. This paper 
on of image 
nique. 
ESS OF 
places” is 
ties of the 
amera. The 
y changing 
ntains the 
of intensity 
ty D(b) 1s 
he inverse 
. Then the 
(1) 
nce in the 
  
  
> fragment, 
othesis Hp 
lue on a 
f MSE on 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
wh 
A NN 
  
= wo ® 
^oodN 
175.5 
140.4| 
105.3 
70.2 
35.1 
  
  
420 840 1260 1680 2100 2520 2940 
  
WY 
SoaNo 
  
Figure 2. Analysis of statistical properties of noise by the 
“wedge” image: a — noise variance along the *wedge"; b — 
smoothed noise variance; c — changing of intensity along the 
“wedge” d — dependence of noise variance on intensity. 
Hypothesis Hy is equivalent for the following: 
Ho: f-&, 6& €N(uO'() 
For variance 
c? 
LION | = 2 2 
eV LS 2) 
ie (e 
(N - o? 
o (u) 
distribution with (N-1) degrees of freedom. So, Hp is equivalent 
to the following expression: 
BRUM ; o 2 
it is known, that the value satisfies to 0" — 
. QN - De: 
o (u) 
€ X’(N-1), (3) 
The estimation of value is formed as follows: 
M (4) 
+ 
N° 
Then, since statistics are calculated, the criterion of signal 
presence is tested as follows: 
If sZ X'(N-1 
then Hy hypothesis to be rejected. 
For N > 30 the quantile of r£ distribution can be estimated by 
formula 
721 
X, (m) - > am—1 +a). 
where Ot, - the quantile of normal distribution. 
Thus, the decision rule has the form: 
as > (ND ray 
which is equivalent to 
co a 
zz VUN- HA ol te. 
ou ER D" ( T) + ho - D 
So, the decision rule takes the form: 
i o>(1+ 
a 
JANCB pou 
then Hy hypothesis to be rejected. 
Because it is useful to process data sets with N>200, this 
criterion is applied in the following form: 
o>(1 xs E don) 
JIN 
where. C= Hos = 2.4. 
2 
So, the thresholding becomes adaptive. It depends both on the 
testing set amount, and on the statistical properties of the signal 
inside the image fragment. If one use the informative function 
1x0 yoN) = O(Xo yo N), 
then the threshold T is obtained as 
T=0 Lada (5) 
JN