MFlenses - 数码时代的新生
MFLenses 2026

手动对焦镜头论坛

通过维修记录、实拍样片和图文丰富的档案帖,探索老镜头的世界。

通过维修记录、实拍样片和图文丰富的档案帖,探索老镜头的世界。
精选

论坛首页的合作伙伴展示位。

raw hardware artifacts from RGGB
查看上一个主题 :: 查看下一个主题  


Reply

I have some strange artifacts with 5DII
I try Bibble, C1, ACDsee, Breeze, pp with 0 sharpening and all give those jigged line & artifacts
I believe that the raw get hardware sharpening that it is not possible to avoid


100% crop, 0 sharpening, look over the eye, like already sharpened
contax 35-70, 200 iso


-------------------------------
edit: I found a solution, scroll down to read it Very Happy


最后由 poilu 于 周二 3月 23, 2010 3:56 pm 编辑;共编辑 1 次


Reply

and with DPP 3.8 ?
Do you have the same artifacts ?


Reply

Those don't look like sharpening artifacts as much as just aliasing. It looks like the 5DII has a weak AA filter, which I'd say is a good thing.

I'd try RPP (if you have access to a Mac) or dcraw, but I don't think that's sharpening.


Reply

Olivier 写道:
and with DPP 3.8 ?
Do you have the same artifacts ?

BreezeBrowser use Canon algo and have the same artifacts

n6mod 写道:
Those don't look like sharpening artifacts as much as just aliasing. It looks like the 5DII has a weak AA filter, which I'd say is a good thing.
I'd try RPP (if you have access to a Mac) or dcraw, but I don't think that's sharpening

I tried raw therapee who use dcraw and got same artifacts
I guess the AA filter is weak, all my lenses are very sharp at 100%


Reply

I believe there is applied some sharpening after (or during) the debayering process. Bayer mask consists of 25% blue, 25% red and 50% green, so the channels have different level of sharpness, which is probably corrected in some way...


Reply

[quote="no-X" so the channels have different level of sharpness, which is probably corrected in some way...[/quote]

Well, there's interpolation going on, and this is deeply related to the sort of aliasing you see. It's not *sharpening* per se.

Certainly these effects will disappear if you avoid interpolation by using the -h option to dcraw (or RPP's "half" mode), which averages the two green sensels and presents a group of four sensels as a single pixel.

And we're dangerously close to awakening the Foveon enthusiasts... Wink


Reply

n6mod 写道:
Certainly these effects will disappear if you avoid interpolation by using the -h option to dcraw

the -h option give the same result as resizing full image
left dcraw -h ... right C1 sized 50% nearest neighbor


as green channel seems affected, it can be result of bayer interpolation
I can live with some artifacts, I just hope Canon will not follow Sony way and alter raw with hardware denoising for marketing
left parrot 100% no sharpening, no problem


Reply

poilu 写道:

as green channel seems affected, it can be result of bayer interpolation


There's no interpolation with -h, just averaging the two values.
poilu 写道:

left parrot 100% no sharpening, no problem

But that's a monochrome target, basically.

Seems I was wrong that the artifacts went away with -h, in fact they moved around. I think that's still consistent with aliasing (fine detail at pixel scale).

I'll say this. That's one heck of a lens to expose aliasing on a 5DII!


Reply

poilu 写道:
I have some strange artifacts with 5DII
I try Bibble, C1, ACDsee, Breeze, pp with 0 sharpening and all give those jigged line & artifacts
I believe that the raw get hardware sharpening that it is not possible to avoid


No, there is no hardware sharpening, just a weak AA-filter and a sharp lens. You may be able to ease this problem by using a worse lens Smile Or maybe add some grease to the front (or even rear) element Smile

This image of a parrot has clear problems induced by a bad demosaicing process, ugly artifacts. Even though Lightroom 1.4's (which is at the moment the tool I use) demosaicing is not that good in general, I've never seen anything like the ugliness in some of the feathers (and Pentax K20D also has rather weak AA-filter - aliasing is occasionally seen).


Reply

I guess I have to put some grease to my lens Smile

I found the solution with dcraw with parameter -f : Interpolate RGGB as four colors
so the problem occur with mixing of the 2 Greens in RGGB matrices

dcraw left with -f , right normal

--- edit ----
but the -f kill resolution
the sensor data with dcraw -D show that the jigged lines are already there
look like a aliasing problem from sharp Zeiss lens
in those condition, the normal dcraw output made a better job than capture one


最后由 poilu 于 周三 3月 24, 2010 10:40 am 编辑;共编辑 1 次


Reply

poilu 写道:

I found the solution with dcraw with parameter -f : Interpolate RGGB as four colors


Would it be possible for you to post the raw file somewhere? A file known to be problematic with most converters would be an interesting way to test the behaviour of different utilities.


Reply

Arkku 写道:
Would it be possible for you to post the raw file somewhere?

http://rapidshare.com/files/367267231/parrot.CR2
if you find a different behavior with your raw converter, share it here


Reply

poilu 写道:

http://rapidshare.com/files/367267231/parrot.CR2
if you find a different behavior with your raw converter, share it here


Thanks, and I will, although I'm waiting to upgrade my computer in the next few weeks/months, and don't have many different converters installed on my current one.


Reply

I noticed your firmware is 2.03.
Try with the new one ?

Do you shoot in sRVB ?

I tried in DPP 3.8 and worked only on the luminance noise reduction. I set it to 6 which is a lot.
Here are the results. I don't know wether it's better or not :


Reply

Olivier 写道:
I tried in DPP 3.8 and worked only on the luminance noise reduction. I set it to 6 which is a lot.
Here are the results. I don't know wether it's better or not

nice try Olivier, but with noise to 6, it look like from a m4:3 Laughing

引用:
I noticed your firmware is 2.03.
Try with the new one ?
Do you shoot in sRVB ?

firmware and srgb doesn't affect the raw who is a copy from the sensor
the jpg could get tweaked but no effects on raw conversion


Reply

You're right, Poilu. Wink


Reply

raw therapee have a demosaicing algorithm called VNG-4 who give good result

left VNG-4 ... right Bibble 5.3 no sharpening


Reply

poilu 写道:
raw therapee have a demosaicing algorithm called VNG-4 who give good result

left VNG-4 ... right Bibble 5.3 no sharpening


VNG four color interpolation? Also in DCRaw (UFRaw).