Note: Descriptions are shown in the official language in which they were submitted.
CA 02369790 2002-O1-31
GR 2001P01774
Description
Method and apparatus for reading dispatch addresses
with postcodes
The invention relates to a method and an apparatus in
accordance with the preambles of the independent patent
claims 1 and 5.
Systems for the automatic reading of addresses (OCR)
are well known in the field of letter processing and
are described e.g. in DE 195 31 392. Processing rates
of 10 letters per second, i.e. 36,000 letters per hour,
or more can be achieved with modern OCR letter sorting
installations. However, the recognition reliability
varies greatly with the script and overall quality of
the address information applied to the surface of the
letters. In the case of successful recognition, the
relevant letter can be provided with a machine-readable
bar code. This bar code allows further mechanical
processing up to an arbitary desired sorting order. In
particular, the use of bar codes enables the letters to
be sorted up to the sorting level of the mail delivery,
in which letters are sorted in accordance with the
order of their distribution by the deliverer.
Since the recognition rates of the automatic reading
systems vary greatly, it is necessary to assist the
latter by various forms of manual intervention.
In this case, the entered address is coded and
converted into a machine-readable sorting bar code,
which is applied to the dispatch, by means of a
directory.
To that end, in video coding systems, video images of
the dispatches are presented to operators for coding
purposes. This can be done online and offline. In
CA 02369790 2002-O1-31
GR 2001P01774 - 2 -
online video coding systems (OVS), the video image is
shown to the operator, while the physical dispatch is
held in delay paths. In these delay paths, the dispatch
is normally kept in motion for a time duration which
suffices for the OVS operator to input the required
sorting information for the relevant image. The
customary delay paths allow a delay of between 10 and
30 seconds. The longer the delay path, the greater the
costs and also the requirements for maintenance and the
physical size of the installation.
When using OVS, the available time often suffices only
for careful inputting of the postcode (PC).
In offline coding systems, as described in
US-A 4 9 92 649, dispatches with addresses that are not
recognized are provided with an additional item of
information, a tracking identification (TID). The
dispatches that are not recognized are stored
externally, while the images of these dispatches are
presented to operators for coding purposes, there not
being any temporal restrictions. The dispatches are
subsequently presented to TID readers. The TID is
linked with the entered address information. On the
basis of this, it is also possible to apply a customary
bar code sorting information item to the dispatch, so
that the relevant dispatch can be processed like
customarily OCR-read dispatches. Although the offline
video coding method represents an effective method for
coding all address constituents, capacities for the
further processing of dispatches provided with non-read
addresses and correspondingly complex logistics are
additionally demanded.
In order to shorten the reading times further,
WO 97/49503 proposed that the image of each complete
address information item that is not unambiguously
recognized within a specific time by means of video
CA 02369790 2002-O1-31
GR 2001P01774 - 3 -
coding is fed with the information about recognized
address parts that is obtained during the video coding
to the OCR unit for a further automatic evaluation for
address interpretation purposes.
This method is used in the case of addresses in which
only the place of destination is also applied as a
readable postcode (e.g. German Postleitzahl) on the
dispatch surface.
However, there are also countries in which further
parts of the address are represented in the postcode,
e.g. a number of small streets, medium-sized streets or
parts of larger streets. If the result of the OCR
reading process is not unambiguous, then the
corresponding image with the address is transmitted to
a video coding station for the purpose of video coding.
The invention specified in the independent claims 1 and
5 is based on the object of providing, in the context
of postcodes used which identify the addresses up to
part of a larger street or a few small streets, a
method and an apparatus which ensure that the address
is determined as quickly and as unambiguously as
possible in an interactive reading process with an OCR
unit and at least one video coding station when the
sender has specified the postcode incorrectly.
If the read result of the OCR unit is not unambiguous,
then firstly the postcode is video-coded. Then, with
this postcode, the streets) or the street region is
determined in a corresponding address database and
compared with the street read by the OCR unit. In the
case of correspondence, the video-coded postcode is
accepted, and, in the case of non-correspondence, the
video coding of an extraction code (AKS code) of the
image is effected by one or two operators, if this has
not already been carried out in previous steps.
