landmarks4711
Leading Member
note: 40D chart changed to blue below
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
D300 vs Canon 40D Noise
- Thread starter ThaQuest
- Start date
note: 40D chart changed to blue below
landmarks4711
Leading Member
so if in PP exposure increased +1EV, it's a one stop Iso.
how much you on camera EC or not to EC doesn't matter.
how much you on camera EC or not to EC doesn't matter.
landmarks4711
Leading Member
Looking at the D300 graphs only, I see almost no reason to ever shoot below ISO 800, since 800 has almost as low of noise as 200.
note: 40D chart changed to blue below
--
3fros
http://www.pbase.com/frofamily/
The D300 has four levels of NR - high, normal, low and off. Often it is said on the Canon forums that the D300 has NR which gives low noise but destroys detail. The setting "normal" is default and is a bit too much IMO. This is from IR:s review of the D300:
Here the Canon shows more detail.
But the review also has samples with the other NR settings of the D300. Here is the setting "low". Now the D300 shows more details and more sharpness than the 40D:
Personally I always use NR low or off.
About ISO 6400 - here are some samples:
--
Small first D300 gallery:
http://www.pbase.com/interactive/d300_first
Small D200 gallery:
http://www.pbase.com/interactive/d200_12
Small D40 gallery:
http://www.pbase.com/interactive/d40_12
Small Nikon P5000 gallery:
http://www.pbase.com/interactive/nikonp5000_12
http://www.pbase.com/interactive
antimonite
Well-known member
Noise reduction starts at ISO 800 with any setting (Normal, Low, High). This is a good reason.
Fred Dominic
Well-known member
Actually, this is something that never really made sense to me both in theory and in practice. I thought that the point of ISO is that since any amplification is analog, there is zero probability of any digital quantization effects causing any additional noise when undergoing the general process of amplification. It is also operating at an earlier stage in the information flow, and as you might know, any time you process a signal, you are distorting it to some extent relative to the "original." Amplifying at the earlier analog stage won't have a chance to enhance any distortion done by the ADC or anything else artifact-producing that might lie in between.
I did some tests and this seemed to work out. That is, if I shoot at 1600 and compare that with the same shot at 800 + EC -1 (everything else equal) and them digital EC up by one stop, or if I shoot at 800 and compare with the same shot at 400 + EC-1 followed by one stop up of EC, and again shot at 400 and same shot at 200 with one stop of EC, where all of the shots have relatively low light, or even if I shoot at 400 EC-2 followed by 2 stops of EC and compare it to the 1600, in each case the shot done with the higher ISO has noticeably less noise than the lower ISO + EC shot (all shots done using RAW, viewed with Adobe lightroom without any additional processing done at all, and all under the "flat" settings).
I have concluded, then, that when people say "but your camera can do 3200 or 6400 (or even 12800)", it is not really true. I would bet that a camera that could do real analog amplification up to the equivalent of those high ISOs would have a correspondingly better noise profile at those ISOs then the method that you and others have espoused here as an alternative.
I think a much better way of simulating a high-ISO shot with a 40D (and if you have the latest photshop cs3) and if you have a relatively stationary object, would be to do the following. Set your ISO as high as you can honestly go (1600), and set to normal exposure (or better, shoot in manual). Shoot an odd number of shots in rapid succession. I usually do 5. Then, in photoshop cs3, load all the shots as separate layers, align the layers, convert them to a smart object, go to the smart-object stacking mode and choose "median" (which takes the median of the shots pixel-by-pixel), and then flatten it back down to a single image. Since the noise is random, the median gets rid of much of the noise. And since you are computing the median not over neighboring pixels but rather over the same pixel on different shots, the result stays sharp. I've been able to take some very low-noise dark or night shots of landscapes and buildings this way, and they come out extremely sharp, even hand-held (thanks to the photoshop align ability, and lens IS helps too). This is, by the way, one of the big reasons fast cameras such as the 40D or D300 are so nice.
Of course, if your subject is moving this does not work.
The sort of noise comparisons (charts or examples) at this and other review sites seem more attuned to point and shoot cameras than to high-end digital SLRs. If you don't need a high-throughput process (if, for example, you're not a wedding or sports specialist who needs to produce lots of jpg images), it's so easy to exceed the results produced by the in-camera settings (e.g, with a batch process in Capture NX) that it's hard to understand why you'd care what the default settings produce; if you are a professional photographer who needs to review and produce a large number of jpgs, I wonder if you'd use the default settings, or even care what they are.
landmarks4711
Leading Member
NickMJr
Senior Member
I completely agree, which I have always been critical of with DPREVIEWS camera comparisons as they always use default settings. What amazes is how most reviewer simply unpack the test camera and take pictures, just like a professional would do huh?
http://nickmjr.smugmug.com/
Nick M
Brad Morris
Senior Member
compatares teh sony with 40D and D300.
here you go:
http://www.dpreview.com/reviews/sonydslra700/page17.asp
http://www.flickr.com/photos/k-blad/
here you go:
http://www.dpreview.com/reviews/sonydslra700/page17.asp
--
http://www.flickr.com/photos/k-blad/
ThaQuest
Senior Member
Thanks for posting the charts and the pictures, everyone. It seems to my eyes and according to the charts that the D300 is a bit better in high ISO. Kudos to Nikon for pulling this off, especially with a 12MP sensor!
