I try to reduce the video from each observing session on that same day. It's almost the only way not to get hopelessly behind a back log of video tape. I use a Digital Origin IEEE 1394 Firewire card in a Dell Pentium II 350 MHz computer to control the VCR and to capture the images. The software I use is the Digital Origin PhotoDV plug-in for Adobe Photoshop LE. This software provides a very nice graphical interface except for a uselessly small preview window.
The preview window is so small that it is impossible to tell
anything about the fine detail in an image. To overcome this I
use a 9" B&W Sony video monitor and a Sony IFT-R20
cordless IR receiver. This allows me to control the VCR from the
computer but watch the image on the video monitor. I normally
play the tape back at normal speed until I see a promising
stretch of video. I then rewind the tape to the beginning of the
good section and go back through it in step mode, looking at
every frame. This is slow but, short of an automated grab and
analyze software package, it is the only way to find the instants
of good seeing. And instants they are. The following three images
illustrate the power of video. These images are three consecutive
interpolated fields. This means that they are consecutive
"half frames" that were recorded 1/60th of a second
from one another. Only the odd or even video lines are recorded
each 1/60 of a second and these two fields make up a video frame
(30 per second). The PhotoDV plug-in can be set to grab only a
field and "de-interlace" the field by various means.
The one that I find works best is "field interpolation"
where the missing scan line is created by interpolating the
values in the adjacent scan lines (for example, when grabbing an
odd field the even lines are created). It is pretty obvious that
the second image of the three below is better than the first and
much better than the third.
7/23/00 15:56:49 UT, Time code: 00:47:58;20 Odd
7/23/00 15:56:49 UT, Time code: 00:47:58;20 Even
7/23/00 15:56:49 UT, Time code: 00:47:58;21 Odd
I then stretch the histogram of all of the captured fields until the full range of values is used. The histogram for a well exposed large spot generally looks like the one shown below. The large peak to the right corresponds to the photosphere, the smaller peak to the right of center is the penumbra and the umbra is the very short wide peak to the left.
I scale the histogram until the input levels are stretched to the full 0 to 255 range.
I save this stretch as a file and apply it to all of the
captured fields. Applying this stretch to the three images above
produces the following:
7/23/00 15:56:49 UT, Time code: 00:47:58;20 Odd
7/23/00 15:56:49 UT, Time code: 00:47:58;20 Even
7/23/00 15:56:49 UT, Time code: 00:47:58;21 Odd
The differences between the images now become obvious. It is at this point that I cull the selections. I usually save only the best of the captured fields. I apply an unsharp mask with the following Photoshop parameters (determined empirically):
This, of course, dramatically improves the image.
7/23/00 15:56:49 UT Time code: 00:47:58;20 Even
Finally, I again adjust the levels to brighten the image a bit.
The final result is then ready to rotate if necessary (images
taken with the sun past local noon must be rotated 180 degrees to
put East to the left and North to the top because of my optical
set-up and German equatorial mount), annotate, and post.
7/23/00 15:56:49 UT Time code: 00:47:58;20 Even
Co-addition of multiple images of the same object is a well know technique for reducing the noise in images. Solar video would seem an ideal application of this technique. However, the images to be stacked must be of comparable quality and not distorted relative to one another for this technique to be effective. I have only had steady enough seeing for this to be true one time in the eleven months that I have been using video on the sun. I obtained a total of five very good fields within seven seconds when I recorded the field above. I processed all of them as described. Then I inserted them all into one Photoshop file as "layers" and registered them using the "exclusion" display. The co-addition was done using Photoshop's "merge down" function with the first field being the background, the second having 50% weight, and the third, fourth, and fifth fields having 33%, 25%, and 20% weight respectively. Interestingly, I found that the registration between individual fields and the co-added image varied by a pixel or two. I therefore registered the second image to the first and co-added them. Then I registered the third to the first co-addition before merging it, and so on. The semi-final image had better signal relative to the noise but some blurring due to differences in the seeing between the fields. Consequently, I again applied the unsharp mask filter. The final result is shown below.
The improvement from the best of the individual fields is subtle but real. The following image highlights some of the features that were most improved. Fragments of the co-added image are shown in the left-hand column. The corresponding fragments from the best individual field are in the right-hand column.
Copyright © 1999, 2000 Arthur L. Whipple, All Rights Reserved
Comments regarding this web site may be e-mailed to:
Page Last Updated: 31 March 2001