U.S. patent application number 15/567567 was filed with the patent office on 2018-09-27 for self-correction recognition method and device for valuable document recognition device.
This patent application is currently assigned to GRG Banking Equipment Co., Ltd.. The applicant listed for this patent is GRG Banking Equipment Co., Ltd.. Invention is credited to Mengtao Liu, Panfeng Luo, Rongqiu Wang, Dong Wei.
Application Number | 20180276491 15/567567 |
Document ID | / |
Family ID | 54576282 |
Filed Date | 2018-09-27 |
United States Patent
Application |
20180276491 |
Kind Code |
A9 |
Luo; Panfeng ; et
al. |
September 27, 2018 |
SELF-CORRECTION RECOGNITION METHOD AND DEVICE FOR VALUABLE DOCUMENT
RECOGNITION DEVICE
Abstract
A self-correction recognition method and device for a valuable
document recognition device is provided. The method includes:
extracting a feature of a valuable document; determining whether
the feature falls in a discriminant library of any preset category,
and obtaining a substitutive centroid according to a feature
centroid and the feature in a case that the feature falls in a
discriminant library of any preset category; and updating the
discriminant library of the category by replacing the feature
centroid of the category with the substitutive centroid. The
discriminant library is pre-constructed according to the feature
centroid of the category, and the feature centroid is calculated
from a feature of the category.
Inventors: |
Luo; Panfeng; (Guangzhou,
Guangdong, CN) ; Wang; Rongqiu; (Guangzhou,
Guangdong, CN) ; Liu; Mengtao; (Guangzhou, Guangdong,
CN) ; Wei; Dong; (Guangzhou, Guangdong, CN) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
GRG Banking Equipment Co., Ltd. |
Guangzhou, Guangdong |
|
CN |
|
|
Assignee: |
GRG Banking Equipment Co.,
Ltd.
Guangzhou, Guangdong
CN
|
Prior
Publication: |
|
Document Identifier |
Publication Date |
|
US 20180107889 A1 |
April 19, 2018 |
|
|
Family ID: |
54576282 |
Appl. No.: |
15/567567 |
Filed: |
March 30, 2016 |
PCT Filed: |
March 30, 2016 |
PCT NO: |
PCT/CN2016/077873 PCKC 00 |
371 Date: |
October 18, 2017 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06K 9/32 20130101; G06K
9/00463 20130101; G06K 9/186 20130101; G06K 9/3208 20130101; G06K
9/6269 20130101 |
International
Class: |
G06K 9/32 20060101
G06K009/32; G06K 9/18 20060101 G06K009/18; G06K 9/00 20060101
G06K009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Apr 23, 2015 |
CN |
201510198411.0 |
Claims
1. A self-correction recognition method for a valuable document
recognition device, comprising: extracting a feature M of a
valuable document; determining whether the feature M falls in a
discriminant library of any preset category i, and obtaining a
substitutive centroid O.sub.i' based on a feature centroid O.sub.i
and the feature M in a case that the feature M falls in the
discriminant library of the preset category i; and updating the
discriminant library of the category i by replacing the feature
centroid O.sub.i of the category i with the substitutive centroid
O.sub.i', wherein, the discriminant library is pre-constructed
based on the feature centroid O.sub.i of the category i; and the
feature centroid O.sub.i is calculated from a feature of the
category i.
2. The method according to claim 1, wherein before extracting the
feature M of the valuable document, the method further comprises:
extracting features of n categories of a valuable document, wherein
each of the n categories comprises one or more features, and n is
greater than 1; calculating the feature centroid O.sub.i of the
category i based on the features of the category i; obtaining a
relative discriminant plane group of the category i based on the
feature centroid O.sub.i, wherein the relative discriminant plane
group of the category i is composed of relative discriminant planes
of the category i relative to the other n-1 categories, and a
relative discriminant plane l.sub.i.eta. of the category i relative
to a category .eta. satisfies that the relative discriminant plane
l.sub.i.eta. is perpendicular to a connection line between the
feature centroid O.sub.i of the category i and a feature centroid
O.sub..eta. of the category .eta., valuable documents of the
category i and the category .eta. are respectively divided to the
two sides of the plane, all valuable documents of the category i
are located on a positive side of the relative discriminant plane
l.sub.i.eta., all valuable documents of the category .eta. are
located on a negative side of the relative discriminant plane
l.sub.i.eta., and a minimum value of distances between valuable
documents of the category i and the relative discriminant plane
l.sub.i.eta. is smaller than a minimum value of distances between
valuable documents of the category .eta. and the relative
discriminant plane l.sub.i.eta., and i is not equal to .eta.;
building a discriminant sphere of the category i with a radius
R.sub.i, wherein R.sub.i is determined by the feature centroid
O.sub.i and the relative discriminant plane group of the category
i; and determining an intersecting area between the relative
discriminant plane group of the category i and the discriminant
sphere of the category i as a discriminant library of the category
i.
3. The method according to claim 2, wherein the relative
discriminant plane is expressed as: w i .eta. T * X + d i .eta. = 0
, ##EQU00027## where X is a space vector of the valuable document
feature, w.sub.i.eta. is a normal vector of the relative
discriminant plane l.sub.i.eta., and d.sub.i.eta. is an intercept
of the relative discriminant plane l.sub.i.eta..
4. The method according to claim 3, wherein the radius R.sub.i is
expressed as: R i = min ( i i + .chi. i 2 ) , ##EQU00028## where
.epsilon..sub.i is a minimum value of distances between the feature
centroid O.sub.i of the category i and the relative discriminant
plane group of the category i, and .chi..sub.i is a maximum value
of distances between the feature centroid O.sub.i of the category i
and all the valuable documents of the category i.
5. The method according to claim 1, wherein the substitutive
centroid O.sub.i' is expressed as:
O.sub.i'=(1-.gamma.)O.sub.i+.gamma.M, where an update coefficient
.gamma. is less than 1 and greater than 0.
6. The method according to claim 2, wherein the calculating the
feature centroid O.sub.i of the category i based on the features of
the category i comprises: obtaining the features of the category i;
and calculating an arithmetic mean of the features of the category
i as the feature centroid O.sub.i.
7. The method according to claim 4, wherein, i = min .eta. .noteq.
i ( O i l i .eta. ) = min .eta. .noteq. i ( w i .eta. T * O i + d i
.eta. w i .eta. ) , .chi. i = max 1 .ltoreq. j .ltoreq. N i ( O i M
ij ) = max 1 .ltoreq. j .ltoreq. N i ( ( k = 1 n ( o ik - m ijk ) 2
) ) , ##EQU00029## where N.sub.i is the number of valuable
documents for the category i, O.sub.ik is a k.sup.th feature value
of the feature centroid O.sub.i of the category i, M.sub.ij is a
j.sup.th feature in the category i, and n is the number of features
of the valuable documents.
8. The method according to claim 7, wherein the determining whether
the feature M falls in a discriminant library of any preset
category i comprises: determining whether the feature M satisfies
both of the following two expressions: w i .eta. T * M + d i .eta.
> 0 ##EQU00030## MO i .fwdarw. = k = 1 n ( m k - o ik ) 2 < R
i ##EQU00031## where .parallel.{right arrow over
(MO.sub.i)}.parallel. is a distance between the feature M and the
feature centroid O.sub.i.
9. The method according to claim 1, wherein before updating the
discriminant library of the category i by replacing the feature
centroid O.sub.i of the category i with the substitutive centroid
O.sub.i', the method further comprises: determining whether the
substitutive centroid O.sub.i' satisfies a preset correction
condition, and performing the step of updating the discriminant
library of the category i by replacing the feature centroid O.sub.i
of the category i with the substitutive centroid O.sub.i' in a case
that the substitutive centroid O.sub.i' satisfies the preset
correction condition, wherein in a case that the substitutive
centroid O.sub.i' does not satisfy the preset correction condition,
the discriminant library is not updated.
