[SciPy-User] Help optimizing an algorithm

[Charles R Harris]

On Thu, Jan 31, 2013 at 10:57 AM, Chris Weisiger <cweisiger at msg.ucsf.edu>wrote:

> On Thu, Jan 31, 2013 at 9:39 AM, Zachary Pincus <zachary.pincus at yale.edu>wrote:
>
>> I presume you've seen this article about some of the sCMOS cameras, but
>> if not:
>>
>> http://www.microscopy-analysis.com/files/jwiley_microscopy/2012_January_Sabharwal.pdf
>>
>> They mention the dual amplifier gain issues, and also point out some
>> potential trouble spots (toward the end in the "unexpected findings"
>> section) with the low-gain amplifier at least for the (unidentified) camera
>> they used. Worth knowing about...
>>
>>
> I hadn't seen that; shame on me for not doing my due-diligence. However,
> their plots look significantly worse than ours do, even if they're
> cherry-picking bad pixels. For comparison, here's our 7 worst (most
> nonlinear) pixels:
> http://derakon.dyndns.org/~chriswei/temp2/badPixels.png
>
> And here's our sensor-wide average low-end nonlinearity (note that camera
> baseline is 100 counts):
> http://derakon.dyndns.org/~chriswei/temp2/lowEndToe.png
>
>
So the light source is held constant here and only the integration time
varied? Due to pipelining, it is possible that polynomial fits might be as
fast as the linear splines you are using. In any case, a polynomial fit to
the inverse function could be used to sample the output to input conversion
at equally spaced output values and the result stored. With proper scaling
you could then determine the table index and offset for the interpolation
using divmod.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20130201/2d096a74/attachment.html>