Analysis: K-7 DNG raw Beta 0.2...

Started May 29, 2009 | Discussions
GordonBGood Veteran Member • Posts: 6,311
Analysis: K-7 DNG raw Beta 0.2...

Some of you may have noted that Imaging Resource (www.imaging-resource.com) have posted a link to the German Pentax website where two DNG raw images from a Beta 0.2 camera are posted (and thus presumably sanctioned by Pentax?). I downloaded these two files (a wide and a telephoto of a church) and can make the following statements after analysis, with the qualifications as stated:

The following observations are very unlikely to change between the beta version and the production versions:

1) These files confirm JohnCPentax's telling us that DNG's from the K-7 are now losslessly compressed. The compression scheme used is not the multi-tiled scheme used by the Adobe DNG Converter application but is compressed as one large block, both of which schemes fit into the DNG compression specification.

2) The embedded colour gamut matrix in the DNG files is again slightly different than the K20D, which was different than anything before it. The chromaticity endpoints as in these beta files indicate that the camera may tend to produce slightly more saturated reds, quite a bit less saturated greens, and about the same blues as the K20D. Pentax colour is generally quite accurate, so this is unlikely to be a problem, although the reds may tend to clip when processed to sRGB or AdobeRGB a little more.

3) The layout of the sensor is very similar to that of the K20D but Pentax no longer encode the completely garbage values on the right side as for the K20D (likely to reduce file size) and now seem to have two rows on the top as well as the 12 on the bottom of the sensor that are masked to light and not part of the images. Also, the dark masked-to-light area on the right side of the sensor (in landscape mode) is now narrower at only 52 columns rather than about 96 columns for the K20D, indicating that the new PRIME II imaging processor may now use a new algorithm to determine the black offset compensation and apply it to the light-exposed photosite rows. Basically all this tells us is that it indeed is a redesigned sensor as to layout, that it may be used somewhat differently as to black compensation, and that it is likely still scanned horizontally as the K20D was, as well as having the required 4 channel readout for the required increased scanning speed.

4) I push processed and contrasted the raw data for flat textured dark patches to extreme levels to see the RGB noise" speckles" and could detect no signs of Vertical/Horizontal Pattern Noise (VPN/HPN) in the images in areas of dark flat texture. Patterns may appear to some extent for extremely high ISO images as it is said to do for the K20D for some cameras, but I could not detect it pushing these low ISO images.

Which brings me to the limitations of testing these images, as both were shot at ISO 200 with Dynamic Range Expansion (DRE) set to On according to the Manufacturer Notes in the EXIF data (PhotoMe); this means that the images were actually shot at an ISO 100 internally. Further, one should not consider the noise found here as final in that all of the firmware, the sensor, and the electronics, may be at a pre-production state and may be improved before the production cameras are released. One would think that production cameras would be better than this and not worse (hopefully). What I found was:

That the black read noise (and all other noises) of this Beta 0.2 camera was very similar to that of the K20D at ISO 100, which was a little disappointing (although still hopeful that production versions will be better).

If the black read noise is not reduced below this level, the Dynamic Range (DR) of the K-7 will not be any better than that of the K20D, nor will the high ISO capability be improved, and if there are any improvements seen in the JPEG outputs it will likely be due to stronger Noise Reduction (NR).

Now I know that a DR limitation does not bother many users who never "push process" their raw images or compress a wide DR into a narrower one for display on a contrast limited viewing device or paper, but unfortunately, if black read noise is not improved this is the one parameter related to image quality across the ISO range that the K-7 does not at least come close to matching for its major competition in the same price range. I still hope that Pentax may be able to improve this without NR in production cameras, but fear that it may be a limitation of the Samsung sensor design itself.

Just for your consideration while we wait for production samples.

Regards, GordonBGood

ozdean
ozdean Forum Pro • Posts: 26,217
Re: Analysis: K-7 DNG raw Beta 0.2...

Thanks for all your effort here Gordon - black read noise is a concern, we'll see.

Am I right in saying that the 10m sensor doesn't suffer this as much, it seems that way to me.
--
Regards Dean - Capturing Creation

 ozdean's gear list:ozdean's gear list
Pentax MX-1 Pentax K-5 IIs Pentax K-50 Pentax smc DA* 55mm F1.4 SDM Pentax smc DA 15mm F4 ED AL Limited +13 more
Jonas B Forum Pro • Posts: 14,596
Re: Analysis: K-7 DNG raw Beta 0.2...

