EP2355056B1

EP2355056B1 - Paper sheet recognition apparatus and paper sheet recognition method

Info

Publication number: EP2355056B1
Application number: EP11152302.3A
Authority: EP
Inventors: Kunihiro Ryou
Original assignee: Glory Ltd
Current assignee: Glory Ltd
Priority date: 2010-01-29
Filing date: 2011-01-27
Publication date: 2018-07-25
Anticipated expiration: 2031-01-27
Also published as: EP2355056A1; JP2011159073A

Description

TECHNICAL FIELD

The present invention relates to a paper sheet recognition apparatus and a paper sheet recognition method for determining fitness of paper sheets such as banknotes. More particularly, the present invention relates to a paper sheet recognition apparatus and a paper sheet recognition method that can determine the fitness of paper sheets with high accuracy which have a wide variety of designs and soiling.

BACKGROUND ART

Paper sheet recognition devices that employ image sensors for determining fitness of paper sheets are known in the art. In such paper sheet recognition devices, an image of a paper sheet that is the target of fitness determination is captured with the image sensor, and the image is analyzed to determine the fitness of the paper sheet.
The term fitness determination refers to assessment of soiling, such as stains or wear and tear, in a paper sheet. The paper sheets devoid of soiling shall hereinafter be referred to as "fit notes" and paper sheets that bear soiling shall be referred to as "unfit notes".
Some of these paper sheet recognition devices employ the following method to determine the fitness of the banknote (see Japanese Patent Application Laid-open No. H4-199489 (Patent Document 1)). That is, a density of a predetermined area of a fit note is registered as a decision criterion density, a density of the corresponding area of a banknote that is the target of fitness determination is compared with the registered decision criterion density, and whether the banknote is fit is decided based on the result of comparison.
Other paper sheet recognition devices employ the following method to determine the fitness of the banknote (see Japanese Patent No. 2587433 (Patent Document 2)). That is, several predetermined areas including different pixel number (i.e., areas of different sizes) are registered beforehand as decision criterion areas, and fitness determination is performed by specifying one of the registered decision criterion areas.
The fitness determination is performed in Patent Document 2 by comparing an average pixel data value in the specified decision criterion area with an average pixel data value in a corresponding area in the banknote that is the target of fitness determination.
However, in the method disclosed in Patent Document 1, because an area that is devoid of print, such as the portion bearing the watermark, is taken as the predetermined area, a banknote that has ink wear (hereinafter, "ink damage") on the predetermined area is not determined as an unfit note.
Moreover, because the density of the area devoid of print is taken as the decision criterion density, this method is not suited for detecting localized soiling in paper sheets with a wide variety of designs, such as banknotes.
In the method disclosed in Patent Document 2, the pixel number varies depending on the size of the specified decision criterion area, and an accuracy of recognition drops, when the decision criterion area is large.
That is, because individual pixel data is more averaged out in case of larger decision criterion areas, this method, for example, is not suited for fitness determination of banknotes bearing designs that have differing contrasts from region to region or banknotes with local soiling (hereinafter, "localized soiling").
The accuracy of recognition is higher for smaller decision criterion areas; however, if the decision criterion area is made of a single pixel, all the pixels on the entire surface of the image need to be analyzed, which increases the processing load.
Due to the above-mentioned reasons, how to realize a paper sheet recognition apparatus or a paper sheet recognition method that can perform a highly accurate fitness determination for paper sheets having a wide variety of designs and soiling has been a big issue.
GB 2 164 442 A discloses an apparatus and method for sensing documents such as banknotes which determines fitness of a document based on a threshold value of elements of the document displaying certain brightness characteristics.
EP 0 680 909 A1 discloses a method and apparatus for identifying printed sheets which uses a fixed area of a printed sheet for identifying the sheets.

DISCLOSURE OF INVENTION

The present invention has been made in order to overcome the drawbacks of the above-mentioned conventional technology.
An object of the present invention is to provide a paper sheet recognition apparatus and a paper sheet recognition method that can perform a highly accurate fitness determination for paper sheets having a wide variety of designs and degree of soiling.
According to an aspect of the present invention there is provided a paper sheet recognition apparatus for determining fitness of a paper sheet based on a captured image of the paper sheet, the paper sheet recognition apparatus with the features claimed in claim 1.
According to an another aspect of the present invention there is a paper sheet recognition apparatus, wherein the valid area determining unit determines an ink damage valid area usable as the valid area for fitness determination for ink damage, and a localized soiling valid area usable as the valid area for localized soiling.
According to an another aspect of the present invention there is a paper sheet recognition apparatus, wherein the valid area determining unit determines the localized soiling valid area based on a minimum value included in the statistics, and the ink damage valid area based on a maximum value included in the statistics.
According to an another aspect of the present invention there is a paper sheet recognition apparatus, wherein the valid area determining unit removes the area, having a difference between the maximum value and the minimum value that is above a predetermined value, from the ink damage valid area and the localized soiling valid area., the area having a difference between the maximum value and the minimum value that is above a predetermined value.
According to an another aspect of the present invention there is a paper sheet recognition apparatus, wherein the threshold value determining unit determines an upper limit of the threshold value that indicates an upper limit of a range within which the paper sheet is determined to be fit when the valid area is the ink damage valid area, and a lower limit of the threshold value that indicates a lower limit of the range within which the paper sheet is determined to be fit when the valid area is the localized soiling valid area.
According to an aspect of the present invention there is a paper sheet recognition method for determining fitness of a paper sheet based on a captured image of the paper sheet recognition method according to claim 6. The above and other objects, features, advantages and technical and industrial significance of this invention will be better understood by reading the following detailed description of presently preferred embodiments of the invention, when considered in connection with the accompanying drawings.

BRIEF DESCRIPTION OF DRAWINGS

FIGS. 1A and 1B are drawings illustrating an overview of a paper sheet recognition method according to an embodiment of the present invention;
FIGS. 2A and 2B are schematic diagrams of a line sensor of a paper sheet recognition apparatus according to an embodiment of the present invention;
FIG. 3 is a block diagram of the paper sheet recognition apparatus according to the present embodiment;
FIGS. 4A to 4C are drawings for explaining overlapping regions of an evaluation area;
FIGS. 5A to 5C are drawings for explaining a valid area learning process performed by a valid area learning unit;
FIGS. 6A to 6D are drawings illustrating a threshold value learning process performed by a threshold value learning unit;
FIG. 7 is a flowchart of a learning process executed by the paper sheet recognition apparatus;
FIG. 8 is a flowchart of a localized soiling learning process executed by the valid area learning unit of the paper sheet recognition apparatus;
FIG. 9 is a flowchart of an ink damage learning process executed by the valid area learning unit of the paper sheet recognition apparatus;
FIG. 10 is a drawing illustrating an example of arrangement of sensors included in the paper sheet recognition apparatus; and
FIG. 11 is a flowchart of a fitness determination process executed by a fitness determining unit of the paper sheet recognition apparatus.

