applied from base ISO. And Samsung sensor have much worse noise reduction library, this is main cause of higher noise.

I agree, you know as well that NR is applied to the Sony 12Mp sensor at base ISO, but I think only in Nikons and maybe Sonys dSLR not Pentax, Pentax uses strong NR at higher ISO, what do you think? I value your input

as you know them at a very low level is Pentax doing it at ISO200 as well?

As for the Samsung its Pentax library, I think they just had a hand off approach to the Samsung as it does not use NR until ISO1600 (K20D) and then its a small amount. We now see it was had less noise when it came out over the D300. As for the K-7 its fine for many. But time is passing by and so is this sensor. But I will keep my K20D for now until a serious breakthrough comes in sensor technology and or for reasons as you state.

Back to the Sony. Would Nikon Sony's biggest customer besides itself get less than the best sensors. Did Sony give Pentax the best ver of that old 12Mp CMOS sensor and made it clean up to ISO6400 and betters the D5000, I don't think so, its all in the image engines as you know PRIME, EXPEED... those microprocessor do all kinds of things to the data coming off the sensor thats for sure.

falconeye · 10-31-2009, 11:03 AM

Originally posted by jamesm007

For RAW converters I am saying one that does not apply any NR.

Yes, I was understanding this. Any kind of Bayer matrix demosaicing however must use luminosity values from at least four raw gray dots. This results in a mild form of noise reduction and mild "Bayer-typical" softness at the pixel level, even without an antialias filter. It cannot be avoided. It may not be called NR.

tr13 · 10-31-2009, 09:23 PM

I believe that next breakthrough will be in software part, and not sensor only hardware.
I bet on new deconvoluton and demosaicing engines that allow to practically remove antialiasing filter blurring. This require manufacturing of special chips with hightly special parts that can do math intensive operations very fast.
Also special large map coding individual pixel behaviour could be used. If you know how each pixel behaves you could reduce noise significantly.

jeffkrol · 10-31-2009, 09:37 PM

Originally posted by falconeye

Yes, GBG deserves our respect. BTW, he did write a RAW file reader but no raw converter. He can manipulate dot data but not create an RGB linear image. He didn't do the demosaicing which is the hard part.

Not 100% sure but I believe he did make one of his own. Emil Martinec is also developing one and I may be confusing him w/ GBG but again I believe he did one for himself. Just an FYI..
Ok, found this:
http://forums.dpreview.com/forums/read.asp?forum=1036&message=32734624&q=gor...converter&qf=m
This was a converter to reduce the magenta cast but it is a converter, demosaic and all...

Hmmm.. it may just correct the numerical RAW data so I may be mistaken here. No time to really check..

falconeye · 11-01-2009, 11:20 AM

Originally posted by tr13

I bet on new deconvoluton and demosaicing engines that allow to practically remove antialiasing filter blurring.

I had a deep look into deconvolution and actually use it regularly on some of my out-of-focus images.

In general, I do agree with your expectation.

However, noise and deconvolution heavily counteract each other. We have developed our own denoise algorithm which we believe is very good. But the same applies again. Where an image is full of detail noise cannot be removed effectively (that is, w/o destryoing the detail).

There are some fundamental theorems of information theory at work. Nothing a fast processor can help with

But I do agree, at least the technically feasible could move soon into firmware

My favorite is multi-frame stacking all at the same time dramatically removing noise, increasing dynamic range and increase resolution.

falconeye · 11-01-2009, 07:07 PM

I thought this thread would be more fun with actual photos

So, I shot a typical high dynamic range scene with the K-x: a sunshine scene from within an inside room sitting in the shadow.

The naked eye had trouble to see that the window frame has blue color rather than black. Look at the magazines in the lower left corner in the unprocessed original ...

I append two versions of the photo:

- first, a standard-processed version showing the dynamic range
- second, a post-processed version where shadows have been pushed but NO noise reduction was applied (lightroom default settings).

I pushed the shadows using photo-shops "convert 32 to 8 bit" dialog, method "local changes" and standard parameters (16 px and 0.5 threshold, linear levels for 16!! EV).

Click on the images for 1920x1275 Pixel versions (the reduction to 2.5 Mp reduces noise but IMHO even at full size noise is not an issue).

Enjoy.