Hi,

I very much appreciated your post. To me this was valuable information.

regards,

-- hide signature --

Jonas

Paul Hunt Contributing Member • Posts: 784
Re: Analysis: K-7 DNG raw Beta 0.2...

Gordon,

A lucid and dispassionate analysis. I always read and appreciate your contributions.

Paul

EnriqueBello Regular Member • Posts: 373
Re: Analysis: K-7 DNG raw Beta 0.2...

GordonBGood wrote:

Some of you may have noted that Imaging Resource
(www.imaging-resource.com) have posted a link to the German Pentax
website where two DNG raw images from a Beta 0.2 camera are posted
(and thus presumably sanctioned by Pentax?).

I can't find the links.

Thanks

-- hide signature --

Enrique

OP GordonBGood Veteran Member • Posts: 6,311
Re: Analysis: K-7 DNG raw Beta 0.2...

ozdean wrote:

Am I right in saying that the 10m sensor doesn't suffer this as much,
it seems that way to me.

Well, Dean, the problem with the Samsung CMOS sensor design hasn't been so much the sensel (the photosite) itself as compared to the 10 MP CCD sensors used in other designs such as the K10D and K200D but rather the electronics and especially the Variable Gain Amplifier (VGA) that Samsung (in their infinite wisdom) decided to implement on the sensor itself. This VGA actually added to the to the black read noise of the sensor of the K20D rather than reduce it as intended. If this circuit has not been removed or improved over that of the K20D, it won't matter how much Pentax work to "improve surface noise injection" in the circuit boards, as this noise is negligible as compared to the noise originating on the sensor.

Noises are added in quadrature, so for instance if the sensor noises combined are about 2.5 Digital Output Units (DOU's) and the noise that could be reduced due to surface circuit board injection was about a half DOU, then the total noise would be the root sum of the squares or about 2.55 DOU's. You can see that reducing the circuit noise has almost no effect on the combined black read noise.

As another example, if the actual photosite black read noise at ISO 100 is actually about 1 DOU as I suspect it is but the VGA noise is about 2.8 DOU's, then we can never get close to seeing the real photosite (sensel) black read noise until the noise of the VGA is greatly reduced or eliminated.

I am hoping that the results from the beta raw data I posted are just from preliminary designs and that the real source of the noise is known and is being improved. I also hope that the solution is not to just mask the noise with (more) Noise Reduction (NR), as is already done to some extent for high ISO's in the K20D, even to the raw data.

Regards, GordonBGood

Duplo Veteran Member • Posts: 6,660
Thank you and a question

hi Gordon,

Always a pleasure to read your posts and you seem to really devote a lot of time and energy on these very technical aspects of things.

I am looking forward to more analysis from you once you get your hands on actual RAW files.

Will you be doing any work on finding a uniWB for the K-7 or tone curve analysis for the embedded camera profiles?
--
Thomas

Nothing is foolproof to a sufficiently talented fool
http://main.duplophotography.com/

OP GordonBGood Veteran Member • Posts: 6,311
Re: Analysis: K-7 DNG raw Beta 0.2...

EnriqueBello wrote:

GordonBGood wrote:

Some of you may have noted that Imaging Resource
(www.imaging-resource.com) have posted a link to the German Pentax
website where two DNG raw images from a Beta 0.2 camera are posted
(and thus presumably sanctioned by Pentax?).

I can't find the links.

http://www.imaging-resource.com/NEWS/1243530291.html

Regards, GordonBGood

OP GordonBGood Veteran Member • Posts: 6,311
Re: Thank you and a question

Duplo wrote:

hi Gordon,

Will you be doing any work on finding a uniWB for the K-7 or tone
curve analysis for the embedded camera profiles?

Thomas, my interests are mainly in the camera characteristics as revealed by the raw image data and not to the JPEG processing done by the camera; thus I don't have that much interest in the Tone Response Curves (TRC's) for the developed JPEG's as per the embedded camera profiles. As well, these will be revealed by DPR's "Dynamic Range Tests" which primarily just determine these TRC's.

