image
as the
ed the
e. Not
ning
ion of
mical
. with.
acy of
a star
'acies,
| from
| main
yds of
image
| more
| of mass
of area.
st taking
intensity
1t to one
mine the
n, 1994;
mate the
) fitting.
r around
seful to
related to
1e values
13). If an
B (x. y)
ion give
nages the
Gonzalez
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
The total intensity of the image is given by Bo :
We find that moments depend on the intensity or
grey level. One important thing about moment
features of objects is that they can be used
regardless of location in the image and size of the
object. Image moments include the following:
centre of mass, variance and orientation.
CENTRE OF MASS: 1f we regard the intensity
or grey level B (x,y) at each point (x,y) of the
given image (B) as the ‘’mass“ of (x,y), we can
define the centroid (centre of mass) as well as the
other moments of B. It is that point where all the
mass of an object could be concentrated without
changing the first moment of the object about any
axis. The centre of mass is the centre of area of a
figure, which in practice is chosen as the position
of the figure. In the two dimensional case, the
centre of mass is given by (B10, Bo1)-
By, = >. 2 x B(x.y)/By Q)
Bu = 2 2.9 B(x,)/Bw
VARIANCE (c | : This is given by the second
s C
moment about the centre of mass (8 KL )
2 : : s
O'. Characterises the extension or spread of the
object in the x direction.
From the above equations the position and
variance of an object can be calculated using first
and second moments only. These moments are
invariant under linear co-ordinate transformations
(Jain 1989 pp. 381). Thus we do not need the
original image to obtain the first and second
moments. Projections.of the image are sufficient.
This is of great interest since the projections are
more compact and suggests faster algorithms
(Horn 1986 pp.54). Existing literature gives the
accuracy of the simple moment centroiding
algorithm to be between 0.05 to 0.1 pixel
(Schildknecht 1994, Chubunichev pp.152-153).
2.2 Point spread function (PSF) Fitting
The basic principle in this technique is that all
images of stars on a CCD frame have, baring
distortion introduced by the camera optics; the
same form but differ from one another in
intensity or scaling ratio and position. (Teuber,
1993). Thus fitting a suitable defined PSF to a
series of images will give relative magnitudes
m — zpt — 2.5 log(scaling ratio) (4)
613
where zpt is the magnitude assigned to the PSF.
Mathematical curves are fitted to the real data until a
good match is obtained. The parameters determined are
the position and the scaling ratio. The mathematical
equation used to model the profile of the stellar image is
usually the Gaussian function (Buil, 1991).
2
I(r) =1(0)exp| — (5)
where:
r is the radius with respect to the centre of the
star image
oc 1s a parameter characterizing the stars spreading
I(0) 1s the maximum intensity
Linearising equation (10) we get
In 7(r) 21n 1(0)+ br’ (6)
The equation above is of the form Y = A + BX. The
coefficients (A, B) can be estimated by linear regression
from (X, Y) pairs.
S X S Xy S XX S y
d= 2
S HS C
Bs SxSy —nS yy
Sy — nS yy
and
>
I
X
Sy = SS DX; IIT mS X. Sy
il i-l
n
SPEM XE.
i=l
Expressing r in terms of image co-ordinates gives
2
m2 = (x =x)" + (v= vo) (8)
Replacing r in equation (11) by equation (13) we get.
in 7(;)-tn (0) ds -x) (7) | ©
The above equation can be written as
E, = GC, + EX, t OY; +ex/ +e,x} (0
where E; = In I(r;) + In I(0), i is the pixel number and
(x;,ÿ;) are pixel co-ordinates. The centroid, (Xo,ÿo), Of the
image is computed after solving for the coefficients, c;
(j=0...4) of equation (10) using the least squares
criterion. Differentiating (10) and using the necessary
conditions for extremum gives;