BEST MODE(S) FOR CARRYING OUT THE INVENTION

Exemplary embodiments of a paper sheet recognition apparatus and a paper sheet recognition method according to the present invention are explained below with reference to the accompanying drawings. An overview of the paper sheet recognition method according to the present invention is explained with reference to FIGS. 1A and 1B followed by an explanation of exemplary embodiments of the paper sheet recognition apparatus, to which the paper sheet recognition method according to the present invention is applied, with reference to FIGS. 2A to 11.
In the following explanation, each pixel of a captured image of a paper sheet is presumed to have a pixel value, which refers to a value, for example, between 0 (black) and 255 (white), according to brightness of its color.
An overview of the paper sheet recognition method according to the present invention is explained below first with reference to FIGS. 1A and 1B. FIG. 1A is a drawing illustrating an overview of statistics relating to pixel blocks and block values used in the present invention. FIG. 1B is a drawing illustrating an overview of a valid area learning process according to the present invention.
In FIGS. 1A and 1B, a paper sheet 1B is presented as an example. In the paper sheet 1B, a pixel value of each pixel in a localized area Pd is 0 (black) and a pixel value of each pixel in the remaining area is 255 (white).
As shown in FIG. 1A, in the paper sheet recognition method according to the present invention, the captured image of the paper sheet 1B is divided into n number of blocks of a predetermined number of pixels. Statistics relating to a block value are calculated for each block (hereinafter, "pixel block"). The block value is based on an average of the values of the pixels constituting the block. The number of pixels constituting a pixel block is constant even if the paper sheets are of different types (for example, banknotes of different denominations).
Specifically, the paper sheet recognition apparatus divides the captured image of the paper sheet 1B into 24 blocks (n=24) of four pixels each (see FIG. 1A). Each pixel block includes information that identifies its position (for example, a block No. such as 17 of a pixel block B₁₇) (see (A-1) of FIG. 1A). For the sake of convenience, in (A-1) of FIG. 1A, after the paper sheet 1B is divided into pixel blocks, rows of the pixel blocks have been assigned row numbers 1 to 6.
The paper sheet recognition apparatus then classifies each pixel block into one of eight levels from 0 (dark) to 7 (bright) (level converting of the pixel value) based on an average value of the pixel values of the pixels constituting the respective block, and treats the gradation value as a block value. For example, as shown in (A-1) of FIG. 1A, a pixel block B₆ is made of four pixels with each pixel having a pixel value of 0, and therefore, the block value of the pixel block B₆ will be 0. Similarly, the pixel block B₁₇ is made of two pixels having a pixel value of 255 and two pixels having a pixel value of 0, and therefore, the block value of the pixel block B₁₇ will be 3.
The paper sheet recognition apparatus reads a predetermined number of paper sheets (denoted by m) and calculates statistics relating to the block value of each pixel block. The statistics calculated in the paper sheet recognition method according to the present invention are a maximum value, a minimum value, an average value, and a standard deviation of the block value.
In the example shown in (A-2) of FIG. 1A, four paper sheets are read and the statistics of the block value of the pixel block B₁₇ are calculated. If the block values of the pixel block B₁₇ sequentially from the first paper sheet are 3, 7, 0, and 4, the statistics of the block values of the pixel block B₁₇ of the lot of four paper sheets will be calculated as follows: maximum value=7, minimum value=0, average value=3.5, and standard deviation=2.5.
In the paper sheet recognition method according to the present invention, as shown in FIG. 1B, learning of a fitness determination area of the paper sheet is carried out based on the statistics of the block value of each pixel block shown in FIG. 1A. In (B-1) of FIG. 1B, the numbers separated by a "/" shown inside each pixel block B_i indicate the statistics, that is, the maximum value and the minimum value, respectively, of the block values of each of the m number of paper sheets. For example, 7/0 indicates that 7 is the maximum value and 0 is the minimum value.
Specifically, as shown in FIG. 1B, the paper sheet recognition apparatus learns, for each category of fitness determination of the paper sheet, a valid area that is usable for fitness determination for the category as well as a threshold value for the fitness determination for each valid area.
In the paper sheet recognition method according to the present invention, learning of two categories of valid area is carried out. The first category of valid area is for detection of localized soiling on the paper sheet, and the second category of valid area is for detection of ink damage on the paper sheet.
The learning of the valid area for the detection of localized soiling, which is the first category of valid area, is explained first. In a pixel block that is originally dark, detection of a so-called soiling, such as localized soiling, is generally difficult because the soiling does not stand out prominently in the area.
Therefore, in the paper sheet recognition apparatus, a stably bright pixel block is identified as the valid area for the detection of localized soiling. The determination, which is related to whether a pixel block is stably bright, is made based on whether the minimum value in the statistics of the block value is greater than or equal to a predetermined threshold value (see (B-1) of FIG. 1B).
For example, if the predetermined threshold value is taken as 7, the paper sheet recognition apparatus regards "minimum value≥7" as a determination condition for determining whether a pixel block is stably bright, and determines a pixel block B₁₉ in (B-1) of FIG. 1B as the stably bright pixel block because the minimum value of the pixel block B₁₉ is 7.
That is, the paper sheet recognition apparatus identifies the pixel block B₁₉ as the valid area for the detection of localized soiling (see (B-1a) of FIG. 1B).
For learning the valid area for the detection of ink damage, which is the second category of valid area, the paper sheet recognition apparatus identifies a stably dark pixel block as the valid area. The reason why is the detection of ink damage in a pixel block that is originally bright is difficult because the ink damage does not stand out prominently in the area.
Therefore, in the paper sheet recognition apparatus, a stably dark pixel block is identified as the valid area for the detection of ink damage. The determination regarding whether a pixel block is stably dark is made based on whether the maximum value in the statistics of the block value is less than or equal to a predetermined threshold value (see (B-1) of FIG. 1B).
For example, if the predetermined threshold value is taken as 0, the paper sheet recognition apparatus regards "maximum value≤0" as a determination condition for determining whether a pixel block is stably dark, and determines a pixel block B₂₄ in (B-1) of FIG. 1B as the stably dark pixel block because the maximum value of the pixel block 24 is 0.
In other words, the paper sheet recognition apparatus identifies the pixel block B₂₄ as the valid area for the detection of ink damage (see (B-1b) of FIG. 1B).
The paper sheet recognition apparatus then performs the process of identification of the two categories of valid areas for all the pixel blocks. As a result, for example, all the pixel blocks in the first column to the fourth column (see (A-1) of FIG. 1A) of the paper sheet 1B are learned as the valid area for the detection of localized soiling, the pixel blocks in the sixth column (see (A-1) of FIG. 1A) are learned as the valid area for the detection of ink damage.
The pixel blocks in the fifth column (see (A-1) of FIG. 1A), which includes the pixel block B₁₇, do not satisfy either of the determination conditions. In other words, these pixel blocks are neither stably bright nor stably dark. The paper sheet recognition apparatus therefore regards these pixel blocks as an invalid area.
In the paper sheet recognition method according to the present invention, after the pixel blocks are identified as the valid area for the detection of "localized soiling" or "ink damage", some of those blocks may be eliminated as the invalid area depending on a difference between the maximum value and the minimum value of the block value. This process is explained in detail later with reference to FIGS. 5A to 5C.
As shown in (B-2) of FIG. 1B, the paper sheet recognition apparatus then learns the threshold value for fitness determination for each valid area learned in (B-1) of FIG. 1B for each category. The threshold value is calculated according to the statistics of the block value of each pixel block. This process is explained in detail later with reference to FIGS. 6A to 6D.
Thus, in the paper sheet recognition method according to the present invention, the valid area for fitness determination as well as the threshold value for the valid area are learned for each category of fitness determination (for example, localized soiling and ink damage).
Moreover, in the paper sheet recognition apparatus according to the present invention, the learning of the valid area as well as the threshold value corresponding to the valid area is performed for different types of paper sheets; for example, if the paper sheets are banknotes, the learning of the valid area and the threshold value corresponding to the valid area is performed for each denomination of the banknotes. A fitness determination is performed based on a learned result obtained for each type of paper sheet. Therefore, the paper sheet recognition method according to the present invention can perform a highly accurate fitness determination for paper sheets having a wide variety of designs and soiling.
