minimising noise in digital cameras

[Loris Medici]

Test finished

Hi all,

I finished my tests. First, the visuals. Follow the links below:

Small jpegs from originals:
http://www.flickr.com/photos/loris-medici/sets/72157623509570487/
(Please inspect the images at their Original/Large size; it's done by clicking on the "All sizes" link above the image.)

Original files (including the RAWs):
Dead Link Removed
Username: hybridphoto
Password: Hp12345
(Please note that both username and password are case sensitive...)

All file names are self explanatory.

Test conditions below: (It's long, go to the bottom for the results in case you're impatient. You can return back here later...)

Camera: Sony A700 - a fine camera for its modest price.
ISO setting: 400 (I've chosen this value because I) it's most probably what we're going to use - at minimum - in casual indoors shooting conditions w/o the aid of a flash, and II) I wanted to avoid noise reduction which automatically activates in case of shutter speed equals or exceeds 1 second.)
Lighting: One 100W incandescent light bulb with a soft envelope (not naked) in the ceiling, at a distance about 180cm/~71" from the subject. For those who are familiar with EV values, incident light metering right on subject indicates EV 3.5 - which corresponds to EV 5.5 at ISO 400 sensitivity.
Lens: Tamron 17-50/2.8 - a very fine lens for its modest price.
Filter: Cokin P 020 Blue (80A) (Valued USD 15, 1.5x the price pellicle had quoted - I mean it's presumably something finer than what he was anticipating... )
Shutter speeds: 1/4 for the image shot w/o a CC filter (About 2/3 stop moderate over-exposure - I mean considering the subject - w/o causing any clipping in non-specular highlights), 0.6s for the image shot w/ a CC filter (About 1/3 stop small over-exposure w/o causing any clipping in non-specular highlights. Please notice that the CC image got 1/3 stop less over-exposure than the no-CC image, in other words it's effectively exposed less.) Both image were shot on tripod, because of the relatively low light level and for sake of consistency in framing and focus etc.
White balance while shooting: Camera's default incandescent white balance setting for the no-CC image, camera's default daylight white balance setting for the CC image.
White balance settings in Adobe Lightroom RAW processor: Left as shot for the no-CC image (2950K / +1) - because it was very close to what I had perceived at time of shooting, set to 4250K / -6 for the CC image - both because I wanted to match it as closely as possible with the no-CC image and was way off. (Too red; probably it's due to the incompatibility in the elements of the system that consists of the light source / CC filter color correction factor and camera's default daylight white balance setting.)
Other RAW processor settings:
1. Exposure: -0.66/no-CC, -0.40/CC
2. Recovery/Fill Light: 0/0 for both no-CC and CC
3. Blacks: 2/no-CC, 3/CC
4. Brightness/Contrast: 50/25 for both no-CC and CC
5. Clarity, Vibrance, Saturation: All 0 for both no-CC and CC
6. Sharpness Amnt.: 0 for both no-CC and CC
7. Noise reduction Luminance/Color: 0/10 for both no-CC and CC

Please ask about any other non specified detail (it any) that you're curious about...

You can download the original RAW files and play with them yourself. (In case you question my credibility and/or you're just curious. Where are yours pellicle? :rolleyes IMPORTANT: Please refrain from posting self-processed files w/o including all the processing parameters exactly. (Please note that I have LR and PS, and can't process RAW files w/ any other software...)

Results:

1. There's very little noise - considering the ISO setting used in the shooting - in both of the images, OTOH, the less exposed image shot "w/ CC filter" exhibit more noise than the more exposed image shot w/o CC filter. See the small 100% crops of the B channel, especially...

2. The image with CC filter is slightly softer than the other one. The filter affected image detail/resolution in a negative way. (I think that would be even more pronounced if I had used ISO 100.) See the face crops, look at the eyes and hair... Each shot was focused individually to the same spot using the center cross sensor (it's a F2.8 lens remember), w/o any camera movements. The camera was stationary / locked to to the tripod from start to end...

Not a surprise I'm afraid; thanks much for taking so much of my time for an exercise in futility...

