(x)
th
—
X
na
i
a, cos 2wkx/M + .b. sin 2wkx/M + r
£
sin 2wkx/M + I(x)
k
uy cos 2mkx/M +
It is natural to choose k in such a way that the sine and cosine
functions are orthogonal. The coefficients a,, b,:; U, ànd v, are then
obtained from the Fourier series of f and g. The matching parameter is
obtained from these coefficients as
M
atk} = i L + a [arctg (b, /4,) - arctg (v, /u,)] (25)
Li. 0. le Arie Az
an expression, which is easily shown to be equivalent to (18). As a
consequence, digital image matching using phase shift methods are
biased by wrap-around errors as already shown.
k (24)
«A
>
u
ati (en
i WM ud Lit e T i n n ih n
: iid Vm : Alt i jun um ir |
CN MNA
ree ibm TR
un p xu 1 i (ur
|l jj i i | ai i ii iil did b iiid A Ab
refte mom";
Figure 2: Row vectors showing possible parallaxes.
Top of aJ, bl, «bh, id) Components Jof rowvectors for each k.
Bottom of al, 50k, xk, idi: $um of. all row. vectors.
a) Cyclically displaced sine curve and real image. N=M=64.
b) Noncyclically displaced sine curve. N=M-64.
c) Noncyclically displaced real image. N=M=64.
d) Noncyclically displaced real image window. N:128, M-:65*4.
- 645 -