10. The method according to claim 9, wherein the determining
whether the substitutive centroid O.sub.i' satisfies a preset
correction condition comprises: determining whether a distance
S.sub.i.eta. between the substitutive centroid O.sub.i' and the
relative discriminant plane l.sub.i.eta. satisfies: S i .eta. >
R i - .DELTA. i .eta. 2 ##EQU00032## where .DELTA..sub.i.eta. is a
distance between the relative discriminant plane l.sub.i.eta. and a
relative discriminant plane l.sub..eta.i, and l.sub.i.eta. is the
relative discriminant plane of the category .eta. relative to the
category i.
11. A self-correction recognition device for a valuable document
recognition device, comprising: a feature extracting unit, a first
determining unit, a substitutive centroid obtaining unit and an
updating unit, wherein the feature extraction unit is configured to
extract a feature M of a valuable document; the first determining
unit is configured to determine whether the feature M falls in a
discriminant library of any preset category i, and activate the
substitutive centroid obtaining unit in a case that the feature M
falls in the discriminant library of the preset category i; the
substitutive centroid obtaining unit is configured to obtain a
substitutive centroid O.sub.i' based on a feature centroid O.sub.i
and the feature M; the updating unit is configured to update the
discriminant library of the category i by replacing the feature
centroid O.sub.i of the category i with the substitutive centroid
O.sub.i'; the discriminant library is pre-constructed based on the
feature centroid O.sub.i of the category i; and the feature
centroid O.sub.i is calculated from a feature of the category
i.
12. The device according to claim 11, further comprising: a feature
extracting subunit, a calculating unit, a plane obtaining unit, a
dividing unit, a constructing unit and a determining unit, wherein
the feature extracting subunit is configured to extract features of
n categories of a valuable document, wherein each of the n
categories comprises one or more features, and n is greater than 1;
the calculating unit is configured to calculate the feature
centroid O.sub.i of the category i based on the features of the
category i; the plane obtaining unit is configured to obtain a
relative discriminant plane group of the category i based on the
feature centroid O.sub.i, wherein the relative discriminant plane
group of the category i is composed of relative discriminant planes
of the category i relative to the other n-1 categories, and a
relative discriminant plane l.sub.i.eta. of the category i relative
to a category .eta. satisfies that the relative discriminant plane
l.sub.i.eta. is perpendicular to a connection line between the
feature centroid O.sub.i of the category i and a feature centroid
O.sub..zeta. of the category .eta.; the dividing unit is configured
to divide valuable documents of the category i and the category
.eta. to the two sides of the plane, wherein all valuable documents
of the category i are located on a positive side of the relative
discriminant plane l.sub.i.eta., all valuable documents of the
category .eta. are located on a negative side of the relative
discriminant plane l.sub.i.eta., and a minimum value of distances
between valuable documents of the category i and the relative
discriminant plane l.sub.i.eta. is smaller than a minimum value of
distances between valuable documents of the category .eta. and the
relative discriminant plane l.sub.i.eta., and i is not equal to
.eta.; the constructing unit is configured to build a discriminant
sphere of the category i with a radius R.sub.i, wherein R.sub.i is
determined by the feature centroid O.sub.i and the relative
discriminant plane group of the category i; and the determining
unit is configured to determine an intersecting area between the
relative discriminant plane group of the category i and the
discriminant sphere of the category i as a discriminant library of
the category i.
13. The device according to claim 12, wherein the calculating unit
comprises: a feature obtaining unit and a calculating subunit, the
feature obtaining unit is configured to obtain the features of the
category i; and the calculating subunit is configured to calculate
an arithmetic mean of the features of the category i as the feature
centroid O.sub.i.
14. The device according to claim 11, further comprising: a second
determining unit, configured to determine whether the substitutive
centroid O.sub.i' satisfies a preset correction condition, and
activate the updating unit in a case that the substitutive centroid
O.sub.i' satisfies the preset correction condition.
Description
[0001] The application claims the priority to Chinese Patent
Application No. 201510198411.0, titled "SELF-CORRECTION RECOGNITION
METHOD FOR VALUABLE DOCUMENT RECOGNITION DEVICE", filed on Apr. 23,
2015 with the State Intellectual Property Office of the People's
Republic of China, which is incorporated herein by reference in its
entirety.
TECHNICAL FIELD
[0002] The present disclosure relates to the technology of valuable
document recognition, and in particular to a self-correction
recognition method and a self-correction recognition device for a
valuable document recognition device.
BACKGROUND
[0003] For a sorter, an ATM, a VTM, a vending machine, an automatic
ticket machine and other intelligent processing devices for
currency or notes, a core module is a valuable document recognition
module. The performance of recognizing a valuable document is a
core index for evaluating a valuable document recognition device.
The index is affected by the following two factors in the
large-scale application.
[0004] 1) A first effect is operating environment. Changes in
temperature, humidity and other environmental changes affect a
precision sensor signal, thus affecting recognition accuracy for
valuable documents.
[0005] 2) A second effect is difference between valuable documents.
There are significant differences between valuable documents of
various areas. For example, the bill quality in urban areas is
generally good, while the bill quality in rural areas is poor. Thus
a same set of recognition standards cannot be well adapted to both
conditions.
[0006] In view of the above problems, the following solutions are
applied in the industry currently.
[0007] 1) A sufficient number of samples of valuable documents are
provided, which normally is required to be one thousand or more
samples per category. Five or more devices are selected to collect
sample signals under conditions of variety of temperature and
humidity changes. The purpose is to collect as many categories and
sample signals as possible for training recognition software, thus
making the recognition adaptable to variety of different
environments.
[0008] 2) Different recognition software is used based on sample
differences in different regions. That is, different versions of
recognition software are customized according to actual needs.
[0009] However, the existing method of valuable document
recognition requires a lot of resources. Furthermore, when the use
environment changes, a response scheme have to be re-developed,
which cannot guarantee long-term stable effect, not only increases
the service cost of a provider, but also affects the market
benefits.
SUMMARY
[0010] A self-correction recognition method and a self-correction
recognition method device for a valuable document recognition
device is provided according to an embodiment of the present
disclosure, which are used for solving the problem that the
existing valuable document recognition method needs to consume a
large amount of resources and a long-term stable use effect cannot
be guaranteed due to the fact that a scheme needs to be renewed
when a use environment changes.
[0011] A self-correction recognition method for a valuable document
recognition device is provided according to an embodiment of the
present disclosure, which includes:
[0012] extracting a feature M of the valuable document;
[0013] determining whether the feature M falls in a discriminant
library of any preset category i, and obtaining a substitutive
centroid O.sub.i' based on a feature centroid O.sub.i and the
feature M in a case that the feature M falls in the discriminant
library of the preset category i; and
[0014] updating the discriminant library of the category i by
replacing the feature centroid O.sub.i of the category i with the
substitutive centroid O.sub.i';
[0015] wherein, the discriminant library is pre-constructed based
on the feature centroid O.sub.i of the category i; and
[0016] the feature centroid O.sub.i is calculated from a feature of
the category i.
