Information Technology Reference
In-Depth Information
10000
3500
TRN
TST
TRN
TST
3000
2500
1000
2000
1500
100
1000
500
10
0
0
2
4
6
8
10
12
14
16
18
20
0
2
4
6
8
10
12
14
16
18
20
(a)
(b)
iteration
iteration
Fig. 8.21.
Recall of iterative low-contrast Data Matrix binarization over time: (a) average
squared output change; (b) average squared output difference to desired output.
adaptive thresholding
iterative binarization
original
degraded
original
degraded
first run
672 (96.8%)
435 (62.7%)
678 (97.7%)
673 (97.0%)
second run
19 (2.7%)
13 (1.9%)
16 (2.3%)
20 (2.9%)
sum 691 (99.6%) 448 (64.6%) 694 (100%) 693 (99.9%)
Table 8.1.
Low-contrast Data Matrix recognition performance: Number of recognized code
images.
the second run is able to reduce the number of rejected images considerably. For
example, all 20 original examples binarized using the iterative binarization method
which were rejected by the first run can be recognized in the second run. In the
sum of both runs the iterative binarization performs slightly better on the original
images than the adaptive thresholding, yielding perfect recognition, compared to
0.4% rejects. On the degraded images its performance is much better than the one
of the adaptive thresholding method. Only one example is rejected, compared to 246
rejects.
Figure 8.22(a) plots the recognition error of iterative binarization against the one
of the adaptive thresholding method. One can observe that there is some potential
for combining both methods since there are examples where one method has a high
recognition error while the other has a low one.
As summarized in Fig. 8.22(b), the recognition error of iterative binarization is
lower for the original images than the one of adaptive thresholding. It is much lower
for the degraded images where the recognition errors differ by a factor of more than
twenty.
Thus, while the iterative binarization method performs only slightly better on the
original images than the adaptive thresholding method for the red ink low-contrast
variant of the Data Matrix codes, it outperforms adaptive thresholding dramatically
on the degraded images.
Search WWH ::
Custom Search