![](bg4.png)
We calculate the median point and area of a
character “a” in digital format. Then, we print it and
scanned and calculated the median point and area
again from the scanned image. Before calculating
the median point and area, we converted the scanned
image to binary image with various thresholds.
Table 1: Invariability during D/A and A/D transform.
thrs
median point
diff area diff
192
(149.12, 146.86)
2.33 260 128
128
(150.28, 148.44)
0.58 186 54
64
(150.02, 148.22)
0.76 152 20
32
(150.49, 148.31)
0.78 116 16
16
(150.41, 148.89)
0.34 108 24
8
(150.46, 148.75)
0.44 104 28
orig (150.08, 148.98) --- 132 ---
The result is shown in Table 1. From Table 1, we
can find that though the threshold varies (thrs), the
median point does not vary very much. On the other
hand, the area varies very much and the difference
of the area of the original image. There are twelve
characters whose area’s difference is less than 54.
Therefore we cannot distinguish these twelve
characters with area.
5.4 Invariability during the Ordinal
Change of Paper
To estimate the invariability during the ordinal
change of paper, we viewed the difference of the
median points’ change between before and after
folding of the paper. The experiment is done as 5.3.
Table 2: Invariability during the ordinal change of paper.
thrs median point diff area diff
192 (146.03, 144.40) 6.11 296 164
128 (150.59, 148.79) 0.53 164 32
64 (150.51, 148.59) 0.57 59 73
orig (150.08, 148.98) --- 132 ---
The result is shown in Table 2. In this
experiment, our median point shows near to original
median point. Therefore we can conclude that our
proposed method has good invariability during the
folding of the paper.
6 CONCLUSIONS
In this paper, we have introduced our feature
extracting method of paper document. Our feature
extracting method is based on the location of the
mean point of each dot, and is expected to be
applicable to home printers such as inkjets.
We have estimated the probability of the
collision and the uniformity of the distribution of our
feature extracting method. We have found that the
feature value extracted from one character is
distributed uniformly and do not collide each other.
We have also checked invariability during D/A
and A/D transform and we have found that almost
every character can be distinguished with our feature
extracting method even after D/A and A/D transform.
We have also found that our feature extracting
method is better than the area.
We have also checked the invariability during
the ordinal change of paper and found that our
method has enough and better invariability than the
area.
Therefore, we can conclude our proposed
method can extract desirable feature value.
REFERENCES
IT Strategic Headquarters, 2007. Priority Policy Program
2007. The Japan Government. Japan.
Ching-Yung Lin, Shih-Fu Chang, 2000. Semi-Fragile
Watermarking for Authenticating JPEG Visual
Content. In SPIE Security and Watermarking of
Multimedia Content II, pp.140-151.
Ke DING, Chen HE, Ling-ge JIANG, Hong-xia WANG,
2005. Wavelet-Based Semi-Fragile Watermarking
with Tamper Detection. In IEICE Trans. on
Fundamentals of Electronics, Communications and
Computer Sciences, Vol.E88-A, No.3, pp.787-790.
Suzaki M., Mitsui Y., Suto M., 2003. New Alteration
Detecting Technique for Printed Documents using Dot
Pattern Watermarking. In SPIE, Volume 5020, pp.
665-676.
Takahashi Y., Yamada T., Ebisawa R., Fujii Y., Tezuka S.,
2007. Research and Development of Dot Pattern
Technology for Commodity Printers. In Proc. of FIT
2007, the Forum on Information Technology, pp. 325-
326.
Takahashi Y., Yamada T., Ebisawa R., Fujii Y., Tezuka S.,
2008. Information Embedding Method for Home
Printing of Certifications. In Proc. of ICACT2008, the
10th International Conference on Advanced
Communication Technologies, pp.2116-2120.
Echizen I., 2007. Digital Watermarking Technique and its
Application. In IPSJ Magazine Vol.47 No.11, pp.1243-
1249.
ICE-B 2008 - International Conference on e-Business
432