50D high iso detail vs noise

[Dave Taylor]

Hey everyone - hope you are all having a good "spring" so far. I'm looking for input from those of you that own a 50D. My high ISO images (on the 50D) - >640 seem to lack the fine detail that I'm getting at low ISO's. I shoot raw and expose to the right. Do you turn on high ISO noise reduction in camera, or just deal with it in post? I know that white balance
How would you describe your experience with the 50D and higher ISOs? The best advice I've ever heard was from Chas - If it's a choice between noise and blur, I choose noise. I'm going to do some tests at different ISO's to compare detail - just for kicks. As much as I hate to admit it, perhaps it's user error - and I'm not as "steady" as I need to be or need to push the ISO a little higher to get a higher shutter speed.

Your thoughts/impressions/musings?

[Dave Taylor]

oops, didn't finish my sentence. I meant to say: I know that settings like white balance are only valid on jpegs and on the camera lcd - once RAW files are brought into LR they are still adjustable. But what about "high ISO noise reduction" in the camera? Is it best to set it to none to keep the detail and then deal with the noise in post (Noiseware, Noise Ninja)?

[Alfred Forns]

Hi Dave I have been using Topaz with great results, seems to do better than NoiseNinja without doing a number of sharpness.

....btw might want to take a look at this link, its the current ETL Selection and I believe it has no noise reduction, check original post !!!! http://www.birdphotographers.net/forums/showthread.php?t=35140

[Rene A]

Topaz also has a new version of Topaz Adjust with some new functions..that compliments Denoise. It can be downloaded as fully function 30 day trial version.

[John Chardine]

Dave- I understand that on the Canon CMOS sensors, a certain amount of NR happens on the sensor itself. My understanding is that if you shoot RAW, the in-camera NR setting is like WB and other settings- it is not interpreted unless you use Canon's Digital Photo Professional. Although I don't use DPP, I will run high-ISO (800 or above) images from the 50D through it because it seems to do a better job of NR apparently without loss of detail, than running NR in ACR or some other program like Noise Ninja. In general I find that the 50D performs well at ISO 800 so long as you get the exposure right.

[Dug Threewitt]

oops, didn't finish my sentence. I meant to say: I know that settings like white balance are only valid on jpegs and on the camera lcd - once RAW files are brought into LR they are still adjustable. But what about "high ISO noise reduction" in the camera? Is it best to set it to none to keep the detail and then deal with the noise in post (Noiseware, Noise Ninja)?

You may want to try DPP. On my 50D, LR is horrible with noise(even at 400ISO). If I have an image I really like, I take it through DPP for processing and just let LR be an asset management program.
Dug

[Alfred Forns]

Haven't updated Rene Will check it out ... any improvement is much appreciated !! Will also check out DPP for comparison !!

.... I think the main point is no under-exposure !

[Stu Bowie]

Hi guys, when I got my 50D, I experimented a lot, as it was a huge upgrade from my 20D. After reading all sorts of info regarding noise, I decided to experiment myself. The attached image is completely untouched apart from cropping to suit the image size. No pp work was done on this whatsoever. I do believe if you have half decent light, the 50D performs well with regards to lack of noise. I remember this shot was taken in late afternoon light.
Settings were as follows: 1/4000 - F/7.1 - ISO 1600.

[Rene A]

Stuart,

I run it through Topaz Denoise..it looks a lot better, but it seems that can NOT post attachments here.

[Dave Taylor]

So do what do you set in the camera to retain the most detail? Do you do noise reduction in the camera (and how much) or do you set it to none and only deal with it in post?
I'm going to try out Topaz, looks like everyone thinks highly of it. I've been using Noise Ninja for a while now - but not ultimately happy with it.
dt

[Stu Bowie]

I would like to clarify that no noise reduction has been done on my attached image. It was posted to show the result of ISO 1600. BTW, all my camers settings are set to default. I do my NR in my pp work, and use neat image.

[Marina Scarr]

So far with my 50D which I have only used a handful of times, I am not happy with the noise compared to the 40D. If I have a clean, light BG, there is no problem. However, if I have a BG which is partly or completely dar, the noise is very apparent even at ISO 400.

[Dug Threewitt]

So far with my 50D which I have only used a handful of times, I am not happy with the noise compared to the 40D. If I have a clean, light BG, there is no problem. However, if I have a BG which is partly or completely dar, the noise is very apparent even at ISO 400.
Are you using lightroom? I haven't had the problem with DPP and I don't think I lose detail through DPP either.

