Re: To David Soderman: A Little Test

[Kennedy McEwen <rkm@xxxxxxxxxxxxxxxxxxxx>]

>[Kennedy McEwen:]
>> Miscalibration would result in banding in either case.
>
>
>Miscalibration could cause irregular streaking in a direction
>perpendicular to the sensor array.
>
>It could not cause periodic banding in a direction parallel to
>the sensor array -- not by any mechanism that I can fathom.
>
>If I'm wrong, please explain.
>
OK, but first it is certainly true that the type of miscalibration that 
Paul & I were discussing would result in banding along the scan 
direction, not along the sensor axis, but that is not the only type of 
miscalibration.

In a three channel CCD and it possible that each channel is calibrated 
to a different reference level - resulting in different brightness or 
response for each channel.  This would give very noticeable bands 
running along the direction of the sensor.

The type of banding depends on the scan step sequence, and thick bands 
will occur if the scanner simply scans the 'obvious' way - ie. the motor 
scans the rows to fill the gap between sensor lines and then moves the 
head forward by three times that number of steps to repeat the process. 
In this case the miscalibration will result in bands equal to the 
distance between the sensor lines, becoming most visible in gradual 
intensity gradient images, such as skies etc.

If the distance between the lines is designed to be an integer number of 
lines plus a fraction equal to one over the number of lines used, then 
the motor steps forward a number of pixels at a time and the lines then 
interleave - resulting in bands along the sensor axis which are only one 
pixel wide.

Just as with the dark current calibration issue, this type of error is a 
well known and very widely documented problem with multichannel scanning 
imager designs!

To avoid this problem it is *traditional* in multichannel scanned 
sensors to use a process known as TDI - Time Delay and Integrate.  In 
this process the total exposure time for each channel is reduced by the 
number of CCD lines used.  Then the motor simply steps forward by a 
single pixel and exposes another line - and so on, until the second row 
of the CCD reaches the same part of the image that has already been 
scanned by the first row, n steps previously.  Now, instead of 
outputting the signal from all three channels, the data from the second 
line is added (integrated) to the output that was obtained n lines 
previously (time delayed) by the first channel.

This repeats on all subsequent scan lines until the point is reached 
where the third channel reaches the position on the image that has 
previously been scanned by the second channel, n steps earlier, and the 
first channel, 2n steps earlier.  Now the signal from the third channel 
is added to the time delayed summation of the second and first channel.

At this point, and only at this point, has the scan actually commenced. 
The image is then scanned line by line with the output of all three 
channels being delayed by n-steps and added to the subsequent channels.

The speed advantage of the multichannel sensor is retained, together 
with the sensitivity of the full exposure time at each pixel which, 
being shared equally by all of the lines in the sensor, are uniform, 
without banding.

Usually this TDI is achieved by having blanked off CCD lines between the 
active lines and transferring the charge from line to line, rather than 
along the line, which is only done on the final line of the sensor, 
however the entire process can be implemented digitally.

Most high performance military surveillance scanners use TDI to marry 
speed and sensitivity - indeed it was one of the earliest imaging 
functions that CCDs were ever used to implement, before they discovered 
that they could be made sensitive to light and image by themselves!  ;-)

So there are plenty of ways for miscalibration issues to cause banding 
in either axis - and their are well proven and documented solutions in 
virtually any optomechanical design textbook.
-- 
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers
-
Turn off HTML mail features. Keep quoted material short. Use accurate
subject lines. http://www.leben.com/lists for list instructions.