Autumn door (raw version)

Autumn door

LeDave · 11-01-2009, 07:10 PM

Wow nice pp'ing falconeye. Whenever I try to PP out a picture that dark and try to brighten it up, noise levels show like hell and even Noise Ninja can't help me with it. How did you manage to get it so clean?

falconeye · 11-01-2009, 07:48 PM

Originally posted by LeDave

How did you manage to get it so clean?

As I said, no noise reduction except from downsizing to 1920px (noise reduction could be used to clean it up further).

I already posted lab results for outstanding dynamic range in K-x images and I only wanted to demonstrate that it shows in real life as well.

High dynamic range means clean pushed shadows. No magic in post processing here

J3ff · 11-01-2009, 08:10 PM

Originally posted by falconeye

As I said, no noise reduction except from downsizing to 1920px (noise reduction could be used to clean it up further).

I already posted lab results for outstanding dynamic range in K-x images and I only wanted to demonstrate that it shows in real life as well.

High dynamic range means clean pushed shadows. No magic in post processing here

What does this mean to newbies?

You took both pictures with a Kx and somehow with settings made it lighter ?

10-31-2009, 09:37 PM	#55
jeffkrol Veteran Member Join Date: Sep 2006 Location: Wisconsin USA Photos: Gallery Posts: 8,434	Originally posted by falconeye Yes, GBG deserves our respect. BTW, he did write a RAW file reader but no raw converter. He can manipulate dot data but not create an RGB linear image. He didn't do the demosaicing which is the hard part. Not 100% sure but I believe he did make one of his own. Emil Martinec is also developing one and I may be confusing him w/ GBG but again I believe he did one for himself. Just an FYI.. Ok, found this: http://forums.dpreview.com/forums/read.asp?forum=1036&message=32734624&q=gor...converter&qf=m This was a converter to reduce the magenta cast but it is a converter, demosaic and all... Hmmm.. it may just correct the numerical RAW data so I may be mistaken here. No time to really check.. Last edited by jeffkrol; 10-31-2009 at 09:54 PM.

11-01-2009, 07:07 PM	#57
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Sample photos I thought this thread would be more fun with actual photos So, I shot a typical high dynamic range scene with the K-x: a sunshine scene from within an inside room sitting in the shadow. The naked eye had trouble to see that the window frame has blue color rather than black. Look at the magazines in the lower left corner in the unprocessed original ... I append two versions of the photo: - first, a standard-processed version showing the dynamic range - second, a post-processed version where shadows have been pushed but NO noise reduction was applied (lightroom default settings). I pushed the shadows using photo-shops "convert 32 to 8 bit" dialog, method "local changes" and standard parameters (16 px and 0.5 threshold, linear levels for 16!! EV). Click on the images for 1920x1275 Pixel versions (the reduction to 2.5 Mp reduces noise but IMHO even at full size noise is not an issue). Enjoy. *Autumn door (raw version)* *Autumn door* Last edited by falconeye; 11-01-2009 at 07:17 PM.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Dynamic Range on the K-x	Shahmatt	Pentax DSLR Discussion	33	04-22-2010 03:51 PM
Lumolabs: Nikon D700 vs. D5000 vs. Pentax K-x, Dynamic range and noise	falconeye	Pentax News and Rumors	12	12-18-2009 05:34 AM
dynamic range K20D	dafiryde	Pentax DSLR Discussion	4	01-08-2009 10:51 AM
Dynamic Range	Rene`	Pentax DSLR Discussion	6	09-25-2008 07:46 AM
Dynamic Range	Vlad	Photographic Technique	5	06-01-2008 10:40 PM

