The power of shooting RAW....

I know that many of you already know and understand the power of shooting in RAW but I just wanted to show it firsthand. This is a fun photo I took of a youth basketball team where my lighting did not fire properly......it was obviously severely underexposed (my camera actually fired twice as I had it in continuous shooting mode so I did capture the photo with lights a split second before....the lights didn't have time to recycle before the second shot). Anyhow, out of curiosity alone, I wanted to see how much detail I would be able to recover in this shot.....and I was quite surprised. I will admit that it's not all that pretty and it does have plenty of noise along with poor lighting quality (only the gym lights).....but I was still amazed at the level of detail and color that still existed in such a poorly exposed photo, especially in the shadow areas. The first pic is as it appeared SOOC.....the second is my edit in an attempt to see just how well I could rescue it. Keep in mind, I'm using a 1DsMkII.....a great camera but it's quite long in the tooth as compared to what's out there now.

great example. good work.

Pretty cool, you might want to post that in our RAW Conversion Thread!

If you haven't checked out the thread, read the thread Intro for our "guidelines", check out some examples, and jump in!

The thing that a lot of people don't realize is that technically there is no bottom to the dynamic range of a CMOS sensor. DR is defined as having a bottom limit that is determined by the point where noise obscures photo detail; in other words, where the noise becomes objectionable. That's a kind of loose concept, one man's objectionable is another man's recoverable. (The DxO Mark web site, in an attempt to set a mathematical definition that can be used to compare cameras, defines it as when the signal to noise ratio is 1, but that mathematical construct is often out of step with the real world.) The fact is that as long as there is even the tiniest bit of light, even just one photon, the sensor will react by outputting voltage. (Actually, it even outputs a random pixel voltage even in complete darkness, the so called "black noise".) Of course as the input light gets weaker its part of the output is smaller and it is submerged in noise, but it is still there.

Is this why I've been advised that it's better to err on the underexposed rather than the overexposed side for Digital? Seems to make sense. (IIRC - I think someone said it was the opposite in the film days, but I wouldn't know myself).

Great example though to the OP! Thanks for sharing.

Wonder what the finished would have looked like had that started out in jpeg?

Elie I have not really delved into DxO mark but from a general Digital Signal Processing perspective SNR's of 1:20 are not that difficult to deal with as long as the signal you are looking for is well below the Nyquist frequency (The Nyquist frequency is the highest frequency any digital system can represent, and is half of the sampling frequency, so at worst you have 2 samples per cycle), and the noise is truly random. This is the big advantage of using DSP over traditional analogue systems, as you can deal with quite large negative SNR's.

Alan

adza77 wrote in post #15731247
Is this why I've been advised that it's better to err on the underexposed rather than the overexposed side for Digital?....

You may want to check out THIS thread.

Elfstop wrote in post #15731269
Wonder what the finished would have looked like had that started out in jpeg?

If you had the original file it might not be too bad, but if all you had to work with was the image in this thread you couldn't do much without serious degradation. I'd say that recovering this photo if it were shot in JPG wouldn't be that hard to do if you had the skills.

adza77 wrote in post #15731247
Is this why I've been advised that it's better to err on the underexposed rather than the overexposed side for Digital? Seems to make sense. (IIRC - I think someone said it was the opposite in the film days, but I wouldn't know myself).

Not really, if you value Image Quality then the best thing is to shoot Raw and "push" your exposure a bit "to the right" (check out the "Expose To The Right" (ETTR) method). In general, it's better to get more light, so that you get a better signal-to-noise ratio in those shadows, than less light.

This assumes, though, that you are not getting too much light so that you are "blowing" highlights.

But as the example shows, Raw also gives you "room to play" with the shadows.

The question was asked about how jpegs would handle those shadows. Well, when jpegs are produced in the camera, 8-bit compression and jpeg compression is applied, and unused Raw data is discarded. In the case of highlights and shadows, if they are considered "out of bounds", they will be clipped by the conversion (even if there is some "info" in the Raw data). If someone is curious, go ahead and try, shoot and underexposed scene in Raw+jpeg and compare how much you are able to recover!

WOW! Perfect example!

In fact, go ahead and download the jpeg that the OP posted, and give it a shot, if you think you can get results that approach the Raw recovery, post!

tonylong wrote in post #15731359
In fact, go ahead and download the jpeg that the OP posted, and give it a shot, if you think you can get results that approach the Raw recovery, post!

The image in the post is too small and too compressed, you'd need the OP to convert the RAW to JPG with all the settings at zero, then upload that file.

katodog wrote in post #15731377
The image in the post is too small and too compressed, you'd need the OP to convert the RAW to JPG with all the settings at zero, then upload that file.

The point is that the image posted was a jpeg with all the compression applied. The fact that it was resized for the Web, sure, more image was "tossed away", but you can still get a general idea of what you can recover in the shadows.

But if you really question it, take a jpeg that is equally unexposed and see (and post)!

At the posted size yo could probably print the image and it would be ok. But you can see the dreaded Canon pattern noise that plagues these cameras and limits the ability to get a useable image fro these situations. This striped noise is not random and is large enough that the pattern in which it appears is very distracting and difficult to remove.

Not to take away from the OPs treatment, but this is one issue that Canon need to address to make the these kind of images useable.
kirk