As to finding a uniWB for the camera, the gain values required are in the raw data tags for each WB preset but I can't really see the purpose to using these given that the camera histogram will be based on the actual WB used in the JPEG development. The camera histograms will never fully reflect the raw histogram unless the camera firmware is designed so that it does so, as colour space treatments also affect the clipping of colour channels.

Regards, GordonBGood

Duplo Veteran Member • Posts: 6,660
Re: Thank you and a question

GordonBGood wrote:

Duplo wrote:

hi Gordon,

Will you be doing any work on finding a uniWB for the K-7 or tone
curve analysis for the embedded camera profiles?

Thomas, my interests are mainly in the camera characteristics as
revealed by the raw image data and not to the JPEG processing done by
the camera; thus I don't have that much interest in the Tone Response
Curves (TRC's) for the developed JPEG's as per the embedded camera
profiles. As well, these will be revealed by DPR's "Dynamic Range
Tests" which primarily just determine these TRC's.

As to finding a uniWB for the camera, the gain values required are in
the raw data tags for each WB preset but I can't really see the
purpose to using these given that the camera histogram will be based
on the actual WB used in the JPEG development. The camera histograms
will never fully reflect the raw histogram unless the camera firmware
is designed so that it does so, as colour space treatments also
affect the clipping of colour channels.

I do know what you mean and appreciate your answer.

The trick with uniWB is that you get a lot closer to reflect what the sensor sees, and with a linear tone curve in camera, you get a decent representation of the actual RAW capture in the embedded JPG that the in camera histograms are based on.

True you have colour space limitations as well, but I have found them to be very minor and for field work the above two things are a decent tool for detecting actual colour chanel clipping in the field and gives a much better in camera representation of what happens in the RAW file.

Or that is the theory at least
--
Thomas

Nothing is foolproof to a sufficiently talented fool
http://main.duplophotography.com/

jamesm007 Veteran Member • Posts: 5,663
Re: Analysis: K-7 DNG raw Beta 0.2...

Thanks for the hard and interesting work. You know over six months ago I read a non-Pentax related article in which a Samsung engineer stated they have already improved the speed of CMOS chips; not speaking of Pentax or the K20 just Samsung in general. I don't think he knew he bascily told me that they did indeed increase the FPS for the new Pentax (?) I made a post on that comment.

But I am a little discouraged if your findings hold because this tells me Pentax/Samsung only saw FPS as the weakness of the new CMOS and did not really think black read noise was a problem. I have to admit not many complained about it and this may have had an affect of making Pentax think its a none issue for most. But I for one know its there and can easily be seen at iso100 versus the K10D at the same. And its disturbing to me because it limits DR at lower ISO.

Do you have any thoughts on why other than the obvious (increases noise) that Pentax did not increase the mega-pixels on the CMOS?

Also if true this is a marketing camera to please the complainers of slow/poor AF; and bring it up to par with it rivals, but! Not exceed its rivals in performance. Which all in all is not a bad deal and still does not reduce the K7s appeal too much.

-- hide signature --
 jamesm007's gear list:jamesm007's gear list
Pentax K20D Pentax K-5 Sigma 10-20mm F4-5.6 EX DC HSM Sigma 17-70mm F2.8-4 DC Macro OS HSM Pentax smc DA 18-55mm F3.5-5.6 AL WR +12 more
Kerusker
Kerusker Senior Member • Posts: 1,285
thanks Gordon, looking forward ...

Thanks for your analysis Gordon.
I shure won't miss any of your upcoming posts.

-- hide signature --

][.Kerusker (was 'no see')
we don't see that we don't see (eye's blind spot)

my albums: http://picasaweb.google.de/Kerusker

 Kerusker's gear list:Kerusker's gear list
Olympus Stylus 1 Sony RX10 IV Pentax K-5 II Pentax K-1 Pentax smc DA 15mm F4 ED AL Limited +1 more
Stomie Veteran Member • Posts: 3,119
Re: Analysis: K-7 DNG raw Beta 0.2...

Good, insightful & thorough work again as always Gordon. Thank you for sharing it here. It's very much appreciated.

Cheers :->
Stomie
--

Please note: all images displayed by me on this or any other site are copyright ©

Stomie Veteran Member • Posts: 3,119
Nice to see you around Jonas...

