Image processing of the
26 February 1998 total solar eclipse
The 26 February 1998 eclipse was observed from Northern Guadeloupe.
A 55 mm diameter fluorite refractor with 440 mm focal lenght was used to
get wide field images of the corona. An equatorial mounting was used to
allow exposures up to 4s. The polar axis was set up in daylight, first
by means of compass and inclinometer and then refined with Bigourdan method
on the sun.
An Olympus OM 1 camera with the mirror locked up in the up position
was used to reduce vibration.
A 24 x 35 mm negative film (Kodak Ektar 100) was chosen because of
the better dynamic range than positive film. Exposure ranged from 4 seconds
to 1/250 s.
And most of all ... the sky was perfectly clear !
The negative film was scanned with a Nikon LS 2000 scanner with
the following settings :
- 2700 dpi resolution : the output file is 2498 x 3762 pixels wide (ie
2.97° x 4.49° field of view with 4.3 x 4.3 arcsec2
per pixel) ;
- 12 bits per color, for better resolution of the shades ;
- 16X multiscan : the output scan is the result of 16 successive scans
for better S/N ;
- cleanimage mode "on": this function proved to be very efficient to
reduce scratches and point like defaults of the film. No loss of resolution
was apparent.
The negative film was scanned as a "positive film" because of the lower
dynamic range of the scanner in the "negative film" mode. The colors were
"reversed" in positive and balanced with Photoshop.
The red, green and blue channels were processed separately with 16 bits
resolution with CCD image processing software (Qmips and AstroArt). The
main steps are described below.
a – Image registration and addition :
Given the pixel size (4.3 arcsec) and the relative motion of the moon with respect to the sun (0.5 arcsec/s), it is not possible to use the moon disk to register images taken in a time span longer than 9 seconds, unless this relative motion is properly taken in account (amplitude and direction).
This problem was handled here by registering the images with respect to the stars (the relative motion of the sun with respect to stars being only 0.04 arcsec/s). Seven 1s exposures images were registered using the two 6.5 magnitude stars of the field with one pixel accuracy in translation and 0.05° accuracy in rotation.
These registered images were then added to improve S/N (see Fig 1a) :
Note :
The earthshine can be seen in this 7 x 1 s composit (click on the image
to get a better view). It can also be seen faintly on the original negatives.
First, I was very upset, I thought it was instrumental reflexion ...
b- Application of a symmetric radial mask :
The purpose of the symmetric radial mask is to compensate the large scale radial decrease of the brightness of the corona. The mask was given the intensity profile of one of the darkest azimuth of the corona as measured from the previous composite image. This profile was smoothed with a gaussian filter to remove noise from the film grain.
There is a different mask for each different color :
This plot can also be understood as the brightness profile along one azimuth of corona as measured by the film and the scanner in the three channels.
The following step is to remove the mask (with a weight factor) from the composite image :
Eight images used : 4 x 1/30 s and 4 x 1/250 s exposures.
Main steps :
a - images registration (translation and rotation) with respect to
prominences and inner corona details,
b - addition of the 8 registered images to improve signal to noise
ratio.
Main steps :
a - Registration of the inner and outer corona composites with respect to prominences and corona details.
b - Addition of the inner and outer corona composites (Fig 2 a) :
Fig 2a : inner and outer corona composite (green channel)
Note :
This image looks darker that Fig 1c. There is no mistake. Remember this image is the sum of two composits. In the first one, the inner and outer corona are visible, but on the second one, only the inner corona. So, the relative value of the outer corona to the inner corona is much lower when the two composits are added. The absolute value remains the same as in Fig 1c.
c - Application of a non symmetric radial mask (Fig 2b and 2c) :
The purpose of the non symmetric radial mask is solve this problem,
ie to reduce the large scale angular and radial brightness variations of
the inner corona, while leaving the outer corona unaffected. Basically,
the mask is computed by giving to each pixel the average intensity of the
pixels at the same distance from the solar center and in the same angular
sector :
pixel value (r, q) = Max (average
{ pixel value (r, q ') } - offset , 0.)
with -a °< q
- q ' <+ a °
where
r = distance of the pixel from the sun center,
q = pixel azimuth,
a = setting parameter (best
result was achieved with a =17°),
offset = setting parameter to limit the action of the
mask to the inner corona.
A gaussian filter was applied to the mask to remove noise from the film grain.
There is a different mask for each different color. The profile of the "green" mask is plotted below along the equatorial an polar azimuths :
Fig 2b : non symmetric radial mask (green channel)
The mask was then removed (with a weight factor) from the previous composite :
Fig 2c : inner and outer corona composite minus non symmetric mask (green channel)
The surface of the moon can been seen in the previous composit images
(Fig 1a - 1c - 2c), but details are blurred because registration was made
with respect to the stars in order to have a sharp corona. To improve resolution
on the moon, the seven 1s exposure and two 4s exposures (unused for the
corona composit) were registered with respect to the moon disk with 0.2
pixel accuracy and added :
Then, a soft unsharp mask was applied to improve contrast :
The resulting "sharp" image of the moon was "inserted" in the fully
processed image of the corona (Fig 2c) in lieu of the "blurred" image of
the moon. This last step was performed with Photoshop.
The earthshine and the main lunar features can be seen very clearly.
More than 14 stars are visible from magnitude 6.5 to fainter than 8.
Original paper published in Pulsar January 2000 (french langage)