Exemplary embodiments of the paper sheet recognition apparatus that employs the paper sheet recognition method explained with reference to FIGS. 1A and 1B are explained below in detail. The paper sheet recognition apparatus explained below captures an image of the paper sheet with an image line sensor (hereinafter, "line sensor").
In the following explanation, the pixel blocks are assigned numbers from 1 to n. A subscript i is used to denote an arbitrary one of the n number of pixel blocks as B_i. Moreover, the paper sheets are assigned numbers from 1 to m. An arbitrary one of the m number of sheets is represented by a subscript j. Thus, a pixel block B_i from any of the m number of sheets is represented as B_i,j.
FIGS. 2A and 2B are schematic diagrams of a line sensor of the paper sheet recognition apparatus according to the present embodiment. In FIGS. 2A and 2B, only a mechanism that is required for explaining the salient features of the line sensor is shown; a shape or the number of constituent parts of the mechanism are not limited to those shown here. FIG. 2A is a schematic diagram of a single surface reflective and single surface transmissive line sensor 100 that has a multi-wavelength light source. FIG. 2B is a schematic diagram of a both surface reflective and single surface transmissive line sensor 200.
An overview of the line sensor is explained first before explanation is given with reference to FIGS. 2A and 2B. A line sensor is a device used for capturing the image of the paper sheet, and includes several detectors arranged in a line and orthogonal direction to the transport direction of the paper sheet to be imaged. The detector is formed of an array of LEDs (Light Emitting Diodes) as light emitting elements and an array of photo diodes as light receiving elements.
The line sensor captures the image of the paper sheet by detecting a distribution of physical quantities of reflected light and transmitted light, etc., at a predetermined position on the paper sheet that is to be imaged. An optical line sensor is explained in the description given below. However, the optical line sensor is not limited to this embodiment.
As shown in FIG. 2A, the line sensor 100 includes an elongated light emitting unit 110, and an elongated light emitting and photodetecting unit 120. The paper sheet that is the target of fitness determination is passed through a gap between the light emitting unit 110 and the light emitting and photodetecting unit 120 in a transport direction 500.
The light emitting unit 110 includes a line-shaped LED array 111 for transmitting light at two wavelengths integrally disposed with a rod lens 112 for collecting light, and uniformly illuminates the paper sheet passing thereby.
The light emitting and photodetecting unit 120 includes a line-shaped LED array 121 for reflecting light at two wavelengths, a photodiode array 123 for photodetection, a SELFOC lens array (SLA) 122 for limiting an incident light angle to the photodiode array 123, increasing directivity, and improving resolution, and a multiplexer circuit 124 for controlling an integration time of each element of the photodiode array 123.
The LED array 111 for transmitting light and the LED array 121 for reflecting light are controlled by a current-controlled drive circuit. Sensor outputs of the photodiode array 123 are appropriately controlled by the multiplexer circuit 124 using the integration time according to the emitted wavelength, and then output.
The light emitting elements LED arrays 111 and 121 can be a combination of light sources that emit invisible light such as infrared light and light sources that emit visible light (for example, green light) or a combination of RGB light sources according to the type of the paper sheet or the purpose for which paper sheet recognition is being performed. Light emitting elements other than LEDs can also be used.
Although an example in which the LED array for transmitting light at two wavelengths and the LED array for reflecting light at two wavelengths is presented here, the configuration is not to be thus limited; the transmitted light or the reflected light can be appropriately combined.
In FIG. 2A, a single surface reflective and single surface transmissive line sensor is used; however, a both surface reflective and single surface transmissive line sensor can be used instead.
As shown in FIG. 2B, the line sensor 200 includes a first line sensor 210 for scanning one surface of a paper sheet 300 that is to be scanned, and a second line sensor 220 for scanning the other surface of the paper sheet 300. The paper sheet 300 is passed through a gap between the first line sensor 210 and the second line sensor 220 in a transport direction 600.
The first line sensor 210 includes a reflective light source 211 for illuminating one surface of the paper sheet 300 with light of a predetermined wavelength (for example, invisible light such as infrared light or visible light such as green light), a lens 212 for collecting the light from the reflective light source 211 that is reflected by the paper sheet 300, a photodetecting unit 213 for converting the light collected by the lens 212 into an electrical signal, an A/D converting unit 214 for converting the electrical signal output by the photodetecting unit 213 into a digital signal, and a screening unit 215 that screens light emitted by a reflective light source 222 of the second line sensor 220 that is explained later.
The second line sensor 220 includes a transmissive light source 221 and the reflective light source 222 for illuminating the other surface of the paper sheet 300 with light of a predetermined wavelength, a lens 223 for collecting the light from the reflective light source 222 that is reflected by the paper sheet 300, a photodetecting unit 224 for converting the light collected by the lens 223 into an electrical signal, an A/D converting unit 225 for converting the electrical signal output by the photodetecting unit 224 into a digital signal, and a screening unit 226 for screening the light emitted by the reflective light source 211 of the first line sensor 210.
The photodetecting unit 213 of the first line sensor 210 detects a part of the light emitted by the transmissive light source 221 of the second line sensor 220 via the lens 212. Accordingly, the transmissive light source 221 is arranged on an optical axis of the lens 212 of the first line sensor 210.
Although LEDs should preferably be used as the transmissive light source 221 and the reflective light sources 211 and 222, other light emitting elements can also be used. An RGB light source can also be used depending on the type of the paper sheet or the purpose for which paper sheet recognition is being performed.
Instead of the line sensors described above, a both surface reflective and both surface transmissive line sensor can also be adapted to the present invention. The both surface reflective and single surface transmissive line sensor 200 shown in FIG. 2B is used in the present embodiment.
A structure of the paper sheet recognition apparatus according to the present embodiment is explained below with reference to FIG. 3. FIG. 3 is a block diagram of a paper sheet recognition apparatus 10 according to the present embodiment. Only structural elements that are required for explaining the salient features of the paper sheet recognition apparatus 10 are shown in FIG. 3, and description of general structural elements is omitted.
As shown in FIG. 3, the paper sheet recognition apparatus 10 includes a line sensor section 11, a control section 12, and a storage section 13. The control section 12 includes a block setting unit 12a, a pixel statistics unit 12b, a valid area learning unit 12c, a threshold value learning unit 12d, and a fitness determining unit 12e. The storage section 13 stores therein setting information 13a, pixel statistics information 13b, valid area information 13c, threshold value correction information 13d, and threshold value information 13e.
The line sensor section 11 is an input device and corresponds with the line sensor 100 or 200 shown in FIGS. 2A and 2B. When learning the fitness determination area range of the paper sheet, the line sensor section 11 sends the captured image of the fit note of the paper sheet to the block setting unit 12a. Furthermore, when performing the fitness determination of the paper sheet, the line sensor section 11 sends the captured image of the paper sheet to the fitness determining unit 12e.
The control section 12 is a processing section that learns the fitness determination area range based on the captured image received from the line sensor section 11, and performs fitness determination of the paper sheet based on the threshold value information 13e, which is an output result of the fitness determination area range learning.
The block setting unit 12a is a processing unit that performs a process of dividing the captured image of the paper sheet received from the line sensor section 11 into pixel blocks of a predetermined number of pixels. The predetermined number of pixels referred to here is stored as the setting information 13a in the storage section 13 beforehand and it is constant even if the paper sheets are of different types (for example, different denominations of banknotes). Moreover, the size of the pixel block can be changed for each type of the paper sheet.
The block setting unit 12a is also a processing unit that performs a process of providing the captured image divided into blocks of a predetermined number of pixels to the pixel statistics unit 12b.
The pixel statistics unit 12b is a processing unit that performs a process of calculating the statistics relating to the pixel block values of a plurality of paper sheets based on the pixel blocks of the captured image received from the block setting unit 12a. The statistics include the maximum value, the minimum value, the average value, and the standard deviation of the block values.
The pixel statistics unit 12b is also a processing unit that performs a process of storing the calculated statistics of the block values as the pixel statistics information 13b in the storage section 13.