[0017] Optionally, before extracting the feature M of the valuable
document, the method further includes:
[0018] extracting features of n categories of a valuable document,
where each of the n categories includes one or more features, and n
is greater than 1;
[0019] calculating the feature centroid O.sub.i of the category i
based on the features of the category i;
[0020] obtaining a relative discriminant plane group of the
category i based on the feature centroid O.sub.i, wherein the
relative discriminant plane group of the category i is composed of
relative discriminant planes of the category i relative to the
other n-1 categories, and a relative discriminant plane
l.sub.i.eta. of the category i relative to a category .eta.
satisfies that the relative discriminant plane l.sub.i.eta. is
perpendicular to a connection line between the feature centroid
O.sub.i of the category i and a feature centroid O.sub..eta. of the
category .eta., valuable documents of the category i and the
category .eta. are respectively divided to the two sides of the
plane, all valuable documents of the category i are located on a
positive side of the relative discriminant plane l.sub.i.eta., all
valuable documents of the category .eta. are located on a negative
side of the relative discriminant plane l.sub.i.eta., and a minimum
value of distances between valuable documents of the category i and
the relative discriminant plane l.sub.i.eta. is less than a minimum
value of distances between valuable documents of the category .eta.
and the relative discriminant plane l.sub.i.eta., and i is not
equal to .eta.;
[0021] building a discriminant sphere of the category i with a
radius R.sub.i, where R.sub.i is determined by the feature centroid
O.sub.i and the relative discriminant plane group of the category
i; and
[0022] determining an intersecting area between the relative
discriminant plane group of the category i and the discriminant
sphere of the category i as a discriminant library of the category
i.
[0023] Optionally, the relative discriminant plane is expressed
as:
w i .eta. T * X + d i .eta. = 0 ##EQU00001##
[0024] where X is a space vector of the valuable document feature,
w.sub.i.eta. is a normal vector of the relative discriminant plane
l.sub.i.eta., and d.sub.i.eta. is an intercept of the relative
discriminant plane l.sub.i.eta..
[0025] Optionally, the radius R.sub.i is expressed as:
R i = min ( i , i + .chi. i 2 ) , ##EQU00002##
[0026] where, .epsilon..sub.i is a minimum value of distances
between the feature centroid O.sub.i of the category i and the
relative discriminant plane group of the category i, and
.chi..sub.i is a maximum value of distances between the feature
centroid O.sub.i of the category i and all the valuable documents
of the category i.
[0027] Optionally, the substitutive centroid O.sub.i' is expressed
as:
O.sub.i'=(1-.gamma.)O.sub.i+.gamma.M,
[0028] where, an update coefficient .gamma. is less than 1 and
greater than 0.
[0029] Optionally, the calculating the feature centroid O.sub.i of
the category i based on the features of the category i specifically
includes:
[0030] obtaining the features of the category i; and
[0031] calculating an arithmetic mean of the features of the
category i as the feature centroid O.sub.i.
[0032] Optionally,
i = min .eta. .noteq. i ( O i l i .eta. ) = min .eta. .noteq. i ( w
i .eta. T * O i + d i .eta. w i .eta. ) , .chi. i = max 1 .ltoreq.
j .ltoreq. N i ( O i M ij ) = max 1 .ltoreq. j .ltoreq. N i ( ( k =
1 n ( o ik - m ijk ) 2 ) ) . ##EQU00003##
[0033] Where, N.sub.i is the number of valuable documents for the
category i, O.sub.ik is a k.sup.th feature value of the feature
centroid O.sub.i of the category i, M.sub.ij is a j.sup.th feature
in the category i, and n is the number of features of the valuable
documents.
[0034] Optionally, the determining whether the feature M falls in a
discriminant library of any preset category i specifically
includes:
[0035] determining whether the feature M satisfies both of the
following two expressions:
w i .eta. T * M + d i .eta. > 0 ##EQU00004## MO i .fwdarw. = k =
1 n ( m k - o ik ) 2 < R i ##EQU00004.2##
[0036] Where, .parallel.{right arrow over (MO.sub.i)}.parallel. is
a distance between the feature M and the feature centroid
O.sub.i.
[0037] Optionally, before updating the discriminant library of the
category i by replacing the feature centroid O.sub.i of the
category i with the substitutive centroid O.sub.i', the method
further includes:
[0038] determining whether the substitutive centroid O.sub.i'
satisfies a preset correction condition, and performing the step of
updating the discriminant library of the category i by replacing
the feature centroid O.sub.i of the category i with the
substitutive centroid O.sub.i' in a case that the substitutive
centroid O.sub.i' satisfies the preset correction condition, where
in a case that the substitutive centroid O.sub.i' does not satisfy
the preset correction condition, the discriminant library is not
updated.
[0039] Optionally, the determining whether the substitutive
centroid O.sub.i' satisfies a preset correction condition
specifically includes:
[0040] determining whether a distance S.sub.i.eta. between the
substitutive centroid O.sub.i' and the relative discriminant plane
l.sub.i.eta. satisfies:
S i .eta. > R i - .DELTA. i .eta. 2 . ##EQU00005##
[0041] Where, .DELTA..sub.i.eta. is a distance between the relative
discriminant plane l.sub.i.eta. and a relative discriminant plane
l.sub..eta.i, and l.sub.i.eta. is the relative discriminant plane
of the category .eta. relative to the category i.
[0042] A self-correction recognition device for a valuable document
recognition device is provided according to an embodiment of the
present disclosure, including: a feature extracting unit, a first
determining unit, a substitutive centroid obtaining unit and an
updating unit.
[0043] The feature extraction unit is configured to extract a
feature M of a valuable document.
[0044] The first determining unit is configured to determine
whether the feature M falls in a discriminant library of any preset
category i, and activate the substitutive centroid obtaining unit
in a case that the feature M falls in the discriminant library of
the preset category i.
[0045] The substitutive centroid obtaining unit is configured to
obtain a substitutive centroid O.sub.i' based on a feature centroid
O.sub.i and the feature M.
[0046] The updating unit is configured to update the discriminant
library of the category i by replacing the feature centroid O.sub.i
of the category i with the substitutive centroid O.sub.i'.
[0047] The discriminant library is pre-constructed based on the
feature centroid O.sub.i of the category i; and the feature
centroid O.sub.i is calculated from a feature of the category
i.
[0048] Preferably, the device further includes: a feature
extracting subunit, a calculating unit, a plane obtaining unit, a
dividing unit, a constructing unit and a determining unit.
[0049] The feature extracting subunit is configured to extract
features of n categories of a valuable document, where each of the
n categories includes one or more features, and n is greater than
1.
[0050] The calculating unit is configured to calculate the feature
centroid O.sub.i of the category i based on the features of the
category i.
[0051] The plane obtaining unit is configured to obtain a relative
discriminant plane group of the category i based on the feature
centroid O.sub.i. Where the relative discriminant plane group of
the category i is composed of relative discriminant planes of the
category i relative to the other n-1 categories, and a relative
discriminant plane l.sub.i.eta. of the category i relative to a
category .eta. satisfies that the relative discriminant plane
l.sub.i.eta. is perpendicular to a connection line between the
feature centroid O.sub.i of the category i and a feature centroid
O.sub..eta. of the category .eta..
[0052] The dividing unit is configured to divide valuable documents
of the category i and the category .eta. to the two sides of the
plane. And all valuable documents of the category i are located on
a positive side of the relative discriminant plane I.sub.i.eta.,
all valuable documents of the category .eta. are located on a
negative side of the relative discriminant plane l.sub.i.eta., and
a minimum value of distances between valuable documents of the
category i and the relative discriminant plane l.sub.i.eta. is
smaller than a minimum value of distances between valuable
documents of the category .eta. and the relative discriminant plane
l.sub.i.eta., and i is not equal to .eta..
[0053] The constructing unit is configured to build a discriminant
sphere of the category i with a radius R.sub.i. Where R.sub.i is
determined by the feature centroid O.sub.i and the relative
discriminant plane group of the category i.
[0054] The determining unit is configured to determine an
intersecting area between the relative discriminant plane group of
the category i and the discriminant sphere of the category i as a
discriminant library of the category i.