... I hope you are keeping well :->

Cheers
Stomie
--

Please note: all images displayed by me on this or any other site are copyright ©

jthommo Veteran Member • Posts: 3,735
Thanks Gordon

I hope the high iso performance is better in the production version...otherwise here's one lost sale.
--
Thommo
http://www.flickr.com/photos/jthommo101/

ozdean
ozdean Forum Pro • Posts: 26,217
Re: Analysis: K-7 DNG raw Beta 0.2...

Thanks Gordon for your explaination, let's hope for a positive result.
--
Regards Dean - Capturing Creation

 ozdean's gear list:ozdean's gear list
Pentax MX-1 Pentax K-5 IIs Pentax K-50 Pentax smc DA* 55mm F1.4 SDM Pentax smc DA 15mm F4 ED AL Limited +13 more
jackbw Senior Member • Posts: 1,702
Re: Analysis: K-7 DNG raw Beta 0.2...

Thank you for your analysis, Gordon. I hope John C. sees your post. I doubted you in the past about the K20D sensor. But history proved you correct; I will never doubt you again. Something that makes me curious is why DXO Mark rates the GX20 sensor (67.7) higher than the K20D sensor (65.4) when they are supposed to be the same. My gut tells me it could be sample variation, so I'm hoping we get better quality sensors from the K-7. I'll be looking out for your posts when the first production samples from the K-7 come out.
- - - - -
Roger

Stig Vidar Hovland Senior Member • Posts: 1,744
Re: Nice to see you around Jonas...

Stomie wrote:

... I hope you are keeping well :->

Yes, nice to see Jonas here. I need him to do some measurements of what whitebalance K-7 should be dialed in to for the RGB histogram to be correct for raw data I have used 5000K on my K10D for nearly three years now. Thanks Jonas !!

alanschamber Senior Member • Posts: 1,200
Re: Analysis: K-7 DNG raw Beta 0.2...

Maybe Pentax needed just to say this time ME TOO, and they would already win this match. Why do I say this?

If you remember a DPR blog article, I think it was R Butler who stated Pentax as the most complete and dedicated APS-C lens system of the market. So... what keeps people from buying Pentax? AF and fps were two of them. High ISO for me is more than excellent in the K20 (still, improvement is always welcome). So... Pentax, for me, had just to offer a camera that could keep up with the specs and AF, and they did it offering also a smaller body, and even probably better controls for most people (not that much for me... but still I can live without complains).

So, at the end of the day... if Pentax Marketing department makes it right this time, I think this camera could mean a relatively high market share for them.

And, I must say... it's not that I'm understanding the way Pentax see the market: make the lenses system, and the keep up with the camera, so if anyone wants to change brands, the lens system won't be a problem.

Alan.

jamesm007 wrote:

Thanks for the hard and interesting work. You know over six months
ago I read a non-Pentax related article in which a Samsung engineer
stated they have already improved the speed of CMOS chips; not
speaking of Pentax or the K20 just Samsung in general. I don't think
he knew he bascily told me that they did indeed increase the FPS for
the new Pentax (?) I made a post on that comment.

But I am a little discouraged if your findings hold because this
tells me Pentax/Samsung only saw FPS as the weakness of the new CMOS
and did not really think black read noise was a problem. I have to
admit not many complained about it and this may have had an affect of
making Pentax think its a none issue for most. But I for one know its
there and can easily be seen at iso100 versus the K10D at the same.
And its disturbing to me because it limits DR at lower ISO.

Do you have any thoughts on why other than the obvious (increases
noise) that Pentax did not increase the mega-pixels on the CMOS?

Also if true this is a marketing camera to please the complainers of
slow/poor AF; and bring it up to par with it rivals, but! Not exceed
its rivals in performance. Which all in all is not a bad deal and
still does not reduce the K7s appeal too much.

-- hide signature --

Progress is not possible without deviation from the norm - Frank Zappa
http://www.schamberalan.blogspot.com

alanschamber Senior Member • Posts: 1,200
Really interesting information

Thanks a lot... and hope to listen about any changes.

Alan.
--
Progress is not possible without deviation from the norm - Frank Zappa
http://www.schamberalan.blogspot.com

Keyboard shortcuts:
FForum MMy threads