KR20160107075A - Method for recoginizing universal bill - Google Patents

Abstract

The present invention relates to a method to recognize a universal bill, capable of determining a forgery or a real bill based on an image obtained by using the lighting of a back light unit (BLU) and capable of accurately recognizing the bill of all countries regardless of the rotation of an inserted bill. The method comprises: (a) a step of providing lighting to obtain the image of the inserted bill; (b) a step of obtaining the image of a bill to which light is provided; (c) a step of estimating a rotation angle of the obtained bill image; and (d) a step of extracting a key point image for recognition from the bill image obtained based on the estimated rotation angle, calculating a characteristic value by performing a ring projection for the extracted key point image, and recognizing the bill by using ANN for the calculated characteristic value. The method has a rapid processing speed and a high recognition rate regardless of the rotation of an input image.

Description

{METHOD FOR RECOGINIZING UNIVERSAL BILL}

More particularly, the present invention relates to a method and apparatus for discriminating counterfeit / pneumatic based on images obtained by using BLU (Back Light Unit) illumination, The present invention relates to a method for recognizing a world banknote.

The bill recognizing apparatus is a device for discriminating a denomination of inputted bills and discriminating counterfeit / pneumatic, and is applied to various terminal devices such as a vending machine, a coin exchanger, and an automatic teller machine.

The bill recognition apparatus scans an image of a bill to discriminate a denomination, and determines whether or not the bill is counterfeit and whether the bill is abnormal.

Various methods for bill recognition have been proposed. Among them, feature vectors of a bill are obtained by using a scale invariant feature transform (SIFT), which is a feature transformation well known in the field of object recognition, Is also proposed.

SIFT computes the difference of the Gaussian (DOG) in the scale space and finds the maximum and minimum pixels on two adjacent neighbors. These maximum or minimum pixels are immediately resolved as the size invariant feature points, and the size space has a structure as shown in FIG.

As shown in FIG. 2, the maximal and the minima mean that the DOG value is the largest or the DOG value is the smallest in the eight neighboring pixels and two neighboring sizes.

The image gradient orientation of each pixel is calculated in the peripheral region of the extracted feature point, and then these orientations are quantized by 36 angles, and the angle with the highest frequency is determined as the reference orientation of the feature point.

In the case of FIG. 3, a feature vector of 32dimension is obtained for each feature point. Generally, for a bill recognition, a feature vector of 128dimension is obtained for each feature point by using a histogram quantized at 8 angles with respect to a 4x4 region.

The acquired feature points are matched using the k-d tree. The k-d tree is a search tree for effectively locating points adjacent to a given point as input in a space with k dimensions. If the number of points composing the tree is N and n points are found within a certain range around the given point as the input, the time taken is O (sqrt (N) + n).

After constructing a k-d tree with the feature vectors obtained for the model images, we find the model feature vectors that are most similar to the respective feature vectors obtained for the input image. The similarity measure used here is the Euclidean distance, and it is judged that the matching is performed only when the Euclidean distance is smaller than 0.4 for the feature vectors equalized to the size 1.

4 is a flowchart of the nationality and census recognition algorithm in the recognition of a bill. FIG. 5 shows an example of the result of applying the SIFT Feature algorithm to the bill data base, which is the extraction result of the SIFT Feature for Hanwha, FIG. 6 is the extraction result of the SIFT Feature for the Euro, .

On the other hand, the result of recognition of the denomination regarding the matching result using the SIFT characteristic value is as follows. The result of matching the extracted SIFT Vector to the corresponding key point of the image on the data base by using K-D tree and ANN algorithm is applied to Hanwha 50,000 won, 500Euro and USD100 dollar bills.

8A is a result (middle column) and a final extraction result (right column) matched to the Hanwha DB for the input image (left column), and FIG. 8B is a result And probabilistic voting (50,000 won front and back).

Figures 9A and 9B show recognition results for the 500 EURO banknote, Figure 9A is a result (middle column) and final extraction result (right column) matched to USD DB for the input image (left column) Results recognized by probabilistic voting (500EURO front and back).

