JP3804725B2 - Bill identification device and method for creating identification reference data - Google Patents

Description

【数２】

【数３】

【数４】

【数５】

【数６】

【数７】

【数８】

[0001]
[Technical field to which the invention belongs]
The present invention relates to a bill recognition device and a method for creating identification reference data that, for example, authenticate or identify a paper sheet having a print pattern such as a bill, a securities, a ticket, or the like.
[0002]
[Prior art]
In the conventional banknote discriminator , in order to develop the discriminating algorithm, an analysis tool, a neural network, fuzzy inference, etc. for setting items for statistical analysis processing are used. In the case of an analysis tool, tuning ( There are many items to be adjusted), and there are problems that are difficult to automate. In the case of neural networks, automation is possible, but there are problems that require a large amount of calculation and time, and fuzzy inference. In this case, the items to be tuned may not be intuitively understood by humans, and may have problems such as failure to determine fuzzy inference, and are general and effective for various applications. The current situation is that there is nothing.
[0003]
In addition, as a method of identification, a normal correlation method or a method of identifying by using the Euclidean distance to determine the similarity may be considered, but the same type such as printing misalignment in the manufacturing stage of paper sheets and dirt in the distribution process, etc. It is difficult to raise the identification accuracy further than a certain level, and it is difficult to raise the accuracy in the identification of very similar print patterns due to printing misalignment. .
[0004]
[Problems to be solved by the invention]
It is an object of the present invention to provide a banknote recognition apparatus that characterizes an image by setting a weight to the image to be identified, enhances the identification effect, and enables effective identification.
[0005]
[Means for Solving the Problems]
The present invention relates to an identification standard created by weighting a pixel of a coordinate with a stable individual difference for an image of one denomination and weighting of a pixel of a coordinate having a large gradation difference from the image of another denomination. It is characterized by using data .
[0006]
[Operation and effect of the invention]
According to the present invention, a feature can be added to an image, and effective identification can be performed.
[0007]
【Example】
An embodiment of the present invention will be described below with reference to the drawings. The drawing shows a paper sheet identification device. In FIG. 1, a paper sheet, for example, a bill 10 is transported in a predetermined transport direction by transport rollers 11 appropriately driven and controlled, and transmitted from a light source 12 in the transport path. The sensor 13 receives the light and reads the transmission image of the two-dimensional print pattern. The above-described sensor 13 is formed in a line shape by a CCD array, for example, and is provided so as to sufficiently cover the width of the bill 10 to be conveyed.
[0008]
The printing pattern may be read by one-dimensional data, and the reading method may be configured to move the sensor 13. Further, the reading may be a reflection image instead of a transmission image.
[0009]
FIG. 2 shows the coordinate data (X-axis 0 to N, Y-axis 0 to M) of the read image. The data is a two-dimensional digital image, and each pixel ranges from 0 (dark) to 255 (bright). Take gradation value.
[0010]
FIG. 3 shows a control configuration diagram. The CPU 20 drives and controls the image reading unit 23 and the conveying unit 24 based on a program stored in the ROM 21, and the ROM 21 has a weighted template (identification reference data) for identification. ) And weight images (weight data).
[0011]
The RAM 22 stores data necessary for the operation, the image reading unit 23 takes in the image data read by the sensor 13, and the conveying unit 24 controls the driving of the conveying rollers 11.
Next, the identification processing of the paper sheet identification apparatus will be described with reference to the flowchart of FIG. The CPU 20 drives and controls the transport unit 24 to drive the transport rollers 11 in a predetermined transport direction. Further, the CPU 20 drives and controls the image reading unit 23 to turn on the light source 12, and the sensor 13 reads the transmitted image of the banknote 10. , Each gradation value (0 (dark) to 255 (bright)) at each position (1 ≦ x ≦ N, 1 ≦ y ≦ M) of the coordinates (X axis 0 to N, Y axis 0 to M) of this transmission image As input image data [Equation 1] (step n1).
[0012]
[Expression 1]
Then normalized by Equation 2] for grayscale image data [Number 1] inputted (Step n2).
[0013]
[Expression 2]
Next, the set weight is applied to each pixel of the input image (step n3), but this weight is set in advance for each class of the banknote 10 (for each denomination, one type of denomination, front / back and front / rear direction). The weight image (weight data) [Equation 3] is stored in the ROM 21.
[0014]
[Equation 3]
The ROM 21 also stores a weighted template (identification reference data) [Equation 4] obtained by renormalizing the normalized image data of each class of the banknote 10 with a set weight.
[0015]
[Expression 4]
The above-described weight setting in [Equation 3] is intended to identify the pixels (coordinates) that are important for identification by increasing the weight, and the setting method is as follows, for example.
As one of them, the weight of the pixel of the coordinate with stable individual differences is increased with respect to the image of the banknote 10 (four types of one denomination, front and back, front and rear direction).
[0016]
As the second, the weight of a pixel having a large difference in gradation value from the image of the banknote 10 of another class (denomination) is increased.
As the third, the weight of a pixel having a large difference in gradation value between classes of banknotes 10 that are difficult to classify (denomination classification) is increased.
[0017]
As the fourth, in a class (for example, a high-priced bill) having a high risk in the case of misidentification, a weight is given so that misidentification is difficult to occur.
[0018]
In the weight setting example described above, the method for creating a weight image is described as an example in Part 1 and Part 2.