CA 02369790 2002-O1-31
GR 2001P01774 - 4 -
Additional reading processes are then performed in the
OCR unit with that part of the postcode which
identifies the streets/street part and is shortened by
one position and with the different AKS codes. An
incorrect postcode may be either an incorrectly chosen
postcode or a postcode written unclearly or confusedly
in a number of positions. If the read results are not
unambiguous, then further reading processes are
performed in the OCR unit with that part of the
postcode which identifies the streets/street part and
is additionally shortened by a further position and
with the different AKS codes. This additional position-
shortening and reading in the OCR unit is continued
until the read result is unambiguous or only a few
candidates remain, the correct one of which is selected
by means of selection coding.
Advantageous refinements are represented in the
subclaims.
In an advantageous refinement, in the case of a
postcode which relates only to the part of a larger
street, during the validation of the video-coded
postcode with the read result of the OCR unit,
comparison is additionally effected as to whether the
house number specified in the read/partly read address
is situated within the house number range determined
from the postcode.
It is advantageous for the additional reading processes
in the OCR unit with a shortened postcode using the AKS
codes, to begin the shortening with the least
significant position. If no unambiguous result is
obtained, the postcode is additionally shortened by the
next less-significant position, etc.
This order of shortening was chosen on the basis of a
statistical error frequency, according to which the
CA 02369790 2002-O1-31
GR 2001P01774 - 5 -
most frequent errors occur in the least significant
position of the postcodes, the second most frequent
errors occur in the second least-significant position,
etc.
If the street read by the OCR unit does not correspond
to the streets) or street region derived from the
video-coded postcode by a search in the address
database, then it is advantageous, for determining
whether or not the postcode situated on the dispatch is
correct, to perform, at one or two video coding
stations, extraction codings (AKS coding) with which
the distribution point is characterized (e. g. the first
two letters of the street and the house number).
The AKS data of the first operator are then compared
with the streets/ street regions derived from the video
coded postcode in the OCR unit. If they match one
another, the postcode is accepted; if not, the AKS data
are video-coded by a different operator and compared
with the streets/street regions derived from the
postcode in the OCR unit. If they match this time, the
postcode is accepted; if not, the postcode is video-
coded by a different operator than in the case of the
first postcode video coding. The twa postcodes are then
compared and, in the case of correspondence, the post-
code situated on the dispatch is identified as
incorrect and, in the case of non-correspondence, the
AKS code or codes is or are compared with the
streets/street region derived from the postcode video-
coded for the second time in the OCR unit. If the codes
do not match in the case of any comparison, then the
postcode situated on the dispatch is classified as
incorrect.
The invention is explained in more detail below in an
exemplary embodiment with reference to the drawing.
CA 02369790 2002-O1-31
GR 2001P01774 - 6 -
In the figures:
FIG. 1 shows a diagrammatic illustration of an
apparatus for reading dispatch addresses
with postcodes,
FIGS 2a-d show the method sequence using a flow
diagram,
FIGS 3a-d show a diagrammatic illustration of the
method sequence in the case of a postcode
which is classified as incorrect.
FIG. l shows a diagrammatic illustration of a letter
distribution installation with which the method
according to the invention can be carried out . The OCR
letter sorter 100 comprises a feed device 110, which
progressively removes dispatches from a magazine and
transports them at approximately 10 dispatches per
second to a high-resolution video scanner 120. The
dispatches are subsequently transported in a delay path
121. The dispatches usually have address information on
their surface. In an OCR unit 130, the address
information of the images of the dispatches which were
obtained by the video scanner 120 is evaluated. In the
case of a complete evaluation, a bar code printer 150
is activated and the dispatch is provided with a
corresponding bar code for the subsequent sorting into
sorting compartments 160. The OCR unit 130 comprises
one or more microprocessors 131 with associated memory
132 for storing images of the dispatches. Furthermore,
the OCR unit comprises an address database 134 with
postcodes, city names and street names and possibly
further address-related information. During the
evaluation of the images having address information, a
feature-controlled reduction of the entry obtained from
the address database 134 is effected., so that a type of
part-dictionary is produced. Credibilities are assigned
CA 02369790 2002-O1-31
GR 2001P01774 - 7 -
in individual entries, so that a number of data of
correctly recognized addresses are produced during the
evaluation. The apparatus furthermore contains an image
controller 170 and also a number of video coding
stations 200, which are connected to the image
controller 170 directly or by a local area network
(LAN) 171. If the OCR evaluation of an image was not
unambiguous, i.e. successful, this image is transferred
from the OCR unit 130 to the image controller 170,
which, on the one hand, controls the TID bar code
printer 151 and, on the other hand, transmits the
corresponding image to one of the video coding stations
200. The TID bar code printer 151 applies to the
corresponding dispatch an identification code TID which
makes it possible, at a later point in time, to link
the evaluated address information with the physical
dispatch. In this case, the images are preferably
evaluated offline, although in principle, with a
sufficiently long delay time, an online evaluation
through video coding is also possible. In the latter
case, the TID can also be applied to the dispatches at
a later point in time, i.e. if the video coding has not
led to a complete evaluation within a specific pre-
determined time.