FIGS. 10A and 10B are recognition results of 100 USD bills, FIG. 10A is a result (middle row) and final extraction result (right column) matched to USD DB for the input image (left column) The results recognized by Probabilistic Voting (USD 100 front and back).

11 is a speed characteristic table obtained when recognizing paper money using the SIFT Feature algorithm using a PC having the specifications of INTEL PC i5 2,8 Ghz, 4 GRAM, and Win732Bit.

SIFT rotation, change of illumination, change of scale, and so on. However, the processing speed is 3.35 seconds as shown in Fig. 11, which is slow in practical application. Therefore, in order to recognize the banknotes quickly using the SIFT Feature algorithm, it is required to develop an algorithm that uses a processor having a performance of about 1 GHz of Embedded ARM (Cortex 8) and a high processing speed.

On the other hand, in order to solve the disadvantages of bill recognition using the SIFT Feature algorithm, there has been proposed a national recognition and denomination recognition algorithm using DFT.

The pixel data of the image is expressed in the form of a two-dimensional array, which is a discrete function having a form of f (x, y) having two variables x and y, and can be expressed as Equation .

Here, the Fourier transform equation of Equation (1) is implemented using Equation (2) below, which is divided into row and column units.

The results of implementing the DFT for Hanwha and Euro and USD are the same as the results in columns 1 and 2 of Figs. 8a to 10b.

The paper recognition result using DFT and cross correlation is as follows.

An optimal matching algorithm is found between the binarization spectrum obtained from the DB and the binarization spectrum of the input image using the linear cross correlation equation (Equation 2), and an algorithm for recognizing the nationality and the denomination has been proposed. Assuming that the input x (k) is a signal composed of L numbers and the signal y (k) is a signal on the DB consisting of M (M L), in this case, the linear cross correlation The value can be expressed as r _xy (i) as shown in Equation (3) below.

FIGS. 12 to 14 illustrate the results of application of CUREAL (Currency Recognition Algorithm using DFT) algorithm by applying DFT to Hanhwa, Euro and USD.

The results of the fourth column in each figure represent the results of the binarization spectrum of the third row in the x direction and y direction, respectively, and Sum of the Projected Amplitude of the PA (Projection Amplitude) along the respective axis.

Using these results, we obtain the correlation with the extracted PA of the input banknote image, extract the image with the minimum value, and recognize the nationality and the denomination.

Figures 15A to 15F show the results of calculating Descriptor for the CUREAL algorithm for KRW 50,000, EURO 500, and USD 100.

The result on the right side is the extracted Descriptor result, and the CUREAL Descriptor of the input banknote image is obtained by using the extracted descriptor, and the line mapped correlation between the value and the DB value is calculated to extract the best-mated banknote image and recognize the bill.

In addition, as a conventional bill recognition method, there is a method of recognizing bills through Template Matching (TM).

FIG. 16 is a diagram showing the contents of the TM, and the matched portion is shown on the right.

TM compares the template of the original image pre-stored in the memory with the template of the acquired image at the central position, and compares the similarity with the reference threshold value.

Then, the similarity degree of the extracted image is compared with a reference threshold value set as a criterion for discriminating the pneumoconiosis, and the bill is discriminated according to whether it is large or small. For example, the similarity degree of the extracted image is compared with a reference threshold value set as a criterion for discriminating the pneumoconiosis. If the similarity degree of the image is equal to or greater than the reference threshold value, the judgment is made as pneumatic, If it is smaller than the reference threshold value set for the hazard reference, it is determined to be counterfeit.

Here, the algorithm for judging the pneumoconiosis by comparing the similarity of the image with the reference threshold value is summarized by the following Equation (4).

Here, T (x, y) represents a template image for the original image, and I (x, y) represents an image template of the inputted bill.

Then, counterfeit and pneumoconiosis are judged with respect to the recognized nationality and denomination recognized by using the above-described SIFT and DFT. The image discriminates the pneumatic / counterfeit by using the image obtained from the transmission image and the infrared image through the UV illumination and the backlight unit.