The valid area learning unit 12c is a processing unit that performs a process of learning the valid area for fitness determination by referring to the pixel statistics information 13b stored by the pixel statistics unit 12b in the storage section 13 (see FIGS. 1A and 1B). Furthermore, the valid area learning unit 12c is also a processing unit that performs a process of storing the learned valid area of the paper sheet as the valid area information 13c in the storage section 13.
The valid area learning process performed by the valid area learning unit 12c is explained in greater detail with reference to FIGS. 4A to 4C and FIGS. 5A to 5C. FIGS. 4A to 4C are drawings for explaining overlapping regions of an evaluation area, while FIGS. 5A to 5C are drawings for explaining the valid area learning process performed by the valid area learning unit 12c.
The overlapping regions of the evaluation area are explained below with reference to FIGS. 4A to 4C. As a prerequisite for the valid area learning process, areas of a fixed size which are divided from the paper sheet are used as objects to be processed from the beginning of the valid area learning process, or identified as areas not to be used in the process. The areas of fixed size serve as the evaluation areas. The size of the evaluation area remains unchanged even if the sizes of paper sheets differ.
In FIGS. 4A to 4C, two paper sheets X and Y differing in size, the paper sheet X being larger than the paper sheet Y, are shown with each of the paper sheets being divided into evaluation areas Ai of fixed size (see (Prerequisite Condition) of FIG. 4A). FIG. 4B is a drawing illustrating the paper sheet X divided into the evaluation areas Ai, while FIG. 4C is a drawing illustrating the paper sheet Y divided into the evaluation areas Ai.
As shown in FIG. 4B, if the size of the paper sheet X is a multiple of the size of the evaluation area Ai, no overlapping region will be present among evaluation areas A1 to A4. In the example shown in FIG. 4A, the size of the paper sheet X is four times that of the evaluation area Ai.
In this case, the valid area learning unit 12c subjects all the evaluation areas A1 to A4 to the valid area learning process.
On the other hand, as shown in FIG. 4C, if the size of the paper sheet Y is not a multiple of the size of the evaluation area Ai, an overlapping region is formed when the paper sheet Y is divided into the evaluation blocks A1 to A4 (see shaded area of FIG. 4C).
In this case, the valid area learning unit 12c excludes the overlapping region from the valid area learning process to prevent incorrect fitness determination. In other words, the overlapping region is regarded as an invalid area from the beginning of the valid area learning process.
To make the explanation easy to understand, the present embodiment is explained taking the case shown in FIG. 4B in which there is no overlapping region among the evaluation areas A1 to A4.
The valid area learning process performed by the valid area learning unit 12c is explained below in detail with reference to FIGS. 5A to 5C. FIG. 5A is a drawing illustrating an example of identifying the valid area for localized soiling and ink damage based on the maximum value and the minimum value of the block values; FIG. 5B is a drawing illustrating an example of identifying the valid area based on the difference value; and FIG. 5C is a drawing illustrating an example of finally identifying the valid area based on results of the examples shown in FIGS. 5A and 5B.
A case is explained below in which the valid area learning unit 12c performs the process using the statistics of the block values of the pixel blocks of the m number of paper sheets 5B (see FIG. 5A) that include localized areas P1 and P2. The localized area P2, the localized area P1, and an area other than the localized areas P2 and P1 (hereinafter "other area") have sequentially increasing brightness. The block value of the pixel blocks in the localized area P2 is 0, and the block value of the pixel blocks in the other area is 7. The block values of the pixel blocks straddling the localized area P2 and the other area range from 0 to 7, and the block values of the pixel blocks that straddling the localized area P1 and the other area range from 3 to 7.
The statistical values of the block values of the pixel blocks B₁₇, B₁₉, B₂₀, and B₂₄ of the paper sheet 5B are presented as an example in the explanation (see FIGS. 5A to 5C).
As shown in FIG. 5A, the valid area learning unit 12c identifies the valid areas for localized soiling and ink damage based on the maximum value and the minimum value of the block value of each pixel block statistically calculated by the pixel statistics unit 12b and stored as the pixel statistics information 13b.
The numbers demarcated by a "/" shown inside each pixel block B_i indicate the statistics, that is, the maximum value and the minimum value, respectively, of the block values of each of the m number of paper sheets. For example, 7/0 indicates that 7 is the maximum value and 0 is the minimum value.
Specifically, for the detection of localized soiling in the paper sheet 5B, the valid area learning unit 12c regards those pixel blocks B_i as the valid area whose minimum value of the block value is greater than or equal to each of the predetermined threshold value set for the pixel block B_i. In other words, for the detection of localized soiling such as stains, the valid area learning unit 12c regards those areas that are pattern-wise darker than the soiling as the invalid area and eliminates such areas from the target of the fitness determination right at the beginning, and only regards stably bright areas as the valid area (see (B-1a) of FIG. 1B).
Similarly, for the detection of ink damage on the paper sheet 5B, the valid area learning unit 12c regards those pixel blocks B_i as the valid area whose maximum value of the block value is less than or equal to the predetermined threshold value set for the pixel block B_i. In other words, for the detection of ink damage, the valid area learning unit 12c regards those areas that are originally brighter than the ink damage area as the invalid area and eliminates such areas from the target of the fitness determination, and only regards stably dark areas as the valid area (see (B-1b) of FIG. 1B).
Accordingly, as shown in FIG. 5A, if the condition for valid area for the detection of localized soiling is "minimum value≥1", the pixel blocks B₁₉ and B₂₀ that satisfy the condition (that is, the pixel blocks are stably bright) are regarded as the valid area, and the pixel blocks B₁₇ and B₂₄ that do not satisfy the condition are regarded as the invalid area.
If the condition for valid area for the detection of ink damage is "maximum value≤4", the pixel block B₂₄ that satisfies the condition (that is, the pixel block is stably dark) is regarded as the valid area, and the pixel blocks B₁₇, B₁₉, and B₂₀ that do not satisfy the condition are regarded as the invalid area.
Thereafter, as shown in FIG. 5B, the valid area learning unit 12c identifies the valid area based on the difference between the minimum value and the maximum value among the block values of the pixel blocks of the m number of paper sheets stored as the pixel statistics information 13b. The number inside each pixel block B_i denotes the difference (maximum value-minimum value).
Specifically, the valid area learning unit 12c regards the pixel block B_i whose difference between the block values is less than or equal to the predetermined threshold value (smaller difference) as the valid area. That is, if the difference between the block values of the pixel block B_i is larger, it would indicate a possibility that the area is a boundary between the patterns with large difference in brightness. Therefore, such areas are eliminated from the target of the fitness determination as invalid areas beforehand, and only the areas with smaller difference in brightness are regarded as the valid area.
Accordingly, as shown in FIG. 5B, if the condition for valid area in terms of the difference is "difference value≤3", the pixel blocks B₁₉ and B₂₄ that satisfy the condition (that is, the pixel blocks in which the difference in brightness is smaller) are regarded as the valid area, and the pixel blocks B₁₇ and B₂₀ that do not satisfy the condition are regarded as the invalid area.
As shown in FIG. 5C, the valid area learning unit 12c then finally identifies as the valid area the pixel block B_i that is regarded as the valid area in FIG. 5A as well as FIG. 5B.
If an area that is regarded as the valid area in FIG. 5A but is regarded as the invalid area in FIG. 5B because of the difference, that area is no longer included in the valid area.
In other words, as shown in FIG. 5C, the pixel block B₂₀ that is regarded as the valid area for the detection of localized soiling in FIG. 5A is regarded as the invalid area due to the large difference in brightness (difference value=4) in FIG. 5B. Therefore, the pixel block B₂₀ is finally removed from the valid area.
Accordingly, as shown in FIG. 5C, the pixel blocks regarded as the valid area in both FIGS. 5A and 5B, the pixel block B₁₉ for the detection of localized soiling and the pixel block B₂₄ for the detection of ink damage are identified as the valid area.
The conditions shown in FIGS. 5A to 5C are presented merely for the purpose of explaining the paper sheet recognition method according to the present invention, and do not limit the process performed by the valid area learning unit 12c. The predetermined threshold values set for each pixel block that is used in the conditions can be stored beforehand as the setting information 13a in the storage section 13.
Returning to FIG. 3, the threshold value learning unit 12d is a processing unit that refers to the valid area information 13c stored in the storage section 13 by the valid area learning unit 12c and the threshold value correction information 13d, and calculates an upper limit and a lower limit of the threshold value according to the statistics of the block value for each pixel block in the valid area. In other words, the threshold value learning unit 12d is a processing unit that learns the threshold values valid for fitness determination.