 

[Mike1234]

Loris,

Thanks for sharing. Could you post the separate R, G, and B channels please? What actions did you take to get the filtered and unfiltered images to appear so much alike from shadow to midtone to highlight?

 

[pellicle]

Hi all,

I finished my tests. First, the visuals. Follow the links below:

slight posterization showing in the sole of the foot
marked reduction in contrast range of the blue (compare deep shadow levels and hands)

Not a surprise I'm afraid; thanks much for taking so much of my time for an exercise in futility...

given your total disregard for my time, your unsupported assertions and that I somehow did this to simply cook the books if wasting your time has occurred you can be assured buddy that that is just Karma at work

As it has already been pointed out in this thread (by me even) that over cooking the exposure (and thus blowing the Red Channel as well as possibly the Green ) is a strategy that will result in raising the Blue levels. You have clearly chosen this path and therefore entered into high light recovery in the RAW processing. So thanks for wasting our time.

But I"m sure many of us have learned something else, and that is what a self important accusing little piece of work you are. I provide something and you sit back and whinge and yell just like the stereo types in the movies. You have earned an entry into my "ignore user file"

 

[Loris Medici]

...
slight posterization showing in the sole of the foot
marked reduction in contrast range of the blue (compare deep shadow levels and hands)
...

Sorry but I don't see any posterization in the foot area, in any of the images; the histograms are intact. OTOH, the no-CC version looks much more cleaner because there's less noise. See the cheek of the baby for a better visualization...

...
As it has already been pointed out in this thread (by me even) that over cooking the exposure (and thus blowing the Red Channel as well as possibly the Green ) is a strategy that will result in raising the Blue levels. You have clearly chosen this path and therefore entered into high light recovery in the RAW processing. So thanks for wasting our time.
...

That's not "over cooking"; it's a) giving correct exposure (especially in case you need to suppress noise as much as possible) since b) there aren't any blown channels in any of the RAW files.

I highly suggest that you shoot RAW and give adequate exposure when using digital cameras, that's how it's done... Those were the two big hints / solutions already given to you right at the beginning of this thread - which became unnecessarily long.

Ciao...

 

[donbga]

Don

I know you have work and everything, but :



if you care to start another thread on that I'd be keen to read it. I've never done any significant testing on the subject and try not to use filters unless they're needed (like when I use HIE on my SLR).

I would suggest searching Photo.net on this topic. IMO, it's not worth protracted testing unless you are shooting film as digital sensors seem to be immune to UV influence though that may not include high altitude shooting. Also I'm only talking about UV 0. A UV 2 may also be a different story.

In short I place a UV filter over the front lens element to protect the lens element and because I shoot film with the same lenses.

 

[pellicle]

That's not "over cooking"; it's a) giving correct exposure (especially in case you need to suppress noise as much as possible) since b) there aren't any blown channels in any of the RAW files.

if you say so ... but much depends on what you do with those raw files. Now, many converters help you to suck your thumb and do channel recovery for you. This helps both sloppy photography and accidents to recover. I guess that the science of blown high light recovery is getting quite good these days. However if you rely on that for your noise boost great ... I don't mind ... just be aware that you're doing that.

Now when I look at tiff converted using camera white balance of your no CC filter TIFF without entering into high light recovery I get this:



when I make a 16 bit linear decode to see the full data set recorded in the channels I get this:



back in the 8 bit sample, even giving the blown high lights I see this in the blue



so if you don't mind your converter hiding things from you and doing recovery of blown high lights for you (and they do this quite well) then by all means go for it. It will drive up your signal to noise ratio in the floor noise area.

Myself I still don't like working that way, but then I guess that there is seldom any significant details you really want to hold in the highlights ... may as well reconstruct them anyway.

I'm willing to continue exploring this with you (as some interesting points have arisen from my dissection of your work) however if you intend to keep acting like an arrogant commanding dictator then that offer is off. If you wish to ask civilised questions about how I came to my conclusions and wish to discuss them, by all means. God knows I'm probably going to regret this (assuming you're a reasonable man that is).