[0055] Preferably, the calculating unit includes: a feature
obtaining unit and a calculating subunit.
[0056] The feature obtaining unit is configured to obtain the
features of the category i.
[0057] The calculating subunit is configured to calculate an
arithmetic mean of the features of the category i as the feature
centroid O.sub.i.
[0058] Preferably, the device further includes: a second
determining unit, configured to determine whether the substitutive
centroid O.sub.i' satisfies a preset correction condition, and
activate the updating unit in a case that the substitutive centroid
O.sub.i' satisfies the preset correction condition.
[0059] As can be seen from the above technical solution,
embodiments of the present disclosure have the following
advantages.
[0060] In the embodiments of the present disclosure, by
establishing a discriminant library with self-correcting ability,
when any valuable document enters the recognition device, a
calculated substitutive centroid is used to realize
self-correction. Therefore, for different environment changes and
differences of valuable documents in different regions, better
recognition results can be achieved. When the use environment
changes, There is no need to re-develop a response scheme, thereby
ensuring long-term stability of using effects and saving costs and
resources.
BRIEF DESCRIPTION OF THE DRAWINGS
[0061] FIG. 1 is a flow chart for establishing a discriminant
library of a valuable document according to an embodiment of the
present disclosure;
[0062] FIG. 2 is a flow chart for realizing self-correction of a
discriminant library of a valuable document according to an
embodiment of the present disclosure;
[0063] FIG. 3 is a schematic diagram of a relative discriminant
plane group according to an embodiment of the present disclosure;
and
[0064] FIG. 4 is a schematic diagram of symmetry of relative
discriminant planes according to an embodiment of the present
disclosure.
DETAILED DESCRIPTION OF THE EMBODIMENTS
[0065] A self-correction recognition method and device for a
valuable document recognition device is provided according to an
embodiment of the present disclosure, which are used for solving
the problem that the existing valuable document recognition method
needs to consume a large amount of resources and a long-term stable
use effect cannot be guaranteed due to the fact that a scheme needs
to be renewed when a use environment changes.
[0066] The technical solution according to the embodiments of the
present disclosure will be described clearly and completely as
follows in conjunction with the accompany drawings in the
embodiments of the present disclosure, so that purposes,
characteristics and advantages of the present disclosure can be
more obvious and understandable. It is obvious that the described
embodiments are only a part of the embodiments according to the
present disclosure. All the other embodiments obtained by those
skilled in the art based on the embodiments in the present
disclosure without any creative work belong to the scope of the
present disclosure.
[0067] In the present scheme, a discriminant library of a valuable
document is firstly established, and the construction process is
shown in FIG. 1. Then, in the process of recognition, the
discriminant library of a valuable document is self-corrected based
on new valuable documents, to finally achieve desired effect. The
correcting process is shown in FIG. 2.
[0068] Table 1 is a summary table of the mathematical symbol
meaning according to the present disclosure.
TABLE-US-00001 number sign meaning remark 1 i, .eta. category index
of a valuable document, different related to a mathematical symbols
represent different categories valuable 2 N.sub.i a sample number
of an i.sup.th category of valuable document document 3 N.sub..eta.
a sample number of a .eta..sup.th category of valuable document 4 j
an index of a valuable document in an i.sup.th category of valuable
document 5 s an index of a valuable document code in a .eta..sup.th
category of valuable document 6 n the number of valuable document
features, i.e., the number of dimension of feature, and in the
embodiments n = 240 7 k a feature value index for a valuable
document 8 X variables of n-dimensional feature spaces 9 M a
valuable document feature in a n-dimensional feature space 10
M.sub.ij a j.sup.th valuable document feature in an i.sup.th
category 11 M.sub..eta.s a s.sup.th valuable document feature in a
.eta..sup.th category 12 m.sub.ijk a k.sup.th feature value of a
j.sup.th valuable document feature M.sub.ij in an i.sup.th category
13 O.sub.i a centroid of an i.sup.th category of valuable document
related to a 14 O.sub..eta. a centroid of a .eta..sup.th category
of valuable document discriminant 15 O.sub.i' a substitutive
centroid of an i.sup.th category of valuable sphere document 16
o.sub.ik a k.sup.th feature value of a centroid O.sub.i of an
i.sup.th category of valuable document 17 R.sub.i a discriminant
radius centered on a centroid O.sub.i of an i.sup.th category of
valuable document 18 R.sub..eta. a discriminant radius centered on
a centroid O.sub..eta. of a .eta..sup.th category of valuable
document 19 L.sub.i a relative discriminant plane group of an
i.sup.th category of related to a valuable document, which is
composed of l.sub.i.eta. relative 20 l.sub.i.eta. a relative
discriminant plane of a cetegory i relative to a discriminant
category .eta., wherein .eta. .noteq. i plane 21 w.sub.i.eta. a
normal vector of a relative discriminant plane l.sub.i.eta. 22
d.sub.i.eta. an intercept of a relative discriminant plane
l.sub.i.eta. 23 l.sub..eta.i a relative discriminant plane of a
category .eta. relative to a category i, wherein .eta. .noteq. i 24
w.sub..eta.i a normal vector of a relative discriminant plane
l.sub..eta.i 25 d.sub..eta.i an intercept of a relative
discriminant plane l.sub..eta.i 26 a.sub.i.eta. a minimum value of
distances between all samples in an i.sup.th related to a category
and a relative discriminant plane l.sub.i.eta. distance 27
b.sub.i.eta. a minimum value of distances between all samples in a
.eta..sup.th category and a relative discriminant plane
l.sub.i.eta. 28 a.sub..eta.i a minimum value of distances between
all samples in a .eta..sup.th category and a relative discriminant
plane l.sub..eta.i 29 b.sub..eta.i a minimum value of distances
between all samples in an i.sup.th category and a relative
discriminant plane l.sub..eta.i 30 .epsilon..sub.i a minimum value
of distances between an i.sup.th category centroid and all relative
discriminant planes of a category i 31 .chi..sub.i a maximum value
of distances between an i.sup.th category centroid and all
developing samples of a category i 32 S.sub.i.eta. a distance
between an i.sup.th category centroid and a relative discriminant
plane l.sub.i.eta. of a category i 33 S.sub..eta.i a distance
between a .eta..sup.th category centroid and a relative
discriminant plane l.sub..eta.i of a category .eta. 34
.DELTA..sub.i.eta. a distance between a relative discriminant plane
l.sub..eta.i and a relative discriminant plane l.sub.i.eta.,
l.sub..eta.i is parallel to l.sub.i.eta., and directions of normal
vector are opposite 35 min calculate a minimum value of two
elements calculation 36 min 1 .ltoreq. s .ltoreq. N .eta.
##EQU00006## for s being 1 to N.sub..eta., calculate a minimum
value and coefficient 37 min 1 .ltoreq. j .ltoreq. N i ##EQU00007##
for j being 1 to N.sub.i, calculate a minimum value 38 min .eta.
.noteq. i ##EQU00008## for all .eta. but .eta. .noteq. i, calculate
a minimum value 39 .gamma. an updating coefficient of a valuable
document centroid, generally 0.0001 < .gamma. < 0.01, in the
embodiment .gamma. = 0.001 40 p a ratio of b.sub.i.eta. to
a.sub.i.eta., generally p .gtoreq. 1 is required, in the embodiment
p = 2
[0069] First, it should be noted that in the embodiments of the
present disclosure, a feature of the valuable document is a set of
vector characterizing the valuable document, and developing samples
of the valuable document are a number of valuable documents for
constructing an initial discriminant library.
[0070] A process for establishing a discriminant library of the
valuable document is specifically described as follows.