Moreover, the threshold value learning unit 12d is also a processing unit that stores the upper and the lower limits of the threshold values learned for each pixel block as the threshold value information 13e in the storage section 13.
A threshold value learning process performed by the threshold value learning unit 12d is explained in further detail with reference to FIGS. 6A to 6D. FIGS. 6A to 6D are drawings illustrating the threshold value learning process performed by the threshold value learning unit 12d. FIG. 6A is a drawing illustrating basic learning examples of the upper and the lower limits of the threshold values; FIG. 6B is a drawing illustrating correction learning examples of the upper and the lower limits of the threshold values; FIG. 6C is a drawing illustrating concrete calculation examples of the upper and the lower limits of the threshold values; and FIG. 6D is a drawing illustrating examples of stored threshold value information 13e.
The predetermined threshold values and calculation formulae appearing in the explanation are merely examples, and do not limit the process performed by the threshold value learning unit 12d. The predetermined threshold values and the calculation formulae are presumed to be already stored as the threshold value correction information 13d in the storage section 13.
As shown in FIG. 6A, for detecting the localized soiling that requires that the pixel block is stably bright, the threshold value learning unit 12d calculates the lower limit of the threshold value of the block value of each pixel block. For detecting the ink damage that requires that the pixel block is stably dark, the threshold value learning unit 12d calculates the upper limit of the threshold value of the block value of each pixel block.
The threshold value learning unit 12d calculates the upper and the lower limits of the threshold value according to the statistics of the block value of each pixel block. The threshold value learning unit 12d uses the maximum value, the minimum value, the average value, and the standard deviation of the block value that are included in the statistics.
For example, as shown in FIG. 6A, the threshold value learning unit 12d can calculate the basic lower limit of the threshold value by subtracting a value that is four times the standard deviation of the block value from the average value of the block value. Similarly, the threshold value learning unit 12d can calculate the basic upper limit of the threshold value by increasing the maximum value of the block value by 20%.
As shown in FIG. 6B, if the minimum value of the block value of the pixel block B_i is less than or equal to the predetermined threshold value, the threshold value learning unit 12d corrects the calculated basic value of the lower limit of the threshold value(see FIG. 6A) in a negative direction. That is, even in the valid area, for the pixel block that is originally dark, the threshold value learning unit 12d lowers (relaxes) the lower limit of the threshold value. Accordingly, determination is performed leniently for areas that are stably dark, and strictly for areas that are not stably dark.
As shown in FIG. 6B, if the maximum value of the block value of the pixel block is greater than or equal to the predetermined threshold value, the threshold value learning unit 12d corrects the calculated basic upper limit (see FIG. 6A) in a positive direction. In other words, even in the valid area, for the pixel block that is originally bright, the threshold value learning unit 12d raises (relaxes) the upper limit of the threshold value. Accordingly, determination is performed leniently for areas that are stably bright, and strictly for areas that are not stably bright.
Thus, for the correction of the upper and the lower limits of the threshold value also, the threshold value learning unit 12d uses the maximum value, the minimum value, the average value, and the standard deviation of the block value that are included in the statistics.
For example, as shown in FIG. 6B, the threshold value learning unit 12d corrects the lower limit of the threshold value by subtracting a value that is the standard deviation of the block value from the basic lower limit of the threshold value shown in FIG. 6A. Similarly, the threshold value learning unit 12d corrects the upper limit of the threshold value by adding 10% of the maximum value of the block value to the basic upper limit of the threshold value shown in FIG. 6A.
An example of calculating the upper and the lower limits of the threshold value shown with reference to FIGS. 6A and 6B is explained concretely with reference to FIG. 6C. It is presumed that four paper sheets are read, and the block values of a pixel block B_i,j, a second pixel block B_i,j+1, a third pixel block B_i,j+2, and a fourth pixel block B_i,j+3 of the first paper sheet are 4, 3, 4, and 3, respectively. The statistics of the block value of the pixel block B_i of the four paper sheets are as follows; maximum value=4, minimum value=3, average value=3.5, and standard deviation=0.5.
Because the minimum value of the block values of the pixel blocks B_i of the four paper sheets is less than or equal to 3, correction of the lower limit of the threshold value up to what is shown in FIG. 6B is required. Therefore, the lower limit of the threshold value is calculated as 1.0 by the following formula: 3.5 (average value)-5×0.5 (standard deviation). Moreover, because the maximum value of the block value of the pixel block B_i is greater than or equal to 4, correction of the upper limit of the threshold value is also required up to what is shown in FIG. 6B. Therefore, the upper limit of the threshold value is calculated as 5.2 by the following formula: 4 (maximum value)×130(%).
Thereafter, the threshold value learning unit 12d stores the calculated upper and lower limits of the threshold value of each pixel block as the threshold value information 13e in the storage section 13. FIG. 6D is a drawing illustrating examples of the threshold value information 13e stored thus.
As shown in FIG. 6D, in the threshold value information 13e, the upper and the lower limits of the threshold value of each pixel block are stored for position information (for example, block No.) of each pixel block that enables the pixel block to be specified.
As shown in FIG. 6D, only the lower limit of the threshold value could be stored for the pixel block. For example, only the lower limit 2.0 is stored for the pixel block B₁, indicating that the pixel block B₁ is the valid area for the detection of localized soiling.
Similarly, only the upper limit 8.4 could be stored for the pixel block B₂, indicating that the pixel block B₂ is the valid area for the detection of ink damage.
Accordingly, the pixel block B₃ for which neither the lower limit nor the upper limit is stored will be the invalid area not to be used for the detection of either localized soiling or ink damage.
The pixel block B_n for which both the lower limit and the upper limit are stored will be the valid area for the detection of both the localized soiling and the ink damage.
Returning to FIG. 3, the fitness determining unit 12e is a processing unit that refers to the threshold value information 13e stored in the storage section 13 by the threshold value learning unit 12d, and performs fitness determination of the paper sheet.
Specifically, the fitness determining unit 12e first divides the captured image of the paper sheet that is the target of fitness determination received from the line sensor section 11 into pixel blocks of predetermined number of pixels. The predetermined number of pixels is the same as the predetermined number of pixels used at the fitness determination area range learning stage.
The fitness determining unit 12e thereafter verifies whether a summation value of all the pixels in the pixel block is within the upper and the lower limits of the threshold value (see FIGS. 6A to 6D) based on the statistics of the block value of each pixel block and the position information in the threshold value information 13e (see FIGS. 6A to 6D). The fitness determining unit 12e may also use only the maximum value or the minimum value of the block value for verification.
Thereafter, the fitness determining unit 12e counts the number of pixel blocks that are outside the range of the upper and the lower limits of the threshold value in the verification. If the count exceeds a predetermined unfit note determining threshold value, the fitness determining unit 12e determines the paper sheet to be an unfit note. The predetermined unfit note determining threshold value can be set separately according to the statistics of the block value of each pixel block.
The storage section 13 is a memory device such as a hard disk drive or a non-volatile memory device. The storage section 13 stores therein the setting information 13a, the pixel statistics information 13b, the valid area information 13c, the threshold value correction information 13d, and the threshold value information 13e.
The setting information 13a and the threshold value correction information 13d are already stored as setting values through pre-operational verification testing of the paper sheet recognition apparatus 10. The setting values can be appropriately changed according to the operation. The settings and change of settings can be accomplished by hardware or software.
The setting information 13a is information relating to the pixel blocks and is referred to by the block setting unit 12a for dividing the image of the paper sheet into pixel blocks. Specifically, the setting information 13a includes predetermined number of pixels constituting the pixel block. The setting information 13a can also include information relating to initial settings of the paper sheet recognition apparatus 10.
The pixel statistics information 13b is statistical information relating to each pixel constituting a pixel block, and is registered by the pixel statistics unit 12b. Specifically, the pixel statistics information 13b includes the maximum value, the minimum value, the average value, and the standard deviation of the block value of each pixel block of the m number of paper sheets. Moreover, the pixel statistics information 13b is referred to by the valid area learning unit 12c when performing the valid area learning process (see FIGS. 5A to 5C).