 

[Bob Carnie]

Pellicle
Bottom sample is brighter over all by around 7-12 L density units if this matters at all.

 

[Bob Carnie]

bottom right ... sorry

 

[pellicle]

Hi Bob

sure, I didn't equalise or disturb them at all. I just converted them to TIFF and viewed. Exposures were not equal between each or in any specific channel.

knotty stuff for a comparison

 

[Loris Medici]

pellicle,

You can't recover "truly" blow out highlights; think of the base of underexposed film as an analogy (to make you understand better); you can't bring something back that wasn't recorded in the first place. (I think we give different meanings to the same concept...)(*)

When I open the no-CC RAW I see what's shown below:

1) An averaged skin patch from below the right (to us) eye gives me a combined levels value of 233 (at time of capture - downsampled to a scale of 0-255), and there are absolutely no blown-out highlights (same as above *), except the specular reflection in the eyes and and hair.

2) Since the combined channels don't give us posterization but smooth gradation in the face, seeing 255 for the Red channel (right there) doesn't mean that it was blown-out actually, it means that was the exact nature of the subject. (Remember, that's a plastic doll "with a totally homogenous" skin color...)

3) As you can notice, there's no detail loss in the "much lighter/closer to white" parts of the dress; you can see texture of knits clearly.

Because the three reasons listed above (but not limited to them), I can't accept your assertion that the image was blown-out. It's simply not...

 

[Loris Medici]

FWIW, see the R channel of the processed "8bit" PSD, with an extreme gamma adjustment (1.00 to 0.20 - in order to make tones more easily observable) right below:

P.S. I don't have the 16bit version on hands right now, in any case, even the R channel of the 8bit file does show a nice gradation after such a drastic tonal manipulation...

 

[Bob Carnie]

Loris

The red channel is usually the smoothest , could you do the same with the blue.
thanks
bob

 

[Loris Medici]

I can do that indeed, but our current context was over-exposure and/or blown highlights and the channel which is most closer to the clipping zone is red; that's the reason of our current focus on the red channel...

We all agree on the fact that usually it's the blue channel which exhibits the most noise. The real issue is: http://www.hybridphoto.com/forums/showpost.php?p=13451&postcount=65

I'm going out right now. Tell me if you still want that, then I will provide the comparison of noise and gradation in the blue channels of each image. OTOH, I think I already have visuals for the blue channels in the flickr gallery...???

 

[Bob Carnie]

I was just interested in the blue with aggressive move on them as you demonstrated with the red.

 

[pellicle]

You can't recover "truly" blow out highlights; think of the base of

I agree ... its extrapolation

I can't accept your assertion that the image was blown-out. It's simply not...

then can you explain why your own red histogram goes off the end of the scale in the screen shot from above?

 

[pellicle]

Since the combined channels don't give us posterization but smooth gradation in the face, seeing 255 for the Red channel (right there) doesn't mean that it was blown-out actually, it means that was the exact nature of the subject. (Remember, that's a plastic doll "with a totally homogenous" skin color...)

interesting assertion

 

[Loris Medici]

...
can you explain why your own red histogram goes off the end of the scale in the screen shot from above?

Because there are specular reflections in the image!??? See the eyes and hair, count them if you like...

Plus, remember it's an incandescent light source, a spectral power distribution towards yellow/orange/red, something around 2700K. Can't measure it exactly since I don't have color temperature meter...

Come on!? That makes me feel like I'm explaining something to a 3 y/o. (You know, it does iterate forever - w/o making any sense...)

Can you please refrain from that and make a proper statement instead? Let's return back to the subject; see my post #65 to refresh your memory.

Thanks in advance.

 

[Mike1234]

This was one of the arguements about overexposing to minimize noise in the blue channel... clipping in the red. The further off the color at capture the worse the issue becomes. Correct the color before capture and this issue disappears provided the scene DR is within limits.

 

[Loris Medici]

I was just interested in the blue with aggressive move on them as you demonstrated with the red.