10-30-2009, 06:54 PM	#46
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by jamesm007 The K-x does use RAW NR starting at ISO3200 as the K-7 this NR is twice as strong as used on the K20D at ISO3200 and probably much stronger than the K-7. This of course does not answer why your ISO1600 pic from the K-7 came out the winner. Try the test at ISO3200 and see what happens. The K-x does measure high DR but its also using much stronger NR. Is this a comment directed at me? What ISO1600 pic do you mean? Where do you know the K-x raw NR from? My test reveals the exact magnitude (of raw NR at ISO 3200 and higher, for K-x and K-7). Are there other sources for it? However, I wouldn't say the K-x NR is stronger than K-7. Because of a flatter SNR slope, noise at ISO 1600 and 3200 turn out to be about the same, for K-x, where K-7 has stronger noise at 3200 than at 1600. But this is due to the steeper curve, not the stronger NR. The raw NR gains about 2.5 dB, the same for both cameras. This gain doesn't seem to apply to the dynamic range though. Because shadows at low ISO are left alone. Originally posted by georgweb I can only imagine how much work this was It's like a mountain climb. Once you bite, you must continue to the summit. My respect for DxO has increased, actually Originally posted by georgweb understand you are doing this with tungsten light and shutter speeds down to 1/15 or 1/10. This would comply with my needs (stage and social stuff without flash) and I just want to ask you whether your high ISO testing includes those settings and is therefore represented adequately in your charts. I did the test as published at 2900°K tungsten, f/4.5 and 1/15s for ISO 100. I decreased shutter for higher ISO. Just today I retested. One variation was to keep shutter at 1/15 and decrease aperture down to f/22. Both SNR curves lie on top of each other. So, I could confirm that short shutter speeds below 1/10s don't differ in noise (I explained in my blog why this is to be expected). So yes, my results should apply to your shooting style. Originally posted by jamesm007 The trick is the RAW converter its needs to be a neutral [...] GBG also writes his own raw converters he has some serious sensor knowledge. Yes, GBG deserves our respect. BTW, he did write a RAW file reader but no raw converter. He can manipulate dot data but not create an RGB linear image. He didn't do the demosaicing which is the hard part. A raw converter cannot be neutral. A raw converter creates 3N color channels from N raw channels. This implies creating cross-correlation between neighboring color channels. By varying amount, I do agree. But never zero. It isn't possible mathematically. At 1/4 the resolution, yes. But not at full resolution. This is why inspecting the standard deviation of rwa dots BEFORE any raw conversion is interesting. I didn't do this though. DxO may do exactly this, don't know though. But interestingly, they don't measure chroma noise ...

10-30-2009, 07:23 PM	#47
rawr Veteran Member Join Date: Jun 2009 Posts: 11,913	jamesm007, I know about Raw Therapee. But where you see coarse NR at 3200 in the K-x, I can only see very fine grained texture with very little NR. So we amateurs could go around and around on this, based on our perceptions of what is going on within little 200x200 pixel image blocks on screen under 'Detail' view in RT. Let's just wait and watch as falconeye, DXOLabs and others with a more scientific approach (and the time/resources to do actual lab measurements) tackle these interesting technical issues. You may also have to accept at some point that, although the K20D is a great camera, it doesn't represent the acme of camera technology and APS-C sensor design. Others - Nikon, Pentax, Sony, Canon - could have legitimately improved on it.

10-30-2009, 07:54 PM	#48
tr13 Veteran Member Join Date: Jul 2009 Location: Russia Posts: 343	I am not a big fan of such threads. As I found that this comparisons have no real point. I prefer camera with better viewfinder, more buttons, more high res screen and most importantly - best handling, to one that have good high ISO performance. Worse part here is that we see same outdated 12Mp sensor with same noise reduction functions. In reality I belive that ALL of cameras using it have noise reduction applied from base ISO. And Samsung sensor have much worse noise reduction library, this is main caouse of higher noise.