The valid area information 13c is information relating to the identification of the valid area and is registered by the valid area learning unit 12c. Specifically, the valid area information 13c includes the block No. for identification of the valid area. The valid area information 13c is referred to by the threshold value learning unit 12d when performing the threshold value learning process (see FIGS. 6A to 6D).
The threshold value correction information 13d is information relating to the upper and the lower limits of the threshold value of the pixel blocks. Specifically, the threshold value correction information 13d includes the predetermined threshold values relating to the calculation of the upper and the lower limits of the threshold values, and condition formulae relating to the calculation and correction of the upper and the lower limits of the threshold value. The threshold value correction information 13d is referred to by the threshold value learning unit 12d when performing the threshold value learning process (see FIGS. 6A to 6D). Because the threshold value information 13e has already been described above (see FIGS. 6A to 6D), its description is omitted here.
The paper sheet recognition apparatus 10 for determining the fitness of the paper sheet is explained so far; however, an apparatus in which the fitness determining unit 12e shown in FIG. 3 is omitted and that functions as a threshold value learning apparatus can also be configured. In this case, the threshold value learning apparatus generates the threshold value information 13e using a predetermined number of fit notes. The threshold value information 13e generated by the threshold value learning apparatus is made available to other devices.
A process procedure of a learning process executed by the paper sheet recognition apparatus 10 is described below with reference to FIG. 7. FIG. 7 is a flowchart of the learning process executed by the paper sheet recognition apparatus 10. The learning process corresponds to the process executed by the threshold value learning apparatus.
As shown in FIG. 7, the paper sheet recognition apparatus 10 sets reference learning data (Step S101). The act of setting the reference learning data includes inputting a plurality of captured images of fit notes, such as brand new banknotes, as the reference learning data for paper sheets.
The paper sheet recognition apparatus 10 then additionally sets printing shift margin data (Step S102). That is, because there is a possibility of printing shift even in case of banknotes, the paper sheet recognition apparatus 10 forcefully adds margin data that corresponds to shifting, and performs setting such that edge regions of the patterns of banknotes, etc., do not get included in the valid area.
The paper sheet recognition apparatus 10 then divides the captured image into blocks of predetermined number of pixels (Step S103). Next, the paper sheet recognition apparatus 10 changes the direction of the captured image that has been divided into the pixel blocks (Step S104).
Thereafter, the paper sheet recognition apparatus 10 calculates the statistics of the block value for each pixel block (Step S105). The statistics include the maximum value, the minimum value, the average value, and the standard deviation of the block values of the pixel blocks of the m number of paper sheets.
The paper sheet recognition apparatus 10 then performs a localized soiling learning process in which the fitness determination area range for localized soiling of the paper sheet is learned (Step S106). The localized soiling learning process is described later with reference to FIG. 8.
The paper sheet recognition apparatus 10 then performs an ink damage learning process in which the fitness determination area range for ink damage of the paper sheet is learned (Step S107). The ink damage learning process is described later with reference to FIG. 9.
The paper sheet recognition apparatus 10 then determines whether the processes are completed for both the faces of the paper sheet (Step S108). If the processes are completed only for one side of the faces (No at Step S108), the paper sheet recognition apparatus 10 repeats all the processes from Step S103 for the other side of the faces. If the processes are completed for both the faces (Yes at Step S108), the paper sheet recognition apparatus 10 ends the process.
A process procedure of the localized soiling learning process of FIG. 7 (see Step S106 of FIG. 7) is described below with reference to FIG. 8. FIG. 8 is a flowchart of the localized soiling learning process executed by the valid area learning unit 12c of the paper sheet recognition apparatus 10.
As shown in FIG. 8, the valid area learning unit 12c of the paper sheet recognition apparatus 10 sets the overlapping region of the pixel blocks (see FIGS. 4A to 4C) as the invalid area (Step S201).
The valid area learning unit 12c then refers to the pixel statistics information 13b and sets a pixel level for each pixel block based on the minimum value of the block value (Step S202). The act of setting the pixel level, for example, includes setting the pixel blocks having the minimum value of the block value ranging from 0 to 7 to eight levels from 0 to 7 depending on the minimum value. The valid area learning unit 12c then sets the pixel blocks with a level that is greater than or equal to a predetermined level as a valid area I (Step S203).
Thereafter, the valid area learning unit 12c refers to the pixel statistics information 13b, and calculates the difference for each pixel block by subtracting the minimum value from the maximum value of the block value (Step S204).
Similar to Step S202, the valid area learning unit 12c then sets a level for each pixel block based on the calculated difference (Step S205). The valid area learning unit 12c then sets the pixel block having a level smaller than or equal to a predetermined level as a valid area II (Step S206).
The valid area learning unit 12c then sets the pixels blocks set both as the valid area I (see Step S203) and the valid area II (see Step S206) as the valid area for the detection of localized soiling and learns the same (Step S207).
Thereafter, the valid area learning unit 12c calculates the basic lower limit of the threshold value of the valid area for the detection of localized soiling (Step S208), and corrects the basic lower limit of the threshold value according to the level of each pixel block (Step S209). The level of each pixel block is set based on the statistics of the block value of the pixel statistics information 13b (see FIGS. 6A to 6D).
The valid area learning unit 12c then stores the lower limit of the threshold value of the valid area for the detection of localized soiling as the threshold value information 13e (Step S210), and ends the process.
A process procedure of the ink damage learning process of FIG. 7 (see Step S107 of FIG. 7) is described with reference to FIG. 9. FIG. 9 is a flowchart of the ink damage learning process executed by the valid area learning unit 12c of the paper sheet recognition apparatus 10.
As shown in FIG. 9, the valid area learning unit 12c of the paper sheet recognition apparatus 10 sets the overlapping region of an evaluation area Ai (see FIGS. 4A to 4C) as the invalid area (Step S301).
The valid area learning unit 12c then refers to the pixel statistics information 13b and sets a pixel level for each pixel block based on the maximum value of the block value (Step S302). The act of setting the pixel level, for example, includes setting the pixel blocks having the maximum value of the block value ranging from 0 to 7 to eight levels from 0 to 7 depending on the maximum value. The valid area learning unit 12c then sets the pixel blocks with a level that is less than or equal to a predetermined level as a valid area I (Step S303).
Thereafter, the valid area learning unit 12c refers to the pixel statistics information 13b, and calculates the difference for each pixel block by subtracting the minimum value from the maximum value of the block value (Step S304).
Similar to Step S302, the valid area learning unit 12c then sets a level for each pixel block based on the calculated difference (Step S305). The valid area learning unit 12c then sets the pixel block having a level less than or equal to a predetermined level as a valid area II (Step S306).
Thereafter, the valid area learning unit 12c sets the pixels blocks set both as the valid area I (see Step S303) and the valid area II (see Step S306) as the valid area for the detection of ink damage and learns the same (Step S307).
The valid area learning unit 12c then calculates the basic upper limit of the threshold value of the valid area for the detection of ink damage (Step S308), and corrects the basic upper limit of the threshold value according to the level of each pixel block (Step S309). The level of each pixel block is set based on the statistics of the block value of the pixel statistics information 13b (see FIGS. 6A to 6D).
The valid area learning unit 12c then stores the upper limit of the threshold value of the valid area for the detection of ink damage in the threshold value information 13e for each pixel block B_i (Step S310), and ends the process.
The paper sheet recognition apparatus 10 that includes the line sensor is explained so far. However, the paper sheet recognition apparatus 10 can include other types of sensors in addition to the line sensor.
The paper sheet recognition apparatus 10 that includes the line sensor and various other types of sensors is described below with reference to FIG. 10. FIG. 10 is a drawing illustrating an example of arrangement of the sensors included in the paper sheet recognition apparatus 10 through a schematic diagram of a top view of a sensor section 50 of the paper sheet recognition apparatus 10 viewed from a positive direction of a Y-axis (upper part of FIG. 10) and a schematic diagram of a side view of the sensor section 50 of the paper sheet recognition apparatus 10, viewed from a negative direction of a Z-axis (lower part of FIG. 10).