OK, see the files below. (They're on my flickr page.) I did another drastic gamma adjustment on the B channel of the files; 1.00 to 2.70. (In order to let you see the tonal structure of lower midtones, shadows and blacks better...)

No CC:

w/ CC:

I see more noise in the image shot w/ CC filter, don't you agree?

 

[Bob Carnie]

I see noise in both, thanks for posting

 

[Loris Medici]

Yes, obviously; it was shot at ISO 400. (I never said there won't be noise in any of the images, BTW - Not that I think you imply that...) The question is: Which has more noise than the other? Or, do they have similar (and non-discernible in practical conditions) noise levels? And, does the CC filter exhibit any advantage - which also cancels out its disadvantages?

I see noise in both, thanks for posting

 

[pellicle]

Because there are specular reflections in the image!??? See the eyes and hair, count them if you like...

Loris when interpreting a histogram the x axis is the value range from 0 to 255 and the y axis is the amount of pixels which have this value. So for the red channel histogram you show to be from spectral high lights it would not be swarming over there on the right hand side, the would be a few pricks and lines in the histogram reflecting the points of the image which were spectral highlights..

taking the red channel from your CC image and converting to keep red under control we get this



thats as a result of the (expectable) 255 values in eye and hair spectral highlights, and a little in some of the face ... more than I'd personally like. This type of skin tone should be high in values but not death mask white in the red channel.

 

[Loris Medici]

:confused: What I get in ACR (for the CC shot) is shown below:

I get a combined levels value of 225 under the right (to us) eye - less than 233 from the no-CC shot. (See my post #85 to refresh your mind...)

Now, I really want to hear something with substance (because I'm starting to get bored from the ignorance / bluntness / dumbness and pathetic behaviour I have been coming up against in this thread, since the beginning...), which directly relates to what I said previously, all quoted below:

"...
The culprit in all of the images we've seen so far is improper exposure / lack of knowledge of giving proper exposure.

I say, (1) when proper exposure is given there shouldn't be any perceptible difference between the image taken w/o CC filter and the one taken w/ CC filter, (2) and with CC filter you loose 2 stops of light which makes your shooting more difficult (forcing to use a tripod -> where you could shoot handholding the camera or forcing you to increase ISO -> where using a tripod isn't possible, which also is something totally against the original purpose of keeping noise levels down) and finally (3) using a USD 10 CC filter (pellicle's definition, not mine...) over a good lens will probably turn it into a mediocre one.
..." (Post #65)

and

"...
Results:

1. There's very little noise - considering the ISO setting used in the shooting - in both of the images, OTOH, the less exposed image shot "w/ CC filter" exhibit more noise than the more exposed image shot w/o CC filter. See the small 100% crops of the B channel, especially...

2. The image with CC filter is slightly softer than the other one. The filter affected image detail/resolution in a negative way. (I think that would be even more pronounced if I had used ISO 100.) See the face crops, look at the eyes and hair... Each shot was focused individually to the same spot using the center cross sensor (it's a F2.8 lens remember), w/o any camera movements. The camera was stationary / locked to to the tripod from start to end.
..." (Post #76)

and

"...
The question is: Which has more noise than the other? Or, do they have similar (and non-discernible in practical conditions) noise levels? And, does the CC filter exhibit any advantage - which also cancels out its disadvantages?
..." (Post #96)

BTW, I still think that:

"...
The issue is rooted in the fact that some people still haven't completely finished their transition / conversion from analogue to digital (or just adaptation if you like), therefore are still stuck in (some) concepts which are meaningful only in the context of analogue photography.
..." (Post #26)

 

[Bob Carnie]

Loris

I am following this thread, asking a few questions and in some cases making a few observations.
I am neither dumb, ignorant of the subject at hand, and sometimes my behaviour could be considered pathetic, but not here..

So far this thread seems to start to get to the point, and then it goes all to shit with nastiness.
I think this thread is doomed.

 

[ann]

i am with bob on this, and in fact to be honest there are times i get lost with what i am supposing to be understanding.