Digital Signal Processing Reference
In-Depth Information
frequency or probability of each grey level in the one image histogram is counted.
The grey level value distribution of the image is offered in the histogram. In the coor-
dinates of the grey level histogram, the abscissa means the grey level of each image
pixel and the vertical coordinate means the frequency or probability of image pixel in
different grey level. Thus the image histogram can be adopted in the light source
detection.
(b)
The detection of light source outline
The shape-recognition method can be summed up for two kinds: one is a kind of dis-
cernment based on object border which include girth, angle, width, height, diameter,
etc. Its description method includes chain-yard, B-spline function, FDs, auto-
regression model (AR) and Hough-vary etc. Among them FDs is one of widely used
features. The other method that is a kind of discernment based on object coverage
areas which include area, roundness and moment features, etc. Its description method
includes Run-length Encoding, Quad-tree and Moment Descriptor etc. The moment
features is one of widely used features among them. Above-mentioned two kinds of
methods are mainly suitable for the discernment of the closed boundary or the region
of a plane.
The overall characteristic of image can be showed in moment features, which of-
fers different kinds of geometry features information. And 0-3 orders features of mo-
ment can describe the overall characteristic of image, and the high-order moment
includes more meticulous detail but is relatively sensitive to the noise. The moment
features has very important application in the image analyzing and discernment. Any
image can be regarded as a two-dimensional probability density function f (x, y), the
general definition of its moment Φpq is as follows:
Φ=
ψ
pq
xyfxydxdy
( ,
)
( ,
)
,
pq
,
=
0,1, 2, 3
(1)
pq
ξ
ψ
pq
xy
(, )
ζ
fxy
,
(, )
is the domain of
is moment weight function, it is
continuously.
In polar coordinates
(, )
r
θ
pq
++
1
Φ=
r
ψθ
() (, )
f r
θ
drd
θ
,
p q
,
=
0,1,2,3
(2)
pq
pq
ξ
The improved algorithms based on the border of region to calculate moment inva-
riants can be applied in the border of the close areas which is not overlapped. But in
practical application, overlap and cross in the structure is unavoidable and the method
of chain-yard to show the structure is often difficult. In addition, in order to unite
structure moment and regional moment, a new moment feature that is suitable to re-
gion, closed and unclosed areas has been proposed, which is defined relative moment
and used in shape-recognition. And the discrete calculating methods of relative
moment have been proposed.
