Jeff
Offline
• • • • • •
Upload & Sell: On
|
DrPablo wrote:
The point is that at the fringes of DR you may still have a significant spike on the histogram, but you lose the ability to discriminate the black from the white. This would obviously occur when the black is on the fringe of complete overexposure or when the white is on the fringe of complete underexposure.
I totally agree with this, however, you'd clearly have a complex set of data, i.e. you'd be able to differentiate 'detail' (in your case black vs. white) of different resolutions at different DRs. That's a lot of work! 
DrPablo wrote:
So 30 lp/mm would be reasonable, because it would require around half the resolving power of full resolution. I'd imagine that when you're operating around the noise floor, or when you're operating at the upper fringe of highlights, resolution of the finest details is lost (by over or underexposure) before resolution of coarser details.
Exactly.
DrPablo wrote:
I just have some doubts that the presence of a spike on the histogram (even when "rescued" in RAW) is a meaningful correlate of detail discrimination -- so what's the point of that spike in the highlights if you can no longer discern cloud textures? What's the point of that spike in the shadows if you can no longer discern shadow detail?
Make sense?
It does make sense, and before actually spending several hours conducting these tests on two bodies, I had precisely the same question. Now that I've done it, I can see why he designed the test as he did.
Think about it this way (and this is admittedly only what I got from it): If a channel is not blowing, you'll get some kind of 'curve' from it; if the channel is blown, you'll get a single line 'spike' in the histogram, as the RAW converter has been able to bring it back, but there is no detail contained in it. As you slide the white- or black-point slider, it immediately saturates the channel. If there is resolvable detail, as you slide it, you should get some sort of continuum across several levels of data before it saturates, indirectly indicating the ability to contain some sort of detail.
The hard part (and the part which you question) is: Is that detail rendered correctly, i.e. is it actually advantageous to the image, or at what point does advantage become disadvantage? Very difficult to answer, as all this rapidly becomes quite subjective to the viewer. For example, many people assert that you can blow one channel, and yet still have image detail from the other two. While this is absolutely true, is that detail that is left accurate, does it help the image (it probably depends upon which channels aren't blown).
Anyway, I completely understand your line of questioning, however after conducting the simple version of the test, I can see it's simplicity and objectivity is its inherent strength. Anything along the lines of conducting multiple tests at different resolutions would rapidly become incredibly tedious and complex. For example, with the 'test' image of the focusing target that I did, see that the RAW data now goes all the way down to the -11.0 stop limit of the chart. Defining where that data becomes garbage would not be fun using this method, lots time I don't have. Paul, you should try conducting the test yourself, it is actually quite enlightening, especially when it gets down to trying to choose what's 'acceptable'.
What I'm going to do at this point is now do a real-world test of my defined underexposure value, and see if that 'level' contributes to the image detail in the shadows (when opened up a bit), or if it is merely digital 'garbage'.
Regards,
Jeff
Edited by Jeff on Jun 25, 2007 at 09:21 AM GMT
|