As shown in FIG. 10, the paper sheet recognition apparatus 10 can include the sensor section 50 including various types of sensors such as timing sensors 51 and 56.
The timing sensors 51 and 56 emit infrared light to detect a transport timing of the paper sheet, and include light emitting units 51a and 56a and photodetecting units 51b and 56b, respectively.
The fluorescence sensor 53 emits ultraviolet light, etc., and detects a fluorescent component included in a printing surface of the paper sheet. The fluorescence sensor 53 is used for determining the authenticity of the paper sheet.
The thickness detecting sensor 54 is a displacement sensor that detects a displacement of the paper sheet when it is being passed through a gap between a roller 54a and a fixed roller 54b. The thickness detecting sensor 54 can be used, for example, in detecting whether more than one paper sheet is being transported at a time or in the case of an altered banknote such as one with an adhesive tape on it.
The magnetic sensor 55 detects a presence or absence of magnetic field, or the intensity of the magnetic field from the printing surface of the paper sheet. The magnetic sensor 55 includes a magnetic head 55a and a hair-planted roller 55b that presses the paper sheet to the magnetic head 55a. The magnetic sensor 55 is used for determining the authenticity of the paper sheet.
The line sensor 52 (including light emitting and photodetecting parts 52a and 52b) corresponds to the line sensor 100 or the line sensor 200 of FIGS. 2A and 2B; hence, the description thereof is omitted.
As shown in the side view drawing of FIG. 10, the paper sheet recognition apparatus 10 transports the paper sheet 300 in a transport direction 700 (in the negative direction of X-axis in the side view drawing), receives detection signals from each of the sensors, and performs various types of recognition, including the fitness determination according to the present invention, based on the received signals.
A process procedure of the fitness determination process executed by the fitness determining unit 12e of the paper sheet recognition apparatus 10 is described below with reference to FIG. 11. FIG. 11 is a flowchart of the fitness determination process executed by the fitness determining unit 12e of the paper sheet recognition apparatus 10.
As shown in FIG. 11, the fitness determining unit 12e of the paper sheet recognition apparatus 10 performs initial settings such as inputting the captured image of a fit note of the paper sheet that is the target of fitness determination as recognition data (Step S401). The fitness determining unit 12e then divides the captured image into pixel blocks of a predetermined number of pixels (Step S402).
Thereafter, the fitness determining unit 12e determines whether the block value of the pixel block of each paper sheet is not below the lower limit of the threshold value stored as the threshold value information 13e (Step S403). If the block value is not below the lower limit of the threshold value (Yes at Step S403), the fitness determining unit 12e carries control process forward to Step S405.
If the determination condition at Step S403 is not satisfied, that is, if localized soiling of the pixel block is detected (No at Step S403), the fitness determining unit 12e increments a NG count for the detection of localized soiling (Step S404).
Thereafter, the fitness determining unit 12e determines whether the block value of the pixel block of each paper sheet is not above the upper limit of the threshold value stored as the threshold value information 13e (Step S405). If the block value is not above the upper limit of the threshold value (Yes at Step S405), the fitness determining unit 12e carries its control process forward to Step S407.
If the determination condition at Step S405 is not satisfied, that is, if ink damage of the pixel block is detected (No at Step S405), the fitness determining unit 12e increments a NG count for the detection of ink damage (Step S406).
The fitness determining unit 12e then determines whether verification has been performed for all the pixel blocks. If verification has been performed for all the pixel blocks (Yes at Step S407), the fitness determining unit 12e carries its control process forward to Step S408.
If the determination condition at Step S407 is not satisfied (No at Step S407), the fitness determining unit 12e repeats all the processes from Step S403.
The fitness determining unit 12e then determines whether the NG count for localized soiling (see Step S404) and the NG count for ink damage (see Step S406) are within a predetermined unfit note determining threshold value (Step S408).
If the NG count for localized soiling and the NG count for ink damage are within the predetermined unfit note determining threshold value (Yes at Step S408), the fitness determining unit 12e determines that the paper sheet to be a fit note with no soiling (Step S409), and carries its control process forward to Step S411.
The fitness determining unit 12e then determines whether the processes have been completed for both the faces of the paper sheet (Step S411). If the processes have been completed only for one side of the two faces (No at Step S411), the fitness determining unit 12e repeats all the processes from Step S402 for the other side. If the processes have been completed for both the faces (Yes at Step S411), the fitness determining unit 12e ends the process.
If the determination condition at Step S408 is not satisfied (No at Step S408), the fitness determining unit 12e determines the paper sheet to be an unfit note with soiling (Step S410) and ends the process.
Thus, in the present embodiment, the paper sheet recognition apparatus is configured in such a way that the block setting unit divides the captured image of the paper sheet into pixel blocks of a predetermined number of pixels, the pixel statistics unit calculates the statistics of the block value for each pixel block B_i of the m number of paper sheets, the valid area learning unit learns the pixel blocks valid for fitness determination as the valid area, the threshold value learning unit learns the threshold values of the valid area, and the fitness determining unit performs the fitness determination of the paper sheet based on the threshold value information output by the threshold value learning unit. Therefore, the paper sheet recognition apparatus can perform a highly accurate fitness determination for paper sheets having a wide variety of designs and soiling.
In the embodiment of the paper sheet recognition apparatus described above, the threshold value learning unit calculates the threshold values for the valid area identified by the valid area learning unit. According to one aspect of the present invention, using a captured image of a paper sheet to be learned as a learning-purpose image that is divided into areas of a predetermined number of pixels, statistics are calculated for each area by statistically calculating a summation value for the pixels included in the area for a plurality of the learning-purpose images, a valid area that is usable for fitness determination of the paper sheet is determined based on the calculated statistics, a fitness determining threshold value is determined, based on the calculated statistics, for each determined valid area, and the fitness of the paper sheet that is the target of fitness determination is determined based on the determined threshold value. Consequently, a highly accurate fitness determination can be performed for paper sheets having a wide variety of designs and soiling. According to another aspect of the present invention, the valid area determination includes determination of an ink damage valid area usable as the valid area for fitness determination for ink damage, and determination of a localized soiling valid area usable as the valid area for fitness determination for localized soiling. Consequently, fitness determination can be performed according to the type of soiling of the paper sheet that is the target of fitness determination.
According to still another aspect of the present invention, the localized soiling valid area is determined based on a minimum value included in the statistics, and the ink damage valid area is determined based on a maximum value included in the statistics. Consequently, accurate fitness determination can be performed according to a brightness of the area for each type of damage of the paper sheet that is the target of fitness determination.
According to still another aspect of the present invention, the area having a difference between the maximum value and the minimum value included in the statistics that is above a predetermined value is removed from the ink damage valid area and the localized soiling valid area. Consequently, areas where there is a large difference between brightness and darkness that could lead to incorrect fitness determination are removed and a highly accurate fitness determination can be performed.
According to still another aspect of the present invention, an upper limit of the threshold value that indicates an upper limit of a range within which the paper sheet that is the target of fitness determination is determined to be fit is determined when the valid area is the ink damage valid area, and a lower limit of the threshold value that indicates a lower limit of the range within which the paper sheet that is the target of fitness determination is determined to be fit is determined when the valid area is the localized soiling valid area. Consequently, a fitness determination range according to the type of the damage of the paper sheet can be learned.
The paper sheet recognition apparatus and the paper sheet recognition method according to the present invention can perform a highly accurate fitness determination for paper sheets having a wide variety of designs and damages, and are particularly suitable in situations that require recognizing whether highly circulated paper sheets, such as banknotes, are fit for circulation.