---------------------------------------------
Images of China: http://www.shenzhenphotos.com
Photos for Charity: http://www.photogiving.com
---------------------------------------------
Images of China: http://www.shenzhenphotos.com
Photos for Charity: http://www.photogiving.com
jp52637
Senior Member
the first is as you said: analog amplification does not introduce quantization noise (in digicams this can be seens as posterizing) while digital amplification (multiplication) does.
The second and more important reason is that the higher the (real) iso setting on your camera, the lower the read noise becomes and this gives vastly better results.
Dr Roger Clark has already performed these experiments and he came to the same conclusion: when you shoot at the highest possible ISO, the results are vastly lower noise than when you use mathematical brighthening of low ISO shots due to lower read noide
--
cameras don't shoot people
People shoot people.
jp52637
Senior Member
Of course this has nothing to do with NEF files or anything, that is not what I meant. LOL
I meant the raw signal performance of the linear AF sensors, ignoring the AF algorithms that are behind it that do cleanup and correlation
As you probably now, an AF sensor also has a noise and DR spec and it was that spec I was talking about.
--
cameras don't shoot people
People shoot people.
jp52637
Senior Member
I was talking about the signal noise ratio of the AF sensors of course, not the image sensors.
My usage of the word "raw" caused apparently caused some confusion
I had better said "bare performance"
--
cameras don't shoot people
People shoot people.
My usage of the word "raw" caused apparently caused some confusion
I had better said "bare performance"
--
cameras don't shoot people
People shoot people.
jtsmall
Senior Member
For each stop of ISO increase one loses a stop of dynamic range. So, there is an engineering trade off to consider. Of course, if one must increase ISO to get the shot, the cleaner sensor is to be perferred all things being equal. So it is of considerable benefit to shoot at the lowest ISO consistent with the artistic result desired.
You can see the interplay of ISO and DR visibly in this crude experiment ...
http://www.pbase.com/jtsmall/nikon_d300
-jts
------------------------
http://www.pbase.com/jtsmall
Canon, Nikon and Olympus
equipment in profile
'From the first moment I handled my lens with a tender ardour.' Julia Margaret Cameron
Bmark
Senior Member
That is a great technique.
Thanks,
Mark
Thanks,
Mark
DRG
Veteran Member
The primary function of ISO is to inform the meter how to expose. Higher ISO simply tells the camera to pretend that the sensor has a lower maximum exposure and thus to shorten the exposure, providing the photographer, at a high level (that is, forgetting implementation details) with a way to use a smaller aperture or shorter shutter speed.
So ISO tells the camera where max. exposure is and the camera then changes its metering.
Once this is done, there are consequences to the signal processing chain which I'll talk about in a sec. First, though, we should note there are immediate consequences to image quality from the simple reduction in exposure: increase in shot noise from the lower total photon dose. This impacts upper shadows through highlights primarily.
Now to the rest of the signal processing chain. The first step is reading the signal as a voltage from each pixel. This step introduces read noise, the key factor in setting shadow noise.
Next comes an analog multiplication. This step can actually be quite clean, not impacting the signal quality much at all (i.e. the read noise is much larger and dominates).
The final step is analog-to-digital conversion. In order to make full use of all the tone values that can be assigned to the signal (i.e. 0-4095 for 12 bits, 0-16383 for 14 bits), the goal of the analog amp stage is to amplify the signal going into the ADC so that the signal from a maximally exposed pixel comes as close as possible to the maximum input voltage of the ADC.
With this in mind, let's take a step back and note some consequences:
1. Shot noise is impacted by ISO via the total exposure.
2. The ISO analog amps don't seriously impact the signal-to-noise ratio. High ISO images don't have higher noise because of these amps but because of the higher shot noise already built in from the lower exposure.
3. Many cameras seem to have different read noise at different ISO values. Canon cameras, in particular, have lower read noise at higher ISO, presumably because the circuits can tune themselves differently when they know that the maximum signal level which they need to amplify will be a smaller value.
4. Multiplying the signal digitally after the ADC stage also doesn't change SNR. However, if you expose at ISO 800 with EC = -1, you end up with the higher read noise associated with the sensor tuning itself for ISO 800. The analog amps, ADC stage and final digital multiplicatio by 2 don't change the SNR much, but this higher read noise is baked in from the start. This is why the ISO 1600 ends up as higher quality.
Note that this same phenomenon isn't true in all cameras. Read noise can stay the same with ISO, making ISO 800/EC=-1 the same as ISO 1600 except for a loss of precision of 1 bit from the digital multiplication (which doesn't matter if you start off with more than enough bits). Or, read noise can go up with higher ISO, making ISO 800/EC=-1 actually better than ISO 1600.
Explained above, for Canon cameras at least.
But it is true, since "doing an ISO" means exposing for it. Whether or not you obtain good quality is up to many factors in the camera's pipeline, not just whether or not there's a dedicated analog amplifier gain just for this ISO. In fact, as long as there's enough bit depth, the key factor in good high ISO shadow noise performance is read noise that drops with increasing ISO.
It's not the analog multiplication. Note that some cameras, like the K10D, have only a single analog gain and implement all ISO values digitally (leveraging their high starting bit depth - 22 bits for the K10D).
This isn't simulating a high ISO shot. It's a way of combining multiple shots, which means you're bringing in more data, the equivalent of shooting at lower ISO but manipulating the statistics a bit differently instead of averaging. Also, this typically won't work in circumstances for which people employ high ISO: handheld, indoor shooting.
David
Keyboard shortcuts
- f
- Forum