Link to noise comparison A500

Specifically for the discussion of the A-mount DSLR range
Forum rules
No more than three images or three external links allowed in any post or reply. Please trim quotations and do not include images in quotes unless essential.
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Sun Aug 30, 2009 7:31 pm

David Kilpatrick wrote:Is what you desribe also what Nikon have been doing all along - something so simple that no-one bothered to think of it at Sony?

It doesn't seem like Nikon/Canon do anything like that. Those two rather play with 'optimizing' the colour matrix to minimize the noise at expense of colour fidelity. Nikon also uses rather heavy threshold filtering on raw in their D90 and especially in D5000 as obvious from the fact that 1px and 2px components of raw noise are far weaker than the lower freq noise components would suggest. The sensor itself in those two is actually nothing better than that in D300 which raw surely can be developed to even better S/N result than D5000/D90 since you can control the strength of that threshold filtering.

On the other hand, Panasonic and Olympus use exactly that kind of developing - separate for luma and chroma, and exactly into the Lab colour internally. But they adjust the 'bin' size for chroma depending on ISO, the higher ISO - the larger the bin. At lowest ISOs they seem to use an almost 'normal' demosaicing.
(E.g. an Lx3 obviously switches to chroma binning at ISO400, while at ISO80-100 it works more like demosaicing; yet the luma is still developed separately from chroma, and the Lab is the only colour space used internally.)

alphaomega
Viceroy
Posts: 1196
Joined: Mon Aug 18, 2008 11:20 pm

Re: Link to noise comparison A500

Unread postby alphaomega » Sun Aug 30, 2009 8:34 pm

Well, I musst admit that agorabasta is way above my basic level of understanding, but I must admit that I have been encouraged by the latest releases by Sony, in particular the A500/550 combo. They are showing they are serious, and whatever tricks they have up their sleeves, they are addressing the higher ISO noise problems. If they could just get out a replacement for the A700 retaining current functionality and perhaps using the new 14.2Mp sensor we might see something resembling a complete line-up short of sports professionals requirements. With a variety of DSLRs, some focusing on speed, others on low ISO max resolving power and yet others on high ISO performance in amongst beginners, enthusiasts and kind of Profs camera offerings, there might besometing for everybody - even Barry Fitzgerald. Time will show, but at least things are happening and progress is being made in the higher ISO stakes it would appear. They need to catch up on lenses as well, but things seem to accelerate a bit. If David's vision of a Sony/DxO relationship is correct, then we might even see a competent IDC at some stage. Keep dreaming. I do like some of the aspects of the A550 and I will trade in my A350 in the spring and acquire an A550 when it is fully supported RAW conversion wise - not just by IDC.

agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Tue Sep 01, 2009 1:59 pm

Ok, here's an example of an A900 ISO6400 no-NR raw image from imagingresource developed using such technique, separately for chroma and luma. Luma was developed with RGB multipliers 1-2-1 and biliear demosaicing, chroma also bilinear, with proper WB. On luma I did only some threshold filtering to cut out stray values; the chroma was given some threshold filtering and then a simple edge-sensitive blur, no chroma binning was used here.

The full image (try this link to download full sized image - http://lh6.ggpht.com/__bYe591qcSo/Sp0MA ... F_dual.jpg - but Picasa will not keep it live for long):

Image

A 100% crop:

Image

(Here's an "official' Picasa link - http://picasaweb.google.ru/lh/photo/mr4 ... directlink - not full size)

User avatar
bfitzgerald
Subsuming Vortex of Brilliance
Posts: 3684
Joined: Fri Sep 26, 2008 10:48 pm

Re: Link to noise comparison A500

Unread postby bfitzgerald » Tue Sep 01, 2009 2:25 pm

Sensor looks good, ISO 6400 might be usable. However 12800 won't be..still we shall see on that one.

Studio shots are not useful to low light shooters though, they give a best case scenario, when many of us look for a worst case one ;-)

agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Tue Sep 01, 2009 3:10 pm

Barry, that was an a900 file 8)

Yagil Henkin
Heirophant
Posts: 86
Joined: Tue Oct 07, 2008 8:29 pm

Agorabasta, could you explain How-to?