An average image is created for each class from the sample images of each class (for each denomination) for creating a template, and further, a variance value in the sample images within each class is obtained, and weight images (weight data) [ Formula 5] is created.
[0019]
[Equation 5]
Further, the above-described weight image is applied to the normalized sample input image for each pixel and re-normalized to create a template image (identification reference data) [Equation 6].
The process of step n3 for applying the weight set to each pixel of the input image described above applies the weight image of each class (each denomination) of the banknote 10 to the normalized input image for each pixel, and normalizes again [Equation 7 ] [0020]
[Expression 7]
Then, a normal correlation value with the weighted template image of [Equation 4] is obtained by [Equation 8] for each class of banknotes 10 (calculation of similarity) (step n4).
[0021]
[Equation 8]
When the normal correlation value (similarity) NC is calculated, the value NC calculated for each class (each denomination) is [Equation 9] −1 ≦ NC ≦ 1
In this case, the closer the normal correlation value (similarity) NC is to “1”, the more the input image is judged to be similar to the corresponding class, and the class closest to “1” among the values NC obtained for each class. Ask for.
[0022]
Next, it is determined whether 1 / | 1-NC | of the class closest to “1” is equal to or greater than a specified value for determining that the corresponding class (step n6). It is determined that there is (step n7), and if it is the following, it is determined that identification is impossible (step n8).
[0023]
In addition, according to the above-described embodiment, as is apparent from the equation shown in [Equation 8], the similarity comparison based on the normal correlation value is based on the difference in the internal angle of the vector, but the similarity is compared based on the vector length. Similarity can also be determined by the Euclidean distance method.
[0024]
As described above, according to the above-described embodiment, it is possible to give a characteristic to an image by applying a weight to the template image that is identification reference data and the read image data, and the template image that is identification reference data. Can be effectively identified when the degree of similarity is obtained.
[0025]
Furthermore, since the weight image can be obtained by setting without repeating the simulation , the development period of the identification algorithm can be shortened.
[0026]
Furthermore, the calculation formula (weight) of the weighted image can be used in common even if the object is changed, and the versatility can be obtained when there is no need to change the basic concept.
Furthermore, by changing the calculation formula (weight) of the weight image, a weight image suitable for the application can be obtained, and the identification accuracy can be improved.
[0027]
If printing misalignment occurs in the bill 10 or other paper sheet manufacturing process, or if dirt is deposited during the circulation of the banknote 10 or other paper sheet, the gradation between the sample (identification reference data) Therefore, the position (coordinates) of printing misalignment and dirt is reduced in weight, and the influence on the identification is reduced. Even in the case of these banknotes 10 and other paper sheets, good identification can be performed.
[0028]
Further, in identifying a paper sheet having a very similar print pattern, the weight of pixels having different gradation values is increased, so that it can be accurately identified.
In the above-described embodiment, weight is applied to both the template image of the identification reference data and the read image data. However, good identification can be obtained even if the weight is applied only to the identification reference data side.
[0029]
In the correspondence between the configuration of the present invention and the above-described embodiment,
The image reading means of the present invention corresponds to the sensor 13 and the image reading unit 23 of the embodiment,
Similarly,
Memorize means corresponds to the weighted template (identification reference data) and the weighted image ROM21 storing the (weight data),
The calculation means and the identification determination means correspond to the CPU 20 that performs the calculation process and the determination process. However, the present invention can be applied based on the technical idea described in the scope of claims, and can be applied only to the configuration of the embodiment. It is not limited.
[Brief description of the drawings]
FIG. 1 is a schematic side view of a paper sheet identification apparatus.
FIG. 2 is a diagram showing coordinates of image data.
FIG. 3 is a configuration diagram of a control circuit of the paper sheet identification apparatus.
FIG. 4 is a flowchart of identification processing.
[Explanation of symbols]
10 ... bill 13 ... sensor 20 ... CPU
21 ... ROM
23 ... Image reading unit [Equation 1]

[Expression 2]

[Equation 3]

[Expression 4]

[Equation 5]

[Formula 6]

[Expression 7]

[Equation 8]

Claims (4)

Image reading means for reading an image of a banknote ;
Storage means you store identification reference data serving as a reference for identification of the bill,
Calculating means for calculating a similarity between the image and the identification reference data read,
An identification determination means for determining the identification result from the similarity ,
The identification reference data is created by weighting a pixel having a stable individual difference with respect to an image of one denomination and weighting a pixel having a large gradation difference from another image of the denomination. It is characterized by
Banknote recognition device. The weight is given by

It is characterized by using
The banknote identification device according to claim 1. A method for creating identification reference data used as a reference for identifying a bill by a bill recognition device,
Weighting for pixels of stable coordinates of individual differences in one denomination image;
The identification reference data is created by weighting pixels with coordinates having a large gradation difference from other denomination images.
How to create identification criteria data. The weight is given by

It is characterized by using
The method of creating identification reference data according to claim 3.

Images