In order to carry out the method explained in more
detail below, the image controller 170 is designed in
such a way that address information that is not
completely evaluated by the video coding is fed, using
the results of the video coding in the OCR unit 130, to
a further automatic address evaluation.
The recorded and stored image of the respective
dispatch with the address, containing a postcode which
identifies the place of destination and the street, the
part of a street in the case of large streets, or a
plurality of small streets, is fed to the OCR unit 130
for a first automatic reading attempt 300. In this
CA 02369790 2002-O1-31
GR 2001P01774 - 8 -
example, said postcode is a 6-position code in
canonical form with the following character sequence
ANA NAN, where A is a letter (alphabetical character)
and N is a number (numerical character).
The concrete embodiment of the postcode is not, of
course, restricted to this example.
If the address can be read unambiguously in the OCR
unit 130 in the first reading attempt 300, the
recognition task is achieved. If not, the postcode is
video-coded 310. Afterward, the street read by the OCR
unit 130 is compared 320 with the streets) or street
region derived from the postcode. This is also
advantageously effected in the OCR unit 130. If the
details match one another, then the video-coded
postcode is accepted. If not, an extraction code (AKS
code), comprising the first two letters of the street
name and the house number, is generated 330 from the
image with the address at a video coding station 200.
These AKS data are then compared 340 with the
streets/street region derived from the video-coded
postcode with the aid of the address database. If they
match, i.e. if the first two letters input in the
course of the AKS coding correspond to the first two
letters of the street determined from the postcode or
to one of the streets determined from the postcode and
if the house number, if appropriate, lies in the range
of numbers of the street determined from the postcode,
the postcode is accepted. If not, a further AKS coding
of the same address is performed 350 at another video
coding station 200 and these AKS data are likewise
compared 360 with the streets/street region derived
from the postcode. In the case of correspondence, the
postcode is accepted; in the case of no correspondence,
the postcode is coded 370 a second time by a different
operator than in the case of the first video coding of
the postcode at a video coding station. Next, the two
CA 02369790 2002-O1-31
GR 2001P01774 - 9 -
postcodes are compared 380. If they correspond, then a
coding error is not present, so that the postcode
situated on the dispatch must have been written
incorrectly (illegible or confusable characters). In
this case, the least significant position is removed
from the postcode and is fed together with the AKS data
of the first AKS video coding to the OCR unit 130 for a
further reading attempt. If the attempt is not
successful, another reading attempt is performed 390
with the reduced postcode and the AKS data of the
second AKS video coding. If this reading attempt does
not produce an unambiguous result either, then the next
less-significant position is additionally removed from
the postcode and further reading attempts are performed
400 with the AKS data, as described, in the OCR unit
130. This procedure (reduction of that part of the
postcode which identifies the streets/street region by
a further position and new reading attempts) is
performed until the address is read unambiguously or
until only a few candidates remain, from which the
correct address is then selected 410 simply and quickly
by means of selection coding, e.g. by clicking on the
correct street name.
If no correspondence could be noted in the case of the
comparison of the two postcodes 380, then the first AKS
data are compared 420 with the streets/street region
derived from the subsequently video-coded postcode (2nd
postcode). If they match, the 2nd postcode is accepted;
if not, the second AKS data are compared 430 with the
streets/street region derived from the second postcode.
If the data match, the video-coded 2nd postcode is
accepted. If there is no correspondence present, then
the above-described process of performing further
reading'attempts in the OCR unit 130 with the AKS codes
and the shortened postcodes is performed. In this case,
the procedure begins with the first postcode 440, and
if the address is not identified unambiguously, with
CA 02369790 2002-O1-31
GR 2001P01774 - 10 -
the second, subsequently video-coded postcode of the
relevant dispatch 450.