10-30-2009, 10:58 PM	#49
jamesm007 Veteran Member Join Date: Jul 2008 Location: Detroit MI, USA Posts: 508	Originally posted by rawr jamesm007, I know about Raw Therapee. But where you see coarse NR at 3200 in the K-x, I can only see very fine grained texture with very little NR. So we amateurs could go around and around on this, based on our perceptions of what is going on within little 200x200 pixel image blocks on screen under 'Detail' view in RT. Let's just wait and watch as falconeye, DXOLabs and others with a more scientific approach (and the time/resources to do actual lab measurements) tackle these interesting technical issues. You may also have to accept at some point that, although the K20D is a great camera, it doesn't represent the acme of camera technology and APS-C sensor design. Others - Nikon, Pentax, Sony, Canon - could have legitimately improved on it. I present what I find to Gordon B Good who with all do respect is probably the most knowledgeable sensor scientist that I know so far... . In respect to a neutral RAW converter I mean one that does not add NR as many do thats all., Until shown other-wise GBG word is finale for me including over DXO. He said he believed something was happening at low ISO with the D90 I was able with pics from my K20D help prove this to be the case NR at base ISO. Now as for the K-x I presented it to GBG and he measured with his tools, not his eyes! "the K-x does apply raw Noise Reduction (NR) to the K-x raw data for ISO's of 3200 and up, and I measure from the French contributor's posted raw files that this raw NR is a little bit heavy handed to reduce the noise by a factor or two in standard deviation. I don't believe that any NR is applied to the raw data for ISO's of 1600 and lower. This NR applied by default (can't be turned off) by the K-x to high ISO raw data looks to be about twice as strong as that applied to the raw data of the K20D for ISO's of 1600 and above, as in the K20D only reducing the standard deviation of the noise by a about a factor of about 0.7 instead of by a factor of 0.5 for the K-x. Of course, it is this fairly heavy handed raw NR that makes the ISO 6400 and 12800 reasonably usable without further application of NR during JPEG processing" This per Gordon B Good, not me. Now overall the Sony sensor has better DR and high ISO DR, noise can get confusing as if you measure it at 18% the Samsung sensor will come out lower or with more detail, if you measure the noise at a lower light level the Sony will have the least noise. Its hard separating whats happening in the image engines (PRIME EXPEED) from what the sensor is giving it, this Gordon does well with his tools. I think the Samsung K20D sensor was great and under rated when it came out but now from what I have seen the Canon 7D may have the best overall APS sensor out, depending on how much noise it does have at lower ISOs. The D300 or Sony 12mp sensor has always been great the differences were in the image engines and Pentax has a great approach making the K-x a wonderful high ISO camera in that you don't have to worry about tinted corners, applying NR, all the work is done for you. Just to make clear I know full well what camera does what and I am not some K20D super fanboy, but... I do call things how I see them and ask am I seeing right, in this one case I did see right. But what does it mean? Last edited by jamesm007; 10-30-2009 at 11:20 PM.

10-31-2009, 05:57 AM	#50
ogl Banned Join Date: Dec 2007 Location: Sankt Peterburg Photos: Gallery \| Albums Posts: 8,382	I've played with RAW from K-x. It seems to me - K-x has strange yellow, red and green. To correct the colour of each photo in convertor is rather lazy for me. Yellow has green cast, red is with raspberry pink cast, green is a bit acid green. The colour are not well-balanced and I can say it's rather pops colours. With higher ISO - more acid. The jpeg from Raw and in-camera jpeg are rather similiar. Like in Olympus.

10-31-2009, 10:13 AM	#51
jamesm007 Veteran Member Join Date: Jul 2008 Location: Detroit MI, USA Posts: 508	Originally posted by falconeye Is this a comment directed at me? What ISO1600 pic do you mean? Where do you know the K-x raw NR from? My test reveals the exact magnitude (of raw NR at ISO 3200 and higher, for K-x and K-7). Are there other sources for it? However, I wouldn't say the K-x NR is stronger than K-7. Because of a flatter SNR slope, noise at ISO 1600 and 3200 turn out to be about the same, for K-x, where K-7 has stronger noise at 3200 than at 1600. But this is due to the steeper curve, not the stronger NR. The raw NR gains about 2.5 dB, the same for both cameras. This gain doesn't seem to apply to the dynamic range though. Because shadows at low ISO are left alone. It's like a mountain climb. Once you bite, you must continue to the summit. My respect for DxO has increased, actually I did the test as published at 2900°K tungsten, f/4.5 and 1/15s for ISO 100. I decreased shutter for higher ISO. Just today I retested. One variation was to keep shutter at 1/15 and decrease aperture down to f/22. Both SNR curves lie on top of each other. So, I could confirm that short shutter speeds below 1/10s don't differ in noise (I explained in my blog why this is to be expected). So yes, my results should apply to your shooting style. Yes, GBG deserves our respect. BTW, he did write a RAW file reader but no raw converter. He can manipulate dot data but not create an RGB linear image. He didn't do the demosaicing which is the hard part. A raw converter cannot be neutral. A raw converter creates 3N color channels from N raw channels. This implies creating cross-correlation between neighboring color channels. By varying amount, I do agree. But never zero. It isn't possible mathematically. At 1/4 the resolution, yes. But not at full resolution. This is why inspecting the standard deviation of rwa dots BEFORE any raw conversion is interesting. I didn't do this though. DxO may do exactly this, don't know though. But interestingly, they don't measure chroma noise ... No comment was directed at you. For RAW converters I am saying one that does not apply any NR.