Claims

A paper sheet recognition apparatus for determining fitness of a paper sheet based on a captured image of the paper sheet, the paper sheet recognition apparatus comprising:
a storage section (13) that stores threshold value information;

a fitness determining unit (12e) that determines the fitness of the paper sheet based on a threshold value in a valid area by referring to the threshold value information in the storage section (13), the fitness including a plurality of categories;

a statistics calculating unit (12b) that uses the captured image of the paper sheet to be learned as a learning-purpose image that is divided into areas of a predetermined number of pixels, and calculates statistics for each of the areas by statistically calculating a summation value for the pixels included in the area for a plurality of the learning-purpose images; and

a valid area determining unit (12c) that determines the valid area of the image for each of the plurality of categories based on the statistics;

characterized by

a threshold value determining unit (12d) that determines the threshold value in the valid area for each of the plurality of categories based on the statistics and stores a determined threshold value as the threshold value information in the storage section (13).
The paper sheet recognition apparatus according to Claim 1, wherein the valid area determining unit (12c) determines an ink damage valid area usable as the valid area for fitness determination for ink damage as one of the plurality of categories, and a localized soiling valid area usable as the valid area for localized soiling as another one of the plurality of categories.
The paper sheet recognition apparatus according to Claim 2, wherein the valid area determining unit (12c) determines the localized soiling valid area based on a minimum value included in the statistics, and the ink damage valid area based on a maximum value included in the statistics.
The paper sheet recognition apparatus according to Claim 3, wherein the valid area determining unit (12c) removes the area, having a difference between the maximum value and the minimum value that is above a predetermined value, from the ink damage valid area and the localized soiling valid area.
The paper sheet recognition apparatus according to any one of Claims 2 to 4, wherein the threshold value determining unit (12d) determines an upper limit of the threshold value that indicates an upper limit of a range within which the paper sheet is determined to be fit when the valid area is the ink damage valid area, and a lower limit of the threshold value that indicates a lower limit of the range within which the paper sheet is determined to be fit when the valid area is the localized soiling valid area.
A paper sheet recognition method for determining fitness of a paper sheet based on a captured image of the paper sheet, the paper sheet recognition method comprising:
determining the fitness of the paper sheet based on a threshold value in a valid area by referring to a threshold value information stored in a storage section (13), the fitness including a plurality of categories;

calculating, using the captured image of the paper sheet to be learned as a learning-purpose image that is divided into areas of a predetermined number of pixels, statistics for each of the areas by statistically calculating a summation value of the pixels included in the area for a plurality of the learning-purpose images;

determining the valid area of the image for each of the plurality of categories based on the statistics; and

determining the threshold value in the valid area for each of the plurality of categories based on the statistics;

characterized by

storing a determined threshold value as the threshold value information in the storage section (13).