Unread postby Yagil Henkin » Tue Sep 01, 2009 6:51 pm

I'm not sure I understood HOW to do that double-developing thing, and it seems useful indeed.

agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Tue Sep 01, 2009 9:30 pm

That's a very simple thing.

First develop for the best BW image possible, use any kind of demosaicing that gives least noise, sway WB at will, whatever - just to get the best quality BW image; then apply some gentle noise filtering to keep the most detail.

Then develop once again, this time for colour, also choose the most suitable demosaicing only to get the cleanest colour regardless of luminance noise or precision. You may reduce the res about 4-fold without loosing much. Apply some heavy noise filtering, but try to retain the stronger colour edges intact.
(In that example I used edge-sensitive blur called 'Smart Blur' in CorelPP, other applications may call it differently. The colour image was developed full-size, then low-passed at 4px, then 100% USM at 4px, then the smart blur applied.)

Then combine the result using the colour channels from the second image for chrominance, and the BW from the first for luminance in some space like Lab or YIQ or any flavour of YCbCr.

All in all, it's not a very computationally intensive process - far less intensive computing than a decent 'regular' NR algorithm requires.
And the result is very suspiciously similar to a500 8)

Yagil Henkin
Heirophant
Posts: 86
Joined: Tue Oct 07, 2008 8:29 pm

Thank you very much!

Unread postby Yagil Henkin » Tue Sep 01, 2009 11:26 pm

Will try.

Vidgamer
Imperial Ambassador
Posts: 527
Joined: Wed Jun 11, 2008 3:37 am

Re: Link to noise comparison A500

Unread postby Vidgamer » Mon Sep 07, 2009 6:10 am

agorabasta wrote:....
Then combine the result using the colour channels from the second image for chrominance, and the BW from the first for luminance in some space like Lab or YIQ or any flavour of YCbCr.

All in all, it's not a very computationally intensive process - far less intensive computing than a decent 'regular' NR algorithm requires.
And the result is very suspiciously similar to a500 8)


Sounds interesting... I decided to try something similar on an A100 ISO 1600 photo. I tried a couple of different RAW converters and different techniques, but it doesn't seem to be working for me. I'm using Paint Shop Pro, and was trying to split into "HSL" in order to combine the luma from one with the chroma from the other. But it didn't seem to matter how much I mangled the color version to reduce the res or blur, the color was still splotchy. (PSP has a smart blur option that they call Edge Preserving Smooth.)

Better to use the normal PSP noise reduction, I think. However, when you're talking about in-camera processing, speed means a lot more....

agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Mon Sep 07, 2009 2:48 pm

Here's a comparison of ISO3200 crops borrowed from http://artaphot.ch/dslrs/202-jpg-low-li ... -dynax-7d-

Image

Here the a500 image is unchanged, just cropped somewhat.

The a700 luma was left intact, separated from original at the first step; then the a700 image processed for chroma as follows:

1. composite image - low-pass 100% at 4px
2. composite image - USM 100% at 4px
3. composite image - edge-sensitive threshold blur
4. chroma separated from the result of the above
5. processed chroma mixed with original luma

Then some channel mixer, saturation and WB adjustments to match the a500 colours somewhat.

If a700 had its internal NR set to 'low' instead of 'normal', the results would be far better (no green/purple colour noise).


Now if the a100 is considered, I don't have any representative samples of its ISO1600 raws. So I borrowed a JPEG image from http://www.dcresource.com/reviews/sony/ ... oosted.jpg

Here are processed crops of that image, the first one processed with my method, then below another crop processed with NeatImage autoprofiled defaults for chroma only

Image
Image


Larger images are here - picasaweb.google.com/lh/photo/LxkKgIl-ZVAJSQ1SLflBMQ?authkey=Gv1sRgCKfmo_PBl9HYIw&feat=directlink - my simple process; and here - picasaweb.google.com/lh/photo/oSUygx2Cn6mHvrkv9FesJg?authkey=Gv1sRgCKfmo_PBl9HYIw&feat=directlink - the one with NeatImage chroma NR. (you have to copy&paste the links due to this board limitations on number of links in a message :twisted: )
Not much difference if you ask me; plus my method masks the colour fringes

Vidgamer
Imperial Ambassador
Posts: 527
Joined: Wed Jun 11, 2008 3:37 am

Re: Link to noise comparison A500

Unread postby Vidgamer » Mon Sep 07, 2009 10:07 pm

Comparing the A100 results, it looks like a wash. Seems like I should be able to do better, so I tried again.