The method is additionally explained briefly in FIGS
3a-d.
In accordance with FIG. 3a, the following address
details are situated on the dispatch:
167 Hunt Street
Ajax, ON LIS 1P6 (Postcode)
With the postcode LIS 1B6, Karl Street was found as an
entry in the address database 134. In the course of the
AKS coding of the address situated on the dispatch, the
operator entered: 167 HU. As can be seen, these details
do not match Karl Street. Since it has already been
determined in previous steps that the postcode is
incorrect, in accordance with FIG. 3b the last position
of the postcode is removed and corresponding entries
are sought in the address database 134 with the
postcode LIS 1P and the AKS code 167 HU. In this case,
the four entries specified were found which, however,
do not correspond to Hunt Street. Therefore, a further
position was removed from the postcode. With the
shortened postcode LIS 1 and the AKS data, the entries
specified in FIG. 3c were found in the now larger
search space in the address database 134. Since no
unambiguous assignment was made during the complete
reading process, rather the two variants Hunt Street
and Hubert Street were determined as the most likely
solutions, in a final step a selection coding was
effected , in which, as illustrated in FIG. 3d, the
complete address label and the two variants to be
selected are presented to the operator on the screen.
By actuating the key for the number 1, he then selects
the correct street (Hunt Street) very quickly.
CA 02369790 2005-02-14
20365-4549
l0a
In accordance with one aspect of this invention,
there is provided a method for reading dispatch addresses
with postcodes which identify the addresses up to part of a
larger street or a few small streets, in which the images of
the dispatch surfaces having the addresses are recorded,
stored and fed to an OCR unit (130) for the reading of
addresses (300) and generation of corresponding address
codes, and in which, in the case of a non-unambiguous read
result of the OCR unit (130), the relevant image with the
address is fed to at least one video coding station (200)
for the video coding of the postcode (310), the following
steps being carried out if the street read by the OCR unit
(130) does not correspond to the streets) or street region
derived from the video-coded postcode which the sender
specified incorrectly: communication of the relevant image
to one or two video coding stations for the inputting of a
respective AKS code (330, 350) which identifies the
distribution point, if this has not already been carried out
in previous steps, performance of additional reading
processes in the OCR unit (130) with that part of the
postcode which identifies the streets/street part and is
shortened by one position and with the AKS codes (390) and,
in the case of non-unambiguous read results: performance of
further reading processes in the OCR unit (130) with that
part of the postcode which identifies the streets/street
part and is additionally shortened by a further position and
with the AKS codes (400), the additional position-shortening
and the reading processes being performed until the read
result is unambiguous or only a few candidates remain, the
correct one of which is selected by means of selection
coding.
In accordance with another aspect of this
invention, there is provided an apparatus for reading
CA 02369790 2005-02-14
20365-4549
lOb
dispatch addresses with incorrect postcodes which identify
the addresses up to part of a larger street or a few small
streets, having an automatic address reading system which
has the following elements: an apparatus for obtaining
images of dispatches (120), an OCR unit (130) for the
automatic evaluation of the images of the dispatch surfaces,
said images having the address information, video coding
stations (200) for the video coding of the images of the
dispatch surfaces, said images containing the address
information, a controller (170) for controlling the sequence
in the OCR unit (130), the video coding stations, and also
the data exchange, which is designed in such a way that
after addresses have not been able to be read unambiguously
in a first reading attempt (310) in the OCR unit (130),
firstly the corresponding images of the dispatch surfaces
with these addresses are transmitted to the video coding
stations (200) for the video coding of the postcodes, the
following steps being carried out if the street read by the
OCR unit (130) does not correspond to the streets) or
street region derived from the video-coded postcode which
the sender specified incorrectly: communication of the
relevant image to at least one video coding station for the
inputting of at least one AKS code which identifies the
distribution point, performance of additional reading
processes in the OCR unit (130) with the postcode which is
shortened by one position and with the AKS codes and, in the
case of non-unambiguous read results: performance of further
reading processes in the OCR unit (130) with the postcode
which is additionally shortened by a further position and
with the AKS codes, the position-shortening and the reading
processes being performed until the read result is
unambiguous or only a few candidates remain, the correct one
of which is selected by means of selection coding.
