Pushing ISO to 3200 and 6400 on Rebel XTi 400D

[BenJohnson]

I saw this thread today, and thought I'd try using Photoshop to push my Rebel XTi to ISO3200 and ISO6400:

http://photography-on-the.net/forum/...d.php?t=667757

I shoot mainly sporting events, so JPEG is my preference. I do realize RAW would be
preferred for this type of post processing, but I normally shoot JPEG, so that is what
these shots were taken with (as opposed to the thread I linked to).

Setup:

Camera: Canon Rebel XTi 400D
Lens: Canon 70-200mm 2.8 IS
ISO: ISO1600

This wasn't really scientific test, just some quick snapshots. I did NOT use a tripod. I
shot them in manual mode: first shot taken at the shutter speed the camera metered to,
and then speed doubled for the subsequent shot(s), ie. stopped down 1 complete stop
each time. I only went to -2 stops (ISO6400 equivalent after post processing). The
exact settings are posted before each set of pictures. I varied the aperture and focal
range a bit, and also used IS for one set of shots.

After I took the shots I brought the JPEG's into Photoshop CS2. I used the "Curves" tool
to adjust the images. This allows the lights and darks to remain mostly unchanged and
bring up the mid levels. Using the "Exposure" tool left my images too dark and if I tried
to increase it more, the highlights (and low lights) looked very poor. If there is a better
way (for JPEG images), please let me know!

Using the "Curves" tool I added two adjustment points to change a -1 stop, ISO1600
image into a properly exposed ISO3200 image:

Input 64, Output 116
Input 128, Output 180

The 0,0 and 255,255 points were left unchanged. To come up with these points I started
at 64,128 and 128,192 because I thought these would give me 1 stop. They looked a
little too light, so I took 12 off each of the output numbers. Nothing scientific, just did
it by eye in 5 min or less.

Same thing for the ISO6400 images, but with these two points:

Input 50, Output 150
Input 150, Output 250

Didn't do anything interesting to come up with these numbers either, basically just slid
the sliders around for a min or two. If someone has better recommendations, please let
me know!

That is the ONLY processing I did. I did not adjust WB, saturation, etc, etc. Just the
"Curves" adjustment and save. In the example photos you can see that I obviously didn't
hit the nail on the head (backgrounds are a bit off in contrast and hue), but I think I was
pretty close. ISO6400 images are pretty terrible, but the ISO3200's seem usable for
posting images online and possibly small prints. The shadowed areas look fairly grainy.

Enough background info, here's some sample shots:

Aperture: f/4
Focal Length: 100mm
Exposure Time: 1/320, 1/640, 1/1250
Image Stabilization: OFF

[BenJohnson]

Aperture: f/2.8
Focal Length: 70mm
Exposure Time: 1/160, 1/320, 1/640
Image Stabilization: OFF

[BenJohnson]

Aperture: f/5.6
Focal Length: 75mm
Exposure Time: 1/40, 1/80, 1/160
Image Stabilization: ON

[tim]

Who needs a 5D2, they look pretty good!

[The Moose]

Originally Posted by tim

Who needs a 5D2, they look pretty good!

I'm thinking the same thing I'll have to try this when I get some spare time.

[JulieNick]

I guess it goes to show that it's the glass that helps get these shots. Should try with the 75-300 and 18-55 kit lenses? lol

[tim]

Originally Posted by JulieNick

I guess it goes to show that it's the glass that helps get these shots.

How do you figure that? Glass makes no difference to noise.

[JulieNick]

Originally Posted by tim

How do you figure that? Glass makes no difference to noise.

can't get 2.8 f-stop with those lenses

[JoYork]

Just goes to show there's not THAT much difference in image quality between cameras. Get the exposure right and noise isn't so much of an issue, get the composition right and people will remark at what good photographs your camera takes

[JoYork]

Originally Posted by JulieNick

can't get 2.8 f-stop with those lenses

I fail to see how the noise levels would look any different at f/4 with a doubling of shutter speed per shot.

[JulieNick]

Originally Posted by JoYork

I fail to see how the noise levels would look any different at f/4 with a doubling of shutter speed per shot.

Depends on lighting situations and what one would be shooting. It would work if camera is on the tripod. But when shooting people in low light it's very hard to use a tripod especially moving children. From what I have read, the faster the glass in those situations, the higher the iso can go and not be too bothered with the noise.

[kirkt]

Just out of curiosity, have you done the experiment where you shoot your scene at, say, ISO 400, and then shoot at 800 and 1600 by adjusting the ISO setting on your camera; then repeat the sequence, only this time shooting ISO 400, then increasing the shutter speed in one-stop increments (like in your experiments above) and then "pushing" those underexposed images in post to the 800 and 1600 equivalent and comparing them to the "real" 800 and 1600 images? I chose those ISO values because most dSLRs have that capability.

Just curious to see how the post-processing "push" compares to the actual increase in sensor sensitivity associated with increased ISO. I would assume that in the post-processing method, you are accentuating the noise inherent at the base ISO you shoot, by shifting the histogram accordingly and revealing the noise in the shadow regions. When you actually increase ISO on camera, you can use the entire histogram to capture your scene, at the expense of more potential noise. At some point, with the post-processing "push", you run out of data in the shadow region (where most of your underexposed image is going to live) whereas, in the actual ISO boosted image, you will have much more latitude if you expose properly.

Interesting! I'd love to see the histograms of the images you posted above - I bet they get a little ugly. For small web images probably not a big deal - for prints and larger images, I am guessing you are going to see a bunch if different artifact (noise, posterization) as you increase the "push" amount. I would be interested to see how pushing the RAW version versus an 8bit JPEG would compare too. There is your homework, get to it! Use a tripod and fix your white balance.

Over and out.

Kirk

[BenJohnson]

I'll try pushing an ISO400 shot to 800 and 1600 and comparing to actual ISO800 and ISO1600 images, but I would suspect that the pushed photos will look worse than increasing the sensor sensitivity.

[tzalman]

Just curious to see how the post-processing "push" compares to the actual increase in sensor sensitivity associated with increased ISO.

In the early days of digital photography all increases in ISO were obtained with software manipulation. Nowadays it is only the "expanded" H settings in models xxD and higher that are done that way. When the camera makers learned to build a variable amplifier into the sensor module it was a big step forward. Aside from satisfying curiousity, there's not much point in reinventing the wheel.

[BenJohnson]

Originally Posted by tzalman

In the early days of digital photography all increases in ISO were obtained with software manipulation. Nowadays it is only the "expanded" H settings in models xxD and higher that are done that way. When the camera makers learned to build a variable amplifier into the sensor module it was a big step forward. Aside from satisfying curiousity, there's not much point in reinventing the wheel.

On the topic of pushing ISO400 to 800 and 1600 I agree. Just curiosity.

On my 400D, pushing to 3200 and 6400 are not options in camera (via sensitivity or otherwhise), so the original of intent of the post has value (to me at least).