Dynamic Range Really isn't as important as you are lead to believe.

Started 4 months ago | Discussion thread
Anders W
Forum ProPosts: 10,749
Like?
Re: Dynamic Range Really isn't as important as you are lead to believe.
In reply to _sem_, 4 months ago

_sem_ wrote:

Anders W wrote:

_sem_ wrote:

Anders W wrote:

_sem_ wrote:

Anders W wrote:

Cameras don't clip shadows. Cameras with poor DR will render the shadows with a lot of noise.

I think they do, and that you can see posterisation due to this if heavily lifting dark shadow areas with smooth gradients. Except that nobody wants to do this in practice because it looks ugly due to the noise and "sparse" shadows raw data. I mean this is relevant in "forensics" when you are trying to recover the last trace of detail DxOMark claimed to exist, not art photography.

No cameras don't clip shadows (except in the sense that they set the black-level offset, which, however, you can move if you want on some cameras, e.g., the E-M5). Posterization is due to quantization errors, not clipping. As I pointed out in my second reply to the OP above, the signal is merely drenched by read noise at the very bottom of the scale. If you can reduce that noise, e.g., by merging a large number of photos, you can record any signal, however weak. If there are no photons at all, you will of course only record complete darkness, as should be the case.

Well they do. This clipping is the ultimate step of quantisation. It is true that the clipping point is also masked by noise. If one made an artificial image mathematically and apply similar quantisation and clipping, then lift the shadows, clipping would be more evident. Most raw converters don't let one lift shadows so much.

Clipping and quantization are different things. On my camera (E-M5), an ADU-level of zero in the RAW file corresponds in terms of signal levels to complete darkness, as should be the case. That's not clipping.

They certainly are different. But the shadows end shows similar issues as the highlights end if you push it high enough, in addition to the noise and the quantisation (relatively "sparse" coverage of the data, due to the relatively linear nature of the sensor but highly nonlinear human perception).

Zero in the raw file is not complete darkness, merely the edge of ideal sensor sensitivity range (in the absence of noise). You can have images with a spike in the raw histogram near zero (not as clear as at the HL edge due to noise). For instance, in case of blue sky at low exposure, so that that goes very dark towards an edge, you can locate bands where the R and G channels reach the zero (clipped plus some noise) region, while B remains obviously positive. If your raw converter lets you push this area so high that it becomes visible, you can notice the edge where you get the fake almost clean blue (save the noise). But at such amplification you may also spot bands due to quantisation. You can use the eyedropper tool to tell which is which, and to distinguish form the effects of lousy 6-bit laptop displays (possibly emphasized by display calibration software); some raw converters also have clipping mask tools for both highlights and shadows.

Yes, zero corresponds to complete darkness (i.e., complete lack of photons hitting the sensor during the exposure). If I shoot my E-M5 with the lens cap or body cap on (at 1/4000 to be on the safe side), the average ADU is equal to the value of the black-level offset.

In your example with the blue sky, R and G wouldn't be zero, merely as close to zero as to be practically indistinguishable from zero. If you merged a lot of shots of the kind you describe, you would eventually be able to distinguish the non-zero R and G signals.

The existence of an absolute zero-point, as we have in this case, is not clipping.

Edited 4 months ago by Anders W
Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow