Fraser Watson
Lyndsay Fletcher
Silvia Dalla
Stephen Marshall
ISSI MEETING – 19th to 21st October 2009 – Bern, Switzerland
In continuum images, we are simply using the sunspot intensity contrast to make
decisions.
This causes difficulty in regions of the Sun where the background changes rapidly and
this is near the limbs.
This is a problem when looking at sunspots close to the limb
of the Sun.
Spots near the solar limbs are the most difficult to detect.
Spots near the solar limbs give the best examples of Wilson depression.
Two sunspots close
to the solar limb.
Simple techniques
such as thresholding
are next to useless
at these solar
longitudes.
The Structuring Element (SE)
Used as the probe in the image. Both the size and shape are crucial
for successful object detection.
The white dots indicate the ‘origin’ of the shape.
The Erosion operator
A is the image we are working on
B is the structuring element we have chosen
 means ‘subset’
Bx is the translation of B, by x
So an erosion can be described as the set of points where the
translation of B by x fits inside of the original image, A.
The Erosion operator
A 2D example
Erosion operations reduce the size of objects, and remove
protrusions smaller than the structuring element.
The Dilation operator
This is the dual operation to erosion and is defined in terms of the erosion operator.
means B is rotated 180 degrees about the origin
means a negation of the image (in a binary
image, all the ones become zeros and viceversa)
The Dilation operator
A 2D example
Image from Handson Morphological
Image Processing
(Dougherty and
Lotufo)
Dilation operations increase the size of objects, and remove
intrusions smaller than the structuring element.
The open top-hat transform
This is the operation that allows sunspot detection.
A ˆ B  A  ( A  B)
○-hat is the top-hat operator
○ is an opening, which is an erosion followed by a dilation
The full sunspot detection is done in three dimensions: the x and y coordinate of
the pixel, and the intensity of the pixel.
These three values create a dome shaped surface due to solar limb darkening,
with ‘spikes’ in the sunspot locations due to their lower intensity.
Original
Original
Original eroded by
circular structuring
element
Original eroded by
circular structuring
element
Previous dilated
with same
structuring element
Previous dilated
with same
structuring element
Previous subtracted
from original
This image contains 4 clear
sunspots.
3
1
4
2
5?
2 with well developed
penumbra and 2 without.
The spots at location 5 are
at a viewing angle of more
than 75 degrees.
2
1
3
4
5?

Can currently record the centroid location of
each sunspot in each image, the spot area, and
flux within that area in a corresponding
magnetogram if necessary.

Processing a full image on a standard desktop
machine takes ~ 4 seconds.

The tracking algorithm takes a list of sunspots in
two consecutive images and checks if any pair
are the same sunspot
IMAGE TWO
IMAGE ONE
6 hours later
SPOT 1
SPOT 1
SPOT 2
SPOT 6
SPOT 3
SPOT 2
SPOT 4
SPOT 3
SPOT 5
SPOT 4
No matches within two
degrees because the Sun
has rotated for the last six
hours
IMAGE TWO
IMAGE ONE
6 hours later
Rotation
model of
Howard,
Harvey
and
Forgach
(1990)
ROTATED SPOT 1
SPOT 1
ROTATED SPOT 2
SPOT 6
ROTATED SPOT 3
SPOT 2
ROTATED SPOT 4
SPOT 3
ROTATED SPOT 5
SPOT 4
This spot is still kept in mind!
Fraser Watson
Lyndsay Fletcher

To create an efficient, automatic method for
detecting magnetic elements in MDI
magnetograms.

To track each element throughout a time
series to determine the fragmentation and
breakup in active regions (possibly associated
with flaring).

MDI magnetograms from SOHO with a 96
minute cadence.

An image is split into two separate images,
one containing all the positive flux and
another containing the negative.

The magnetic elements in each image are
detected using a ‘downhill’ algorithm.
The ‘biggest’ pixel in the image is labelled as region 1.
1
Then, the second biggest pixel is examined. If
it is next to region 1, it is labelled as region 1,
otherwise, it becomes the starting pixel of
region 2.
1
Then, the second biggest pixel is examined. If
it is next to region 1, it is labelled as region 1,
otherwise, it becomes the starting pixel of
region 2.
1
2
This continues until a defined magnetic field
strength threshold has been reached (150 gauss,
although this is still in testing and would be better
if it was not constant but related to the data).
1
2
3
150 gauss

A list of all magnetic elements is produced and
currently contains their area and centroid.

Very small elements are excluded from the
catalogue (< 10 pixels in area).

This method detects small magnetic elements
and so may compliment Paul Higgins’ or Tufan
Colak’s work.

Takes around 5 seconds per magnetogram on a
standard desktop machine.

The magnetic elements are tracked in the
same way as I track sunspots but with a
smaller margin for error as the time cadence
is better (elements must be within 1 degree
after differential rotation is applied).
Some examples......
Image from April 2001
Image from August 2004
September 2001 - active regions
September 2001 - active regions and sunspots





The morphology algorithm detects sunspots with a high
degree of accuracy when compared to other catalogues
(no false detections, 4% false positive pixels when
compared to 100 images with human detection).
Speed can be improved by optimisation, or a faster
programming language.
The sunspots can be easily tracked and their evolution
studied.
Magnetic elements work still in progress and currently
lots of transient features appear, likely due to the
current simplicity of the algorithm.
Any ideas for improvements?