[0071] In step S1, multi-dimensional valuable document features are
extracted. Feature extraction methods for valuable documents vary
slightly depending on different application scenarios. Generally,
extracted features of same categories of valuable documents are
similar, and extracted features of different categories of valuable
documents are significantly different. In the solution, the number
of features of the valuable document is not strictly required,
which is generally from dozens to hundreds. The combination of all
the features of the valuable documents constitutes a
super-high-dimensional feature space. The following two specific
embodiments are provided for the method of extracting the valuable
document, which however is not limited thereto.
[0072] A first specific embodiment is provided as step S1-1. Based
on a color image signal of a valuable document, an image is divided
into several equal parts. For example, the valuable document is
divided into 80 equal parts (10*8), and after the division, average
gray values of three color components of RGB are calculated as
feature values, totaling 80*3=240 values, that is, a feature with
240 elements is obtained. Different banknotes have different
features, and all valuable document features constitute a
240-dimensional feature space.
[0073] A second specific embodiment is provided as step S1-2.
Different feature information of valuable document, such as
ultraviolet optical feature information, infrared optical feature
information, fluorescence feature information, magnetic feature
information, white light image feature information, infrared image
feature information, ultraviolet image feature information,
magnetic image feature information, size feature information, and
thickness signal feature information, is extracted. One or more
features are extracted from signals of each of the above ten
different types of information. For example, the size feature
information may include a length feature and a width feature of a
valuable document, a magnetic coding feature of a security line for
a valuable document may be extracted from the magnetic feature
information, a fluorescence response feature for a valuable
document may be extracted from the fluorescence feature
information. Generally, hundreds of features may be extracted from
the ten types of information. It is acceptable to select more or
less features, or to select only a few of the ten types of
information, or to select a type of information other than the ten
types of information, which are all included in the present
disclosure.
[0074] The quality of the feature extraction will affect the
recognition effect and the speed of self-correction, but the
valuable document discriminant library can be self-corrected by
following the self-correction method.
[0075] In step S2, a feature centroid of each category of valuable
document is calculated. For any category i of valuable documents, a
multi-dimensional valuable document feature is extracted in S1. A
feature centroid is represented by O.sub.i, and the feature
centroid is shorted as a centroid. O.sub.i is equal to an
arithmetic mean of all features of the valuable document. If
N.sub.i represents a sample number of an i.sup.th category of
valuable document, and M.sub.ij represents a j.sup.th valuable
document feature in the i category, then
O i = 1 N i j = 0 N i M ij . ##EQU00009##
[0076] In particular, O.sub.i and M.sub.ij are both one-dimensional
vectors, and the number of elements is equal to a feature number of
the valuable document. Obviously for any i and j, a dimension
number of a feature vector M.sub.ij is the same as a dimension
number of O.sub.i, which is equal to the number of features
extracted in S1 and is denoted as n. Then,
O.sub.i=(o.sub.i1,o.sub.i2, . . . o.sub.in).sup.T
M.sub.ij=(m.sub.ij1,m.sub.ij2, . . . ,m.sub.ijn).sup.T.
[0077] In the solution, the number of categories of valuable
documents, the number of features of valuable documents, and the
number of developing samples of each valuable document category can
be defined depending on different processed objects. The valuable
document categories may include 16 categories, including: [0078] 1)
a front-viewed normal orientation of RMB 100 yuan; [0079] 2) a
front-viewed upside-down orientation of RMB 100 yuan; [0080] 3) a
back-viewed normal orientation of RMB 100 yuan; [0081] 4) a
back-viewed upside-down orientation of RMB 100 yuan; [0082] 5) a
front-viewed normal orientation of RMB 50 yuan; [0083] 6) a
front-viewed upside-down orientation of RMB 50 yuan; [0084] 7) a
back-viewed normal orientation of RMB 50 yuan; [0085] 8) a
back-viewed upside-down orientation of RMB 50 yuan; [0086] 9) a
front-viewed normal orientation of RMB 20 yuan; [0087] 10) a
front-viewed upside-down orientation of RMB 20 yuan; [0088] 11) a
back-viewed normal orientation of RMB 20 yuan; [0089] 12) a
back-viewed upside-down orientation of RMB 20 yuan; [0090] 13) a
front-viewed normal orientation of RMB 10 yuan; [0091] 14) a
front-viewed upside-down orientation of RMB 10 yuan; [0092] 15) a
back-viewed normal orientation of RMB 10 yuan; and [0093] 16) a
back-viewed upside-down orientation of RMB 10 yuan.
[0094] A first category of valuable document is a front-viewed
normal orientation of RMB 100 yuan, and the number of features may
be 240 features described in step S1-1 based on specific embodiment
1, that is n=240. The number of developing samples for each
category may be different, and the number of developing samples of
the first category of valuable document, which is a front-viewed
normal orientation of RMB 100 yuan, may be 50. A centroid of the
first category of a front-viewed normal orientation of RMB 100 yuan
is:
O 1 = 1 50 j = 0 50 M 1 j . ##EQU00010##
[0095] Through the above method, feature centroid of all categories
may be obtained.
[0096] In step S3, a relative discriminant plane group is
calculated. Relative discriminant plane groups of all categories of
valuable documents are calculated, the number of categories may be
16 as examples of step S2, and it may also be other number of
categories. For any category i of valuable document, a relative
discriminant plane group L.sub.i is composed of planes defined as
follows.
[0097] In step S3-1, for any category .eta..noteq.i, a relative
discriminant plane of a category i relative to a category .eta. is
defined as l.sub.i.eta.. L.sub.i is composed of relative
discriminant planes of the category i relative to all other
categories, that is,
L i = .eta. .noteq. i l i .eta. . ##EQU00011##
[0098] For the 16 categories of valuable documents in the example
of step S2, each category has 15 relative discriminant planes,
which together form a relative discriminant plane group of the
category, so there are 16 relative discriminant plane groups.
[0099] In step S3-2, for any .eta..noteq.i, based on the centroid
defined in step S2, the relative discriminant plane l.sub.i.eta.
defined in step S3-1 needs to satisfy that l.sub.i.eta. is
perpendicular to a connection line between a centroid O.sub.i of
the category i and a centroid O.sub..eta. of a category .eta., and
thus normal vector of l.sub.i.eta. is parallel to the connection
line between O.sub.i and O.sub..eta.. The method also defines the
normal vector of l.sub.i.eta. is from O.sub..eta. to O.sub.i, and
the normal vector of l.sub.i.eta. is denoted as w.sub.i.eta.,
then,
w.sub.i.eta.=O.sub.i-O.sub..eta.=(o.sub.i1-o.sub..eta.1,o.sub.i2-o.sub..-
eta.2, . . . ,o.sub.in-o.sub..eta.n).sup.T
[0100] In step S3-3, the relative discriminant plane l.sub.i.eta.
defined in step S3-1 also needs to satisfy that the category i and
the category .eta. are completely divided by l.sub.i.eta. to the
two sides of the plane. According to step S3-2, the normal vector
of l.sub.i.eta. is defined from O.sub..eta. to O.sub.i, which
actually requires that all sample features of the category i is on
the positive side of l.sub.i.eta., and all sample features of the
category .eta. is on the negative side of l.sub.i.eta., which is
equivalent to that values obtained by respectively substituting all
the sample features of the category i into an equation of the
relative discriminant plane l.sub.i.eta. are greater than 0, and
values obtained by respectively substituting all the sample
features of the category .eta. into an equation of the relative
discriminant plane l.sub.i.eta. are less than 0.
[0101] If there is no relative discriminant plane satisfying the
condition S3-3, it is indicated that the category i and the
category .eta. are not completely separated to each other in the
feature space, which is rare, and it is necessary to adjust the
feature extraction method to be adapted to the model.