A method of detecting a counterfeit through a security fiber image specifying value is shown in Figs. 17A to 17E. It is the case of using the gift image characteristic value among the results using the input image by irradiating UV light to the Hanwha 50,000 won ticket. TM is applied to the gift image detected by UV photograph to distinguish counterfeit / pneumatic. 17A is a UV channel image, FIG. 17B is a RED channel image data, FIG. 17C is a green channel image, FIG. 17D is a blue channel image, FIG. 17E is a result of applying a TM to a red channel It is a video.

18A to 18C show a method of detecting a counterfeit by using a transmission image specification value. TM is applied to the projected image obtained by using backlight illumination in Hanwha 50,000 won to distinguish counterfeit / pneumatic. Fig. 18A shows the result of the transmitted image spotted hill TM, Fig. 18B shows the TM result of the letter of the transmitted image, and Fig. 18C shows the TM result of the UV fluorescent solid part.

On the other hand, the following Patent Documents 1 and 2 disclose an example of a bill recognition apparatus according to the prior art.

Patent Document 1 discloses a method and apparatus for recognizing a banknote denomination recognizing a denomination as an output value by simultaneously using a function in a method of receiving a banknote image, processing the denomination, and recognizing the denomination.

Patent Document 2 discloses a method of checking the LED (Light Emitting Device) transmittance (hereinafter referred to as "CIS transmittance amount") in a CIS (Contact Image Sensor) by a predetermined white medium and 1) The image output value of the CIS is multiplied by the parameter value set by the CIS transmittance as another one of a predetermined value to obtain the corrected image, Discloses a configuration of a bill image recognition apparatus that obtains an image of a uniform bill by obtaining an output value.

Korean Patent Registration No. 10-0718728 (published on May 16, 2007) Korean Patent Registration No. 10-1014974 (issued on February 16, 2011)

However, the above-mentioned prior art has the following disadvantages.

For example, the bill recognition method using SIFT and TM is advantageous in that it is resistant to illumination change, rotation, and scale change, but it has a low recognition speed and is difficult to apply to a DSP-based embedded system.

In addition, the bill recognition method using FFT can not obtain the processing speed as much as applicable to the embedded system, but it is difficult to obtain the characteristics for recognition from the specific value obtained by the FFT, and the threshold value differs according to the type of the bank. There is a disadvantage that it is necessary to designate a value, and if there is a rotation in the input part of the bill, an error occurs in recognition.

In addition, the conventional technique disclosed in the patent document also takes a long time to recognize the bill, and there is a disadvantage that an error occurs in the recognition of the bill.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method and apparatus for discriminating counterfeit / pneumatic based on an image obtained by using BLU (Back Light Unit) illumination, The present invention provides a method of recognizing banknotes of various countries around the world.

Another object of the present invention is to provide a method of recognizing a world banknote which can speed up processing speed and accurately recognize a bill even in a rotated image.

According to an aspect of the present invention, there is provided a method of recognizing a world banknote according to the present invention includes the steps of: (a) providing illumination for acquiring an image of an inputted bill; (b) obtaining an image of the bill provided with the illumination; (c) estimating a rotation angle to find a key point in the obtained paper money image; (d) extracting a key point image for recognition from a banknote image acquired based on the estimated rotation angle, performing a ring projection on the extracted key point image to calculate characteristic values, and calculating ANN And recognizing the banknote using the banknotes.

In the step (a), the illumination is provided using at least one of a white LED, an infrared ray (IR), an ultraviolet ray (UV), and a red LED.

In the step (b), the bill image according to the change of illumination is acquired through the bill scanning apparatus.

In the step (b), a transmission image or an infrared image using illumination is acquired as a banknote image.

The step (c) includes: (c1) binarizing the image acquired in the step (b); (c2) expanding or contracting the binarized image; (c3) extracting an outline from the inflated or contracted banknote image; (c4) estimating the rotation angle of the bill inserted based on the outline extracted in the step (c3).

In the step (d), the obtained key point image is converted into a one-dimensional image through ring projection, and the result is normalized to acquire characteristic values of the input banknote image.

The step (d) constitutes a kd tree, and calculates the Euclidean distance between two points between the feature points in the multidimensional space of the kd tree to obtain the smallest index as a recognition result.

