‘Therefore a window is defined in a very homogeneous area of the image. 
Within this window mean values of the lines are computed. A functional model is ob- 
tained by parameterization of these mean values. For the common case of a continuous 
two-dimensional image signal s(x,y) this parameterization is described by 
Yı 
; S(x.y) dy 5: f:(a12:7.:: s Ayaı X). for xe [x95 X11 (9) 
0 : 
Grauwert 
  
80 4 
60 | 
40 - 
  
20 + 
Fi 
  
0 T T T T T T Zeile 
80 | 
60 Pa NE 
40 - 
  
  
  
  
  
ins EC eT Tr 20 T T T T T 7 Spalte 
0 80 160 240 320 400 480 560 
  
Fig. 8: Scene after FFT-filtering Fig. 9: Profile of image row (upper) 
and line 
with [Xo, yg, X1, y1] defining the window coordinates. For a discontinuous image 
signal integration has to be replaced by summation. 
Fig. 10 and 11 show the thermal image with the used window and the line mean 
valües within this window. : 
  
A linear formulation is used for the approximation of function f: 
f(x) = a, tax 
  
Fic 
Parameter a, and a, are computed using least square fit. The SRF is obtained by the 
residuals 
Yi The & 
SRF (x) = a; + a, x = J sQuy) dy (10) meas. 
Ya An ir 
tion 
Now the whole image is filtered with the SRF according to (8). Fig. 12 shows the far 
result of filtering. In fig. 13 the same column as in fig. 9 is presented. Scan- reci 
line noise has nearly almost disappeared. d qu 
.140 
ELM Hi QUE oat 
ER Rs maa cs AN GS NL  S 
EEE