[0102] In step S3-4, based on the relative discriminant plane
l.sub.i.eta. defined by the above S3-1, S3-2, S3-3, a minimum value
of distances between all samples in the category i and the plane
l.sub.i.eta. is denoted as a.sub.i.eta., and a minimum value of
distances between all samples in the category .eta. and the plane
l.sub.i.eta. is denoted as b.sub.i.eta.. In the solution, and
b.sub.i.eta. need to meet the following condition:
b.sub.i.eta.=p*a.sub.i.eta..
[0103] Where, p is a constant and not less than 1, generally p is
selected in the range of 1 to 100. In fact, when p=1, l.sub.i.eta.
is located between the category i and the category .eta., the
restriction is relatively relaxed. When p.gtoreq.10, the
restrictions is relatively strict. When features of different
categories are highly distinct, a larger p may be selected.
Generally p=2, and for .eta..noteq.i, it is satisfied that:
b.sub.i.eta.=2*a.sub.i.eta..
[0104] It is to be noted that at least the condition
b.sub.i.eta..gtoreq.a.sub.i.eta. needs to be satisfied in the
present embodiment. As shown in FIG. 5, the category i and the
category .eta. can be separated by either l.sub.i.eta. or
l.sub.i.eta.. But only when b.sub.i.eta..gtoreq.a.sub.i.eta.,
I.sub.i.eta. is more close to the category i, l.sub.i.eta. is more
close to the category .eta., so that the two discriminant planes
will not be overlapped, thereby ensuring uniqueness of
discrimination. On this basis, the relationship is described more
scientifically with b.sub.i.eta.=p*a.sub.i.eta., while having
better effect in fact if b.sub.i.eta.=2*a.sub.i.eta.. If
b.sub.i.eta.=a.sub.i.eta., l.sub.i.eta. and l.sub.i.eta. are
overlapped, i.e., l.sub.i.eta. and l.sub.i.eta. are same planes,
the degree of discrimination is poor. When p is too large, the
restriction is harsh, and the training is difficult to
converge.
[0105] In step S3-5, based on the above steps S3-1, S3-2, S3-3 and
S3-4, the relative discriminant plane l.sub.i.eta. may be obtained.
The number of features is n, in n-dimensional space X=(x.sub.1,
x.sub.2, . . . , x.sub.n).sup.T, an expression of the relative
discriminant plane l.sub.i.eta. may be defined as:
R i = min ( i , i + .chi. i 2 ) , ##EQU00012##
[0106] Based on step S3-2, a normal line w.sub.i.eta. of the
relative discriminant plane l.sub.i.eta. is obtained as
follows:
w.sub.i.eta.=O.sub.i-O.sub..eta.=(o.sub.i1-o.sub..eta.1,o.sub.i2-o.sub..-
eta.2, . . . ,o.sub.in-o.sub..eta.n).sup.T
[0107] An intercept d.sub.i.eta. of the relative discriminant plane
l.sub.i.eta. is solved. According to step S3-4,
b.sub.i.eta.=p*a.sub.i.eta. stands. According to step S3-4, for
a.sub.i.eta., b.sub.i.eta., as well as the relative discriminant
plane l.sub.i.eta. defined above, a relationship is provided as
follows:
min 1 .ltoreq. s .ltoreq. N .eta. ( w i .eta. T * M .eta. s + d i
.eta. w i .eta. ) = p * min 1 .ltoreq. j .ltoreq. N i ( w i .eta. T
* M ij + d i .eta. w i .eta. ) . ##EQU00013##
[0108] Where,
min 1 .ltoreq. s .ltoreq. N .eta. ##EQU00014##
represents calculating a minimum value of a function of s valuing
from 1 to N.sub..eta., and
min 1 .ltoreq. j .ltoreq. N i ##EQU00015##
represents calculating a minimum value of a function of j valuing
from 1 to N.sub.i. According to step S3-3, the values obtained by
substituting all the sample features of the category i into an
equation of the relative discriminant plane l.sub.i.eta. are
greater than 0, and the values obtained by substituting all the
sample features of the category .eta. into the equation of the
relative discriminant plane l.sub.i.eta. are less than 0, which is
expressed as:
w i .eta. T * M .eta. s + d i .eta. < 0 ##EQU00016## w i .eta. T
* M ij + d i .eta. > 0. ##EQU00016.2##
[0109] Then,
min 1 .ltoreq. s .ltoreq. N .eta. ( - w i .eta. T * M .eta. s - d i
.eta. ) = p * min 1 .ltoreq. j .ltoreq. N i ( w i .eta. T * M ij +
d i .eta. ) . Thereby , d i .eta. = 1 p + 1 * ( min 1 .ltoreq. s
.ltoreq. N .eta. ( - w i .eta. T * M .eta. s ) - p * min 1 .ltoreq.
j .ltoreq. N i ( w i .eta. T * M ij ) ) . ##EQU00017##
[0110] In summary, a normal and an intercept of the relative
discriminant plane l.sub.i.eta. are obtained, that is, an
expression of the relative discriminant plane l.sub.i.eta. is
obtained. In a same way, the relative discriminant plane groups of
all categories can be obtained.
[0111] In step S3-6, although based on the above steps, the
relative discriminant plane group of all categories can be
obtained, the symmetry derivation relation of the relative
discriminant plane is also provided.
[0112] A normal of the relative discriminant plane l.sub..eta.i it
of the category .eta. relative to the category i is defined as
follows:
w.sub..eta.i=O.sub..eta.-O.sub.i=-w.sub.i.eta..
[0113] Therefore, w.sub..eta.i is parallel to w.sub.i.eta., and
thus the relative discriminant plane l.sub..eta.i is parallel to
the relative discriminant plane l.sub.i.eta.. As shown in FIG. 4, a
sample having the minimum value of distances between all samples of
the category i and the relative discriminant plane l.sub.i.eta. is
the same as a sample having the minimum value of distances between
all samples of the category i and the relative discriminant plane
l.sub..eta.i. Similarly, a sample having the minimum value of
distances between all samples of the category .eta. and the
relative discriminant plane l.sub.i.eta. is the same as a sample
having the minimum value of distances between all samples of the
category .eta. and the relative discriminant plane l.sub..eta.i. So
there is:
a.sub.i.eta.+b.sub.i.eta.=a.sub..eta.i+b.sub..eta.i.
[0114] Due to p>1, it is obtained that
b.sub.i.eta..gtoreq.a.sub.i.eta. and
b.sub..eta.i.gtoreq.a.sub..eta.i, then l.sub.i.eta. is closer to
O.sub.i than l.sub..eta.i, and l.sub..eta.i is closer to
O.sub..eta. than l.sub.i.eta.. A distance between l.sub..eta.i and
l.sub.i.eta. is denoted as .DELTA..sub.i.eta., which satisfies:
.DELTA..sub.i.eta.=b.sub.i.eta.+b.sub..eta.i-a.sub.i.eta.-a.sub..eta.i=(-
p-1)(a.sub.i.eta.+a.sub..eta.i).
[0115] It should be noted that, when a spatial dimension is n=2,
the relative discriminant plane l.sub.i.eta. is degenerated into a
straight line of a plane space. As shown in FIG. 4, the spatial
dimension is n=2 (in general, n>10), the relative discriminant
planes l.sub.i.eta. and l.sub..eta.i are both degenerated into
straight lines of the plane space. When n=3, the relative
discriminant plane l.sub.i.eta. is degenerated into a plane of a
normal three-dimensional space. When n=4, the relative discriminant
plane l.sub.i.eta. is a three-dimensional space. In step S1-1, when
n=240, l.sub.i.eta. is a 239-dimensional hyper-plane.