According to the present invention, it is possible to speed up the processing speed of the bill recognition, to improve the accuracy of bill recognition, and to accurately recognize the bill even in a rotated image.

FIG. 1 is an exemplary diagram illustrating generation of a size space of a general SIFT,
FIG. 2 is a diagram showing a maximum / minimum detection example of a DOG image,
3 is a diagram illustrating a method of detecting a size invariant feature point,
FIG. 4 is a flowchart of a process of an existing object recognition algorithm,
FIG. 5 is a diagram illustrating a SIFT feature extraction result for Hanhwa,
6 is a diagram showing a result of SIFT feature extraction for the euro,
FIG. 7 also shows the SIFT feature extraction result for USD,
8A is an illustration of recognition results for 50,000 won notes,
FIG. 8B shows the result of recognition of the 50,000 won note by the ANN and the probabilistic voting,
9A is an exemplary recognition result for a 500-euro banknote,
FIG. 9B shows the results of recognition of the 500-euro banknote by ANN and probabilistic voting,
FIG. 10A is an exemplary recognition result of a 100 USD bill, FIG.
FIG. 10B shows the results of recognition of 100 USD bills by ANN and probabilistic voting,
11 shows a result table of SIFT matching,
FIGS. 12 to 14 show the results of applying the DFT to the Hanwha, Euro, USD, and the CUREAL (Currency Recognition Algorithm using DFT)
Figures 15A-15F illustrate CUREAL Descriptor results for Hanhwa, USD and Euro,
Figure 16 is an illustration of an existing TM algorithm,
Figs. 17A to 17E are diagrams showing an example of a method of detecting a counterfeit through conventional gift detection,
18A to 18C illustrate the result of recognizing the pneumoconiosis by applying the image by the backlight unit to the TM,
19 is a schematic configuration diagram of a world bill recognition apparatus to which the present invention is applied;
FIG. 20 is a flowchart illustrating a method for recognizing a world banknote according to a preferred embodiment of the present invention. FIG.
21 is a view of a metal contact PCB of UV light according to the present invention,
Fig. 22 is a block diagram of a white LED illumination,
Figs. 23A to 23G are photographs of manufacture and assembly of the BLU,
FIG. 24 is a flowchart related to extraction of a rotation angle of an input image,
25A to 25C are views showing examples of implementation of a ring projection image,
26 is an ANN data structure diagram using a kd-tree,
FIGS. 27A and 27B are diagrams showing feature extraction results using ring projections of rupiah,
28 is a diagram showing recognition results based on ANN using characteristic values using ring projections of rupiah,
FIG. 29 is a view showing a result of recognizing banknotes of Indian bills rotated at an arbitrary position based on ANN based on feature values using Ring Projection,
30A and 30B are diagrams showing ring projection characteristic values of USD,
FIGS. 31A and 31B are diagrams showing a banknote counterfeit / pneumatic template characteristic value (banknote through transmitted light) using the ring projection of the Indian banknote (rupee)
32 is a table showing the measurement speed according to the recognition method of the Indian banknote (rupee)
FIG. 33A is a graph of execution speed according to the recognition method of the Indian bill (rupee), 33b is the execution speed graph by the ring projection of the Hanwha bills, and FIG. 33C is the execution speed graph by the ring projection of USD bills.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A method of recognizing a world banknote according to a preferred embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

FIG. 19 is a schematic block diagram of a banknote recognition apparatus to which a banknote recognition system according to a preferred embodiment of the present invention is applied. The banknote recognition system includes a transfer module 10, a lighting apparatus 20, an image acquisition unit 30, A display unit 50, a discharge module 60, and a deposit module 70. [

The transfer module 10 transfers the bill to a position where image scanning is possible. The illumination device 20 provides a BLU to acquire a transmission image or an infrared image.

The image acquiring unit 30 acquires a bill image by scanning the bill in a state where illumination is provided on the bill.

The banknote recognition unit 40 extracts the feature value image around the central position through ring-projection, and applies the ANN algorithm to discriminate the counterfeit banknotes or the pneumatic banknotes do.