Ok, I made one high detail, noisy pic, and pulled the luma from that. Then I took a softer raw conversion for the color. First, I ran median filter of 5. Then, USM of radius 2 @ 100%. I first tried 4, but that really seemed to increase the noise again. Seems to defeat the purpose... Anyway, then did the Edge Preserving Smooth.

Merged with the other luma layer (from the "sharper" conversion)... and I get these yellow splotchy areas. The sharper luma seems lighter, so I added some contrast and merged again, but it still looks kinda bad. I think it's just that PaintShopPro works differently ...or certainly not as I'd expect. Hmm. (I have Photoshop Elements, but it has limitations; not sure if this technique will work there either.) Just to be sure, I went back to the original "color" file and re-extracted its luma, before the other changes, remerged, and still have a bit of slplotchiness, although not quite as bad.

At any rate... I see what you're saying, though. You can get virtually the same results as a complicated algorithm by some simple manipulation.

I do think that it's easy for people, especially techies, to overengineer things. Sometimes it pays to Keep It Simple.

agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Tue Sep 08, 2009 9:14 am

Vidgamer wrote:First, I ran median filter of 5. Then, USM of radius 2 @ 100%. I first tried 4, but that really seemed to increase the noise again. Seems to defeat the purpose... Anyway, then did the Edge Preserving Smooth.


Median is not exactly low-pass - that's the most important part here - median creates some low-intensity high-frequency artifacts that get then exaggerated in further processing.
Then the USM naturally increases the apparent colour noise, but it doesn't degrade the final outcome if you use a true low-pass because the 'smart blur' keeps only those chroma transitions that are spatially correlated with luma transitions if the blurring threshold is set correct (i.e. chroma-only transitions should universally fall below the threshold).

But I'd advise you try it all with the same software that I use - they have a free trial.

colorpenguin
Acolyte
Posts: 3
Joined: Fri Dec 26, 2008 1:33 am

Re: Link to noise comparison A500

Unread postby colorpenguin » Tue Sep 08, 2009 10:49 pm

agorabasta wrote:But I'd advise you try it all with the same software that I use - they have a free trial.


sorry if I missed an earlier mention, what software is that?
Carl

agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Link to noise comparison A500

Unread postby agorabasta » Wed Sep 09, 2009 6:41 am

colorpenguin wrote:sorry if I missed an earlier mention, what software is that?
Carl


That's Corel PHOTO-PAINT from CorelDRAW Graphics Suite X4

Mr_Canuck
Initiate
Posts: 30
Joined: Fri Jun 27, 2008 5:36 pm
Location: Canada

Does DXO solve all???

Unread postby Mr_Canuck » Fri Sep 11, 2009 9:21 pm

RAW conversion technical neophyte here (and an a850 "leaner" towards upgrading from my a700)....

I'm using Photoshop CS4 and Camera Raw for conversion with my a700, and have never used DXO. So a couple of questions...

1. Should I be using DXO instead?
2. If I'm shooting RAW and converting with DXO, is the (alleged) bad Sony JPG engine and perceived hi-iso noise issue with the a850/a900 a moot point because I could get all the NR goodness I want with DXO? Are you guys saying that a850 + DXO = 550 + internal engine (possibly DXO-based)???

Meaning, is my assumption true... that I can get every bit as good hi-iso performance out of an a850/a900 RAW file as I might out of a 5DMkII or D3s, and therefore that would eliminate my remaining concern about upgrading from my a700?
a850 | 28-135 | 70-300G | 20/2.8 | 35/2 | 50/2.8M | 100/02 | 200f2.8 | HVL-20FA | 3600HS | Border Collie X


Return to “Alpha A-mount System”

Who is online

Users browsing this forum: No registered users and 2 guests

cron