Database Reference
In-Depth Information
Table 6.1 Comparison of
rotation parameters for image
registration for real values of
rotation parameter (
ˆ
ˆ
ʸ 0 ˁ m
0 0 0.00 90 90.5 2.49
5 5.0 0.02 95 95.5 2.62
10 10.1 0.06 100 100.6 2.75
15 15.1 0.12 105 105.6 2.90
20 20.1 0.20 110 110.6 2.99
25 25.1 0.30 115 115.6 3.07
30 30.2 0.41 120 120.7 3.18
35 35.2 0.54 125 125.7 3.30
40 40.2 0.67 130 130.7 3.44
45 45.3 0.80 135 135.8 3.66
50 50.3 0.97 140 140.8 3.77
55 55.3 1.14 145 145.8 3.90
60 60.3 1.33 150 150.8 4.04
65 65.4 1.54 155 155.9 4.17
70 70.4 1.74 160 160.9 4.30
75 75.4 1.96 165 165.9 4.43
80 80.4 2.15 170 170.9 4.57
85 85.5 2.33 175 176.0 4.73
The table also shows the mean square
error of image registration (
ʸ 0
ʸ 0
ˁ m
ʸ 0
ʸ 0 )
versus the estimated values
( ˆ
0 ) obtained by phase
correlation
ʸ
ˁ m )
comparison of rotation parameters between the real value and the estimatedvalue
for all samples, which shows that the error rate for parameter estimation was
ˁ p =
37 %.
Table 6.2 shows the experimental result for registration with scaling parameters.
The proposed algorithm estimates all 20 scale factors at the error of
0
.
ˁ p =
0
.
01 %.
This causes the mean square error for image registration to be
ˁ m
5
.
3% for all
samples.
According to the above results for all three cases, it can be observed that the
estimated parameters for translation for the registration of the 480
640 image were
the perfect result. However, the estimation of the rotation and scaling parameters
produced error rates of less than 4.73 % and 5.3 %, respectively. The factor that
influenced the rotation error arose because the dimension of the two images was
not rectangular, as assumed by the estimation process described by Eqs. ( 6.24 )
and ( 6.30 ).
×
 
Search WWH ::




Custom Search