Recognition of Meter Values - Hierarchical Neural Networks for Image Interpretation - page 135

7. Recognition of Meter Values
The remainder of the thesis applies the proposed Neural Abstraction Pyramid to sev-
eral computer vision tasks in order to investigate the performance of this approach.
This chapter deals with the recognition of postage meter values. A feed-forward
Neural Abstraction Pyramid is trained in a supervised fashion to solve a pattern
recognition task. The network classifies an entire digit block and thus does not need
prior digit segmentation. If the block recognition is not confident enough, a second
stage tries to recognize single digits, taking into account the block classifier output
for a neighboring digit as context. The system is evaluated on a large database.
7.1 Introduction to Meter Value Recognition
Meter stamps are commonly used in many countries to mark letters. They are printed
by a postage meter that is part of a mailing machine. The postage meter prints the
stamp usually with red ink in the upper right corner of the letter, at the location
where otherwise adhesive stamps would be placed. In addition to the postage value
the stamp usually contains the identification number of the postage meter, the date
and location of sending, and possibly some advertisements of the sender. The sealed
postage meter keeps track of the postage used and must be refilled from the postal
company when the stored postage has been used. Figure 7.1 shows some historical
meter stamps from different countries.
The first meter machines were installed in Scandinavia and the United States of
America in the beginning of the 20th century. Figure 7.2 shows the Pitney Bowes
Model M postage meter from 1920 which was the first to be licensed by the U. S.
Post Office Department. Nowadays, about half the mail in the United States is me-
Fig. 7.1. Historical meter stamps from different countries (converted to grayscale).