Binarization of Matrix Codes - Hierarchical Neural Networks for Image Interpretation - page 156

(a) (b)
Fig. 8.2. Different meter marks: (a) Meter impression design used by Canada Post; (b)
StampIt design used by Deutsche Post AG.
Data Matrix is designed with a fixed level of error correction capability using the
Reed-Solomon [188] method. Reconstruction of the content is possible if less than
one quarter of the bits have been destroyed. Data Matrix supports industry standard
escape sequences to define international code pages and special encoding schemes.
It is used for small item marking applications using a wide variety of printing and
marking technologies. The code size is variable. Up to 2,334 ASCII-characters can
be stored in one symbol.
8.2 Canada Post Database
For the experiments described below, a variant of the Data Matrix code is used that
was selected for the electronic letter franking method which is offered by Canada
Post as an alternative to postage meter machines and stamps. Figure 8.2 shows an
example of such a meter mark, along with a StampIt meter mark used by Deutsche
Post AG. Both contain human-readable fields, as well as a Data Matrix symbol.
The meter mark can be printed either directly on the letter, such that it is visible
through the address window in the envelope, or it can be printed on the envelope
at the upper right corner where the stamp would be placed otherwise. In this case,
it is printed using fluorescent red ink to allow for automatic up-right placement of
letters. Some examples of the address window variant are shown in Figure 8.3(a).
They have a relatively high image contrast. The symbol consists of four quadrants
with 22 × 22 bits each, framed by black or alternating finder patterns. This allows
for the storage of 1,936 raw bits. Figure 8.3(b) shows example images from the
red ink variant of Canada Post meter marks. Here, each quadrant consists of only
18 × 18 bits. These images are considerably brighter and have a much lower contrast
compared to the address window code variant.
The code matrix contains information about:
- the meter value along with cryptographic key to ensure validity,
- the date of sending,
- the sender, such as the serial number of the meter machine, and
- the addressee, such as the zip code and a short form of the address.
Automatic reading of the Data Matrix code requires localizing and binarizing
the symbol, locating finder patterns, reading the bits, correcting for errors, and vali-
dating the result.
Here, the focus is on the binarization step only. Because of noise, difficult light-
ing, printing errors, and the low-contrast of red ink on dark paper, this problem is