The display unit 50 displays the result recognized by the bill recognition unit 50 on the screen.

When the present invention is applied to the banknote processing apparatus, the discharging module 60 discharges the inputted banknotes when it is determined that the banknotes are abnormal banknotes as a result of the recognition of the banknotes. The banknote depositing module 70, when the present invention is applied to a banknote processing apparatus , And when it is recognized as a result of recognition of banknotes, the banknote is transferred to the depositing apparatus.

FIG. 20 is a flowchart illustrating a method for recognizing a world currency bill according to a preferred embodiment of the present invention, wherein S represents a step.

A method for recognizing a world banknote according to the present invention includes the steps of: (a) providing illumination for acquiring an image of a bill to be input (S101 to S102); (b) obtaining an image of the bill provided with the illumination (S103); (c) estimating a rotation angle of the obtained banknote image (S104 to S107); (d) extracting a key point image for recognition from a banknote image acquired based on the estimated rotation angle, performing a ring projection on the extracted key point image to calculate characteristic values, and calculating ANN (S108 to S109).

The method for recognizing the world currency bill according to the preferred embodiment of the present invention will be described in detail as follows.

First, when the bill to be recognized is inserted as in step S101, the bill inserted in the transfer module 10 is transferred to the image acquisition position.

When the inserted bill is transferred to the image acquisition position, the lighting apparatus 20 provides illumination to the inputted bill in step S102. In bill recognition, the role of lighting is very important. The lighting is BLU (Back Light Unit), which can use illumination for backlight LED, infrared (IR) illumination and ultraviolet (UV) illumination. Backlight LEDs can use white LEDs or red LEDs.

21 is a view of a metal contact PCB of UV light for providing UV illumination in the present invention. As a specification for such UV illumination, the current is 800 mA, the applied voltage is 3.9 VDC, the wavelength is 375 nm, and the maker is Seoul Optpdevice.

In addition, when white LED illumination is used, a bar-shaped LED is used, and FIG. 22 shows a jig of Bar illumination. With this white LED lighting as a position specification, the current is 20 mA, the applied voltage is 3.2 VDC, and the Maker is L Fine.

23A to 23G are photographs of fabrication and assembly of a backlight unit applied to the present invention. FIG. 23A shows the appearance of the BLU applied in the present invention, FIG. 23B shows the scattered light of the BLU after the power supply, FIG. 23E shows the banknote of 500EURO placed on the BLU, FIG. 23F shows the banknote of 500EURO It shows captured image using camera. 23G is a view showing the overall appearance of an AIIS (Automated Illumination and Imaging System) combined with a camera and infrared (IR) illumination.

Next, in step S103, the system of FIG. 23G is used to capture and recognize a bill image according to a change in illumination. The banknote scanning device can use the PCSU100C banknote scanning device developed by Blue Technology Co., Ltd. Here, it is preferable that the banknote image is acquired through a banknote scanning apparatus in accordance with a change in illumination. It is further preferable to acquire a transmission image or an infrared image using illumination as a bill image.

The PCSU 100C banknote scanning device includes a scanning module for scanning reflection images and transmission images of the top and bottom sides of a banknote, a first sensing module for sensing magnetic characteristic information provided on the banknote, and a second sensing module for sensing ultraviolet characteristic information provided on the banknote . The scanning module may include first and second contact image sensors (hereinafter, referred to as 'CIS') installed at the upper and lower portions of the conveying path of the bank notes.

The first CIS scans the reflection image of the upper surface of the banknote by the reflected light source and the transmission image transmitted through the banknote by the transmission light source provided below the banknote conveying path. Here, the reflected light source includes a plurality of visible light LEDs (light emitting diodes) for irradiating light in a visible light region, for example, red and green light, and an infrared LED for irradiating light in an infrared ray wavelength . It is preferable that these reflective light sources are installed at the upper and lower portions of the banknote conveying path, respectively. The transmission light source may include an infrared LED installed at a position corresponding to the first CIS below the banknote conveying path to transmit the banknote and to irradiate light of an infrared wavelength. Accordingly, the first CