sub-pixel resolution vs intensity resolution

[arnieb]

The general rule-of-thumb for sub-pixel resolution is that with ideal back lighting, you can achieve 1/10 pixel resolution.   NI states that their algorithms can achieve 1/25 pixel resolution.   How does an increase in intensity resolution affect sub-pixel resolution?   I presume the rule-of-thumb was based on 8-bit (256 gray levels).   So what would it be if the image resolution is increased to 10-bit (1024 gray levels).  My gut feel is that 8-bit to 10-bit is a 1024/256 = 4 fold increase.  So that 1/10 becomes 1/40 pixel resolution.   Does anyone have any more enlightened thoughts or will they concur with me?

Thanks for any help,

[BruceAmmons]

It feels right to me.  It seems like the larger the difference between foreground and background intensities, the more sub-pixel resolution you can get.  With 8 bit the max range of the difference is smaller than the max range for 10 bit.

Now, the increase in sub-pixel resolution is questionable.  It may be a square relationship, so it might just double instead of quadruple.  That would still be a nice increase, though.

Bruce

[arnieb]

Here is a counter-argument from an associate of mine:

"I suspect going from 8 bits to 10 wouldn't make much difference. Usually when you're making measurements with sub-pixel resolution you're doing some
kind of line or curve fit over a multi-pixel region. Increasing the dynamic range isn't going to change the fitted curve very much. It might shift it a little
one way or the other, but it probably would be a very small percentage change."

Obviously, if you go from 1-bit to 2-bits or 2-bits to 3-bits you are doubling the resolution, but he's saying at some point there are diminishing returns.  It may be more like statistical sample sets.  I heard a rule-of-thumb for statistics, for example gage R&R studies, is that the data is meaningless unless there is at least a sample set of 25.  We know that if you flip a coin 10 times there is still a strong possibility to get 60% heads and 40% tails, but if you flip it 100 times the two outcomes will approach 1% difference and 1000 times will change the outcome less significantly.   Hence, the change from 100 to 1000 is much less than the change from 10 to 100.  (Essentially, the intensity value is an A-to-D conversion which is another form of sampling.)

[BruceAmmons]

That argument sounds reasonable too.  I tend to agree with both sides of an argument as they are presented - I would probably make a terrible juror.  I never would be able to win a debate, either.  My problem is I don't know anything about the subject, and both sides sound right.  I am much better at arguing a side when I know what I am talking about.

Bruce