[Elphel-support] FPN correction

[Andrey Filippov]

Jens,

those three LSBs were removed (and you are right - there was a typo - 3 bits
wide/ 2 bits explained) a while ago, that comment just remained from the
times before LUT-based gamma correction was implemented. The actual Verilog
code that processes that word is in the sensorpix353.v file:

http://elphel.cvs.sourceforge.net/viewvc/elphel/elphel353-8.0/fpga/x3x3/sensorpix353.v?revision=1.1.1.1&view=markup
258 always @ (negedge sclk) if (wfpn) begin  259 testmode <= wdd[10];
260 submode[2:0]
<= wdd[9:7];  261 mpymode[2:0] <= wdd[6:4];  262 wdth <= wdd[3];  263 end

To be honest - the FPN correction was neglected for some time - we did not
use/test it for a long time and I'm not sure that everything works. In any
case you'll need to prepare the 16-bit array (8 bit for scaled subtraction
and 8 - for multiplication) and load it to the video memory (it should
dimensions matching the acquired image ones).

The reason we did not use is that it is not so easy to measure the
background image without mechanical shutter - you need to have the same gain
settings, exposure time and the same temperature of the sensor during
background image acquisition as during the acquisition of the live image. Of
course, there are ways how to measure background images fro several
conditions (at leas for the short and long exposures) and then being able to
synthesize the background image fro the current conditions without actual
measuring it - i.e. just finding these two (or more) coefficients that
provide the lowest amount on high frequency components (or just minimal
image size) - the FPN noise of the pixels is has much more high frequency
components than the real ones.
Additionally - the short exposure FPN of the sensors we are using now is
very low, high is the one during log exposures ("warm" pixels, mostly). And
when you have long exposure - FPGA subtraction does not give you much of the
advantage compared to subtraction of the background image on the host PC -
this is our excuse for neglecting the FPN correction code, that may be
broken by now.

Andrey
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://support.elphel.com/pipermail/support-list_support.elphel.com/attachments/20101008/eaed0bd9/attachment-0002.html>