[Peter Hawrylyshyn]

dave
this has been covered in another thread which you may find useful:
http://www.birdphotographers.net/forums/showthread.php?t=29716

given the increased number of pixels packed onto the same sensor size, it should be expected that there will be more noise with the 50D compared to the 40D

i've starting using Topaz Denoise on Alfred's suggestion and have been very pleased with the results. I feel overall the 50D is a superior camera to the 40D, but the 1D Mk3 still has the best noise overall preformance

[Roger Clark]

Dave- I understand that on the Canon CMOS sensors, a certain amount of NR happens on the sensor itself.

Hi John,
I have not seen any evidence of this. Perhaps you mean their overall low noise sensors? Ten years ago (approx), people didn't think CMOS could cut it for imaging and that CCDs would always outperform. Canon refined the technology, and made CMOS mainstream. They applied technology like double correlated read to reduce the low end noise, then refined the active area and micro-lenses on each pixel for the brighter signals (where photon noise is the dominant noise source). They seem to be refining these things with each new generation, as are other manufacturers. This has helped the low end uniformity (e.g. banding) and read noise which is so important for high ISO performance.

Noise reduction is a processing step (which could in theory be on chip, or in post processing) that trades spatial resolution for noise. All noise reduction algorithms make this trade. Different algorithms do it differently in different situations (meaning better/worse performance, and/or more/less artifacts). Same with true sharpening algorithms: improved sharpness results in more noise/artifacts.

[John Chardine]

Hey Roger- I was referring to Canon's own write-up in their fall 2008 EOS EOS system catalogue. On p. 6 they say:

"Canon's CMOS sensors incorporate a unique on-chip noise reduction technology to deal with both fixed pattern and random noise."

Is this what the "double-correlated read" et al. is all about?

[Bob Blanchard]

I'm a slow burner when it comes to high ISO photography. If I creep above ISO 400, I feel like I've sinned or something. I tend to live mostly at ISO 200. That said, I have found that the best settings (for me anyway) on the 50D is to turn off all the in camera NR, and ESPECIALLY the Highlight Tone Priority (I disabled mine). I have found my images are great with these settings at low ISO. The camera also performs extremely fast with all this stuff off.

I find this thread a very interesting read. I usually use Noise Ninja for NR, but will give Topaz a try (I just use PS for any selective sharpening anymore, and I've always disliked Noise Ninja's attempt at sharpening). I may also play with DPP and some high ISO stuff just to see what happenes. I love the 50D, and if I felt good with it over ISO 400, I'd love it even more.

[Roger Clark]

Hey Roger- I was referring to Canon's own write-up in their fall 2008 EOS EOS system catalogue. On p. 6 they say:

"Canon's CMOS sensors incorporate a unique on-chip noise reduction technology to deal with both fixed pattern and random noise."

Hi John,
(I was off in Canyonlands so it has taken me a while to notice your question.)

They are probably referring to their reset technology. Before an image is taken, reset electronics drains off all the electrons in the wells. A lot of the low level fixed pattern noise we see in digital camera images has to do with how stable the voltage levels can be maintained. That is difficult because with all the electrons moving around, voltages are induced. And then consider the signals in this fixed pattern noise is on the order of 1 electron
and less! I haven't done the calculation, but I bet that translates to less than 1 billionth of one billionth of a volt. Digital cameras and the performance levels achieved in todays cameras are truly astounding. Twenty years ago such performance was difficult to achieve in a laboratory with liquid nitrogen cooled electronics, yet now we see such performance on all kinds of conditions outside.




Is this what the "double-correlated read" et al. is all about?

Some image readout technology is non-destructive. So when you read out the image, the image still exists on the chip until it is erased by bleeding off the electrons. So you can read the image data multiple times and average them so you reduce the noise from the downstream electronics. There was a cool animated gif on the web about this but I can't find it (I found a couple of links that are no longer active--might have been one of those). Double correlated read refers to reading the image twice. You could do more, but it takes more time.

An analogy: a bucket where rain fills the bucket (analogous to photons filling the well in the sensor). After the image is taken, you must empty the well. Imagine spigot on the side of the bucket near the bottom. After you take the image (the rain drops), you open the spigot and drain out the water (photoelectrons). But you can't get every drop of water out of the bucket. The best you can hope for with repeated fills of the bucket is to know the residual level very accurately so you can subtract it from your measured amount of water (or photoelectrons the sensor). In the sensor, the reset voltage represents the residual signal level in the pixel and is subtracted from the image.

Again, it is simply astounding that consumer digital cameras control that signal as well as they do.

Think about it. Every time you take a picture with a digital camera, you are doing a quantum physics experiment!

Roger