[0116] In step S4, a valuable document discriminant library is
established. The valuable document discriminant library is composed
of discriminant domains of all the valuable document categories. A
discriminant domain of a category is described as follows. For
example, for the category i, whether a feature M of a n-dimension
space belongs to the category i is determined by the following
steps.
[0117] In step S4-1, for the category i, the discriminant radius
R.sub.i of the category is calculated by the method as follows.
First, based on step S2, a minimum value e of distances between the
centroid O.sub.i of category i and all relative discriminant planes
of the category i is calculated, and a maximum value .chi..sub.i of
distances between O.sub.i and all developing samples of a category
i is calculated, that is:
i = min .eta. .noteq. i ( O i l i .eta. ) = min .eta. .noteq. i ( w
i .eta. T * O i + d i .eta. w i .eta. ) , .chi. i = max 1 .ltoreq.
j .ltoreq. N i ( O i M ij ) = max 1 .ltoreq. j .ltoreq. N i ( ( k =
1 n ( o ik - m ijk ) 2 ) ) . ##EQU00018##
[0118] Where, l.sub.i.eta. is the relative discriminant plane of
the category i relative to the category .eta. defined in step S3,
N.sub.i is the number of samples of valuable documents of the
category i, M.sub.ij represents a j.sup.th valuable document
feature in the category i, and n is the number of features. Then
the calculation method for R.sub.i is defined as follows:
R i = min ( i . i + .chi. i 2 ) . ##EQU00019##
[0119] The expression of the radius R.sub.i provided in the
embodiment takes into account both a relationship between a
centroid and a relative discriminant plane, and a relationship
between a centroid and a sample. This can ensure that a category is
distinguishable from other categories and control the radius of the
category (that is, separability between the category and abnormal
data is ensured as far as possible).
[0120] In step S4-2, a discriminant sphere is constructed for each
category i of valuable document with the feature centroid O.sub.i
defined in step S2 as a center and a discriminant radius R.sub.i
defined in step S4-1 as a radius. If the feature M is in the
discriminant sphere, it means that M satisfies the definition of
S4-2, and M may belong to the category i. If the feature M is
outside the discriminant sphere, M is not belong to the category i.
That is, the distance between M and O.sub.i is required to be less
than R.sub.i, which is expressed as follows:
MO i .fwdarw. = k = 1 n ( m k - o ik ) 2 < R i .
##EQU00020##
[0121] In step S4-3, it is also required that the result obtained
by substituting the feature M into any relative discriminant planes
of the relative discriminant plane group L.sub.i in the category i
defined by step S3, must be positive, otherwise M does not belong
to the category i, that is, for .eta..noteq.i, it is satisfied
that:
w i .eta. T * M + d i .eta. > 0. ##EQU00021##
[0122] A region where a discriminant sphere of a category defined
in step S4-2 intersects a relative discriminant plane group of the
category defined in step S3 is a discriminant domain of the
category. If a feature of a valuable document is in the
discriminant domain of the category, the valuable document belongs
to the category, otherwise, the valuable document does not belong
to the category.
[0123] A specific self-correction process of a valuable document
discriminant library is further provided, which is described as
follows.
[0124] In step S1, in an actual operating process of a valuable
document recognition device, for a newly added valuable document
sample, a valuable document signal is collected, and a
multi-dimensional valuable document feature is extracted, referring
to the above step S1.
[0125] In step S5, the category of the feature is determined. For
the newly added valuable document sample, a valuable document
feature M is extracted in step S1, and the category to which the
valuable document feature belongs is determined based on the
valuable document discriminant library defined in the above step
S4. That is, according to S4, for any input feature M, if M meets
both of the following two formulas:
w i .eta. T * M + d i .eta. > 0 ##EQU00022## MO i .fwdarw. =
.tau. = 1 n ( m .tau. - o i .tau. ) 2 < R i , ##EQU00022.2##
it is determined that M belongs to the category i, otherwise, M
does not belong to the category i.
[0126] In step S6, a substitutive centroid is calculated. For any
newly inputted valuable document feature M, if it is determined
that M belongs to the category i of valuable document in step S5, a
substitutive centroid O.sub.i' of the category i of valuable
document is calculated. A calculation formula of the substitutive
centroid is provided as follows:
O.sub.i'=(1-.gamma.)O.sub.i+.gamma.M=((1-.gamma.)o.sub.1+.gamma.m.sub.1,-
(1-.gamma.)o.sub.2+.gamma.m.sub.2, . . .
,(1-.gamma.)o.sub.n+.gamma.m.sub.n).sup.T.
[0127] Where, .gamma. is a update coefficient, and
0<.gamma.<1. If the update coefficient is larger, the update
rate is faster, and if the update coefficient is smaller, the
stability is better. Generally, 0.0001<.gamma.<0.01.
Commonly, in the present embodiment, .gamma. may be selected as
.gamma.=0.001.
[0128] In step S7, legitimacy of correction is determined. That is,
it is determined that whether a preset condition is met. A preset
condition in the method may be the following correction rules.
Based on the substitutive centroid O.sub.i' of the category i
calculated in step S6, for .eta..noteq.i, a distance S.sub.i.eta.
between O.sub.i' and the relative discriminant plane l.sub.i.eta.
of the category i relative to the category .eta. is calculated, to
determine whether it satisfies:
S i .eta. > R i - .DELTA. i .eta. 2 ##EQU00023##
[0129] where, R.sub.i is a discriminant radius of the category i of
valuable document defined in step S4-1, and .DELTA..sub.i.eta. is a
distance between the relative discriminant plane l.sub.i.eta. and
the relative discriminant plane l.sub..eta.i defined in step
S3-6.
[0130] For all categories .eta..noteq.i other than the category i,
if the substitutive centroid O.sub.i' defined in step S6 can meet
the condition:
S i .eta. > R i - .DELTA. i .eta. 2 , O i ' ##EQU00024##
meets the condition of legitimacy of correction, otherwise,
O.sub.i' does not meet the condition of legitimacy of correction.
It is to be understood that the condition of the legitimacy of
correction ensures that a valuable document can only belong to one
category, and will not belong to two different categories, that is,
any two different categories do not overlap. Specific conditions
may be preset by coordinating distances between each centroid and
relative discriminant planes.
[0131] Further, the present method can ensure stability during the
model correction process while ensuring the legality. That is,
under the principle of legality of correction, any valuable
document only belongs to at most one category, and cannot belong to
two different categories during the correction process.
[0132] The course of proof is described as follows.
[0133] In order to prove that for any .eta..noteq.i, a valuable
document M cannot belong to both the category i and the category
.eta., it only needs to prove that the valuable document M cannot
be in both the discriminant sphere of the category i and the
discriminant sphere of the category .eta.. It is equivalent to
prove that the two discriminant spheres do not intersect each
other, that is, a distance between the centers of the two
discriminant spheres is not less than a sum of radiuses of the two
discriminant spheres, which is expressed as:
.parallel.{right arrow over
(O.sub.iO.sub..eta.)}.parallel..gtoreq.R.sub.i+R.sub..eta..
[0134] The above equation is proved as follows.
[0135] Based on step S3-3, O.sub.i is on a positive side of
l.sub.i.eta., O.sub..eta. is in a negative side of l.sub..eta.i, so
a distance between O.sub.i and O.sub..eta. is not less than a sum
of a distance between O.sub.i and l.sub.i.eta., and a distance
between O.sub..eta. and l.sub.i.eta.. As shown in FIG. 4, the
distance between O.sub..eta. and l.sub.i.eta. is equal to a
distance between O.sub..eta. and l.sub..eta.i plus a distance
between l.sub.i.eta. and l.sub..eta.i, so:
.parallel.{right arrow over
(O.sub.iO.sub..eta.)}.parallel..gtoreq.S.sub.i.eta.+S.sub..eta.i+.DELTA..-
sub.i.eta..