10-31-2009, 10:33 AM	#52
jamesm007 Veteran Member Join Date: Jul 2008 Location: Detroit MI, USA Posts: 508	Originally posted by tr13 I am not a big fan of such threads. As I found that this comparisons have no real point. I prefer camera with better viewfinder, more buttons, more high res screen and most importantly - best handling, to one that have good high ISO performance. Worse part here is that we see same outdated 12Mp sensor with same noise reduction functions. In reality I believe that ALL of cameras using it have noise reduction applied from base ISO. And Samsung sensor have much worse noise reduction library, this is main cause of higher noise. I agree, you know as well that NR is applied to the Sony 12Mp sensor at base ISO, but I think only in Nikons and maybe Sonys dSLR not Pentax, Pentax uses strong NR at higher ISO, what do you think? I value your input as you know them at a very low level is Pentax doing it at ISO200 as well? As for the Samsung its Pentax library, I think they just had a hand off approach to the Samsung as it does not use NR until ISO1600 (K20D) and then its a small amount. We now see it was had less noise when it came out over the D300. As for the K-7 its fine for many. But time is passing by and so is this sensor. But I will keep my K20D for now until a serious breakthrough comes in sensor technology and or for reasons as you state. Back to the Sony. Would Nikon Sony's biggest customer besides itself get less than the best sensors. Did Sony give Pentax the best ver of that old 12Mp CMOS sensor and made it clean up to ISO6400 and betters the D5000, I don't think so, its all in the image engines as you know PRIME, EXPEED... those microprocessor do all kinds of things to the data coming off the sensor thats for sure.

10-31-2009, 11:03 AM	#53
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by jamesm007 For RAW converters I am saying one that does not apply any NR. Yes, I was understanding this. Any kind of Bayer matrix demosaicing however must use luminosity values from at least four raw gray dots. This results in a mild form of noise reduction and mild "Bayer-typical" softness at the pixel level, even without an antialias filter. It cannot be avoided. It may not be called NR.

10-31-2009, 09:23 PM	#54
tr13 Veteran Member Join Date: Jul 2009 Location: Russia Posts: 343	I believe that next breakthrough will be in software part, and not sensor only hardware. I bet on new deconvoluton and demosaicing engines that allow to practically remove antialiasing filter blurring. This require manufacturing of special chips with hightly special parts that can do math intensive operations very fast. Also special large map coding individual pixel behaviour could be used. If you know how each pixel behaves you could reduce noise significantly.

11-01-2009, 11:20 AM	#56
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by tr13 I bet on new deconvoluton and demosaicing engines that allow to practically remove antialiasing filter blurring. I had a deep look into deconvolution and actually use it regularly on some of my out-of-focus images. In general, I do agree with your expectation. However, noise and deconvolution heavily counteract each other. We have developed our own denoise algorithm which we believe is very good. But the same applies again. Where an image is full of detail noise cannot be removed effectively (that is, w/o destryoing the detail). There are some fundamental theorems of information theory at work. Nothing a fast processor can help with But I do agree, at least the technically feasible could move soon into firmware My favorite is multi-frame stacking all at the same time dramatically removing noise, increasing dynamic range and increase resolution.

Thread Tools	Search this Thread
	Search this Thread: Advanced Search

11-01-2009, 07:10 PM	#58
LeDave Veteran Member Join Date: Jun 2009 Location: Minneapolis - St. Paul Photos: Gallery Posts: 3,067	Wow nice pp'ing falconeye. Whenever I try to PP out a picture that dark and try to brighten it up, noise levels show like hell and even Noise Ninja can't help me with it. How did you manage to get it so clean?

11-01-2009, 07:48 PM	#59
falconeye Veteran Member Join Date: Jan 2008 Location: Munich, Alps, Germany Photos: Gallery Posts: 6,871 Original Poster	Originally posted by LeDave How did you manage to get it so clean? As I said, no noise reduction except from downsizing to 1920px (noise reduction could be used to clean it up further). I already posted lab results for outstanding dynamic range in K-x images and I only wanted to demonstrate that it shows in real life as well. High dynamic range means clean pushed shadows. No magic in post processing here

11-01-2009, 08:10 PM	#60
J3ff Inactive Account Join Date: Oct 2009 Posts: 9	Originally posted by falconeye As I said, no noise reduction except from downsizing to 1920px (noise reduction could be used to clean it up further). I already posted lab results for outstanding dynamic range in K-x images and I only wanted to demonstrate that it shows in real life as well. High dynamic range means clean pushed shadows. No magic in post processing here What does this mean to newbies? You took both pictures with a Kx and somehow with settings made it lighter ?