[0136] Based on S7, it is provided:
S i .eta. > R i - .DELTA. i .eta. 2 ##EQU00025## S .eta. i >
R .eta. - .DELTA. i .eta. 2 , ##EQU00025.2##
[0137] Then,
O i O .eta. .fwdarw. .gtoreq. S i .eta. + S .eta. i + .DELTA. i
.eta. > ( R i - .DELTA. i .eta. 2 ) + ( R .eta. - .DELTA. i
.eta. 2 ) + .DELTA. i .eta. = R i + R .eta. . ##EQU00026##
[0138] Therefore, the two discriminant spheres of the category i
and the category .eta. do not intersect each other, that is, a
valuable document can only belong to one category.
[0139] In step S8, the discriminant library is self-corrected. When
a substitutive centroid calculated based on step S6 satisfies the
condition of legitimacy of correction defined by S7, the
discriminant library is self-corrected with the correction method
of directly replacing O.sub.i with O.sub.i', that is
O.sub.i=O.sub.i'.
[0140] With steps of S1, S5, S6, S7, and S8, the self-correction of
valuable document discriminant library is realized on the basis of
the valuable document discriminant library established in step S4,
and it is proved based on step 7, that the self-correction process
according to the present embodiment is stable.
[0141] Based on the above steps, self-correction of the valuable
document discriminant library can be realized. According to the
present disclosure, through the self-correcting of discriminant
library, a good recognition effect can be achieved for different
environmental changes and the difference between valuable documents
in different regions. At the same time, the relative discriminant
plane group is introduced in the present disclosure for supervising
the self-correction process to avoid misrecognition of the valuable
documents due to over-correction. Therefore, the whole
self-correction process is supervised and stable.
[0142] The method according to present disclosure may not have a
self-correcting termination condition for the above self-correction
process of the discriminant library. As an extension and refinement
of the application, it is possible to define a self-correcting
condition as required in practices, that is, when a certain
recognition effect is achieved, the self-correction may be
stopped.
[0143] Based on the above self-correction recognition method for a
valuable document recognition device, a self-correction recognition
device for a valuable document recognition device is provided
according to an embodiment of the present disclosure.
[0144] The self-correction recognition device for a valuable
document recognition device provided according to the embodiment of
the present disclosure includes: a feature extracting unit, a first
determining unit, a substitutive centroid obtaining unit and an
updating unit.
[0145] The feature extraction unit is configured to extract a
feature M of a valuable document.
[0146] The first determining unit is configured to determine
whether the feature M falls in a discriminant library of any preset
category i, and activate the substitutive centroid obtaining unit
in a case that the feature M falls in the discriminant library of
the preset category i.
[0147] The substitutive centroid obtaining unit is configured to
obtain a substitutive centroid O.sub.i' based on a feature centroid
O.sub.i and the feature M.
[0148] The updating unit is configured to update the discriminant
library of the category i by replacing the feature centroid O.sub.i
of the category i with the substitutive centroid O.sub.i'.
[0149] The discriminant library is pre-constructed based on the
feature centroid O.sub.i of the category i; and the feature
centroid O.sub.i is calculated from a feature of the category
i.
[0150] In the embodiment of the present disclosure, by establishing
a discriminant library with self-correcting ability, when any
valuable document enters the recognition device, a calculated
substitutive centroid is used to realize self-correction, and thus
a good recognition results can be achieved for different
environment changes and differences between valuable documents in
different regions. When the use environment changes, there is no
need to re-develop a response scheme, thereby ensuring long-term
stability and saving costs and resources.
[0151] In another preferable embodiment, the device further
includes: a feature extracting subunit, a calculating unit, a plane
obtaining unit, a dividing unit, a constructing unit and a
determining unit.
[0152] The feature extracting subunit is configured to extract
features of n categories of a valuable document, where each of the
n categories includes one or more features, and n is greater than
1.
[0153] The calculating unit is configured to calculate the feature
centroid O.sub.i of the category i based on the features of the
category i.
[0154] The plane obtaining unit is configured to obtain a relative
discriminant plane group of the category i based on the feature
centroid O.sub.i. The relative discriminant plane group of the
category i is composed of relative discriminant planes of the
category i relative to the other n-1 categories, and a relative
discriminant plane l.sub.i.eta. of the category i relative to a
category .eta. satisfies that the relative discriminant plane
l.sub.i.eta. is perpendicular to a connection line between the
feature centroid O.sub.i of the category i and a feature centroid
O.sub..eta. of the category .eta..
[0155] The dividing unit is configured to divide valuable documents
of the category i and the category .eta. respectively to the two
sides of the plane. All valuable documents of the category i are
located on a positive side of the relative discriminant plane
l.sub.i.eta., all valuable documents of the category .eta. are
located on a negative side of the relative discriminant plane
l.sub.i.eta., and a minimum value of distances between valuable
documents of the category i and the relative discriminant plane
l.sub.i.eta. is smaller than a minimum value of distances between
valuable documents of the category .eta. and the relative
discriminant plane l.sub.i.eta., and i is not equal to .eta..
[0156] The constructing unit is configured to build a discriminant
sphere of the category i with a radius R.sub.i, where R.sub.i is
determined by the feature centroid O.sub.i and the relative
discriminant plane group of the category i.
[0157] The determining unit is configured to determine an
intersecting area between the relative discriminant plane group of
the category i and the discriminant sphere of the category i as a
discriminant library of the category i.
[0158] In one of possible implementations, the calculating unit may
include: a feature obtaining unit and a calculating subunit.
[0159] The feature obtaining unit is configured to obtain the
features of the category i.
[0160] The calculating subunit is configured to calculate an
arithmetic mean of the features of the category i as the feature
centroid O.sub.i.
[0161] In another embodiment, the device further includes: a second
determining unit, configured to determine whether the substitutive
centroid O.sub.i' satisfies a preset correction condition, and
activate the updating unit in a case that the substitutive centroid
O.sub.i' satisfies the preset correction condition.
[0162] It is to be known clearly by those skilled in the art that,
for convenient and clear description, for specific operation
processes of the above system, device and unit, reference may be
made to the corresponding process in the above method embodiment,
which is not described here.
[0163] It should be noted that, the embodiments of the present
disclosure are described herein in a progressive manner, with an
emphasis placed on explaining the difference between each
embodiment and the other embodiments; hence, for the same or
similar parts among the embodiments, they can be referred to from
one another. For the modulator disclosed in the embodiments, the
corresponding descriptions are relatively simple because the
modulator corresponds to the methods disclosed in the embodiments.
The relevant portions may be referred to the description for the
method parts. The above device embodiment is only illustrative. The
unit described as a separate component may be or may not be
separate physically, and the component which is displayed as a unit
may be or may not be a physical unit, that is, may be located at a
position, or may be distributed over multiple network units. Some
or all of the units may be selected as required to implement the
solution of the embodiment. Those skilled in the art will
understand and practice without any creative work.
[0164] As described above, the above embodiments are only intended
to describe the technical solutions of the disclosure, but not to
limit the scope of the disclosure. Although the disclosure is
described in detail with reference to the above embodiments, it
should be understood by those ordinal skilled in the art that
modifications can be made to the technical solutions recorded in
the above embodiments or equivalent replacements can be made to
some or all of the technical features thereof, which modifications
and equivalent replacements will not make the corresponding
technical solutions deviate from the scope of the technical
solutions of the embodiments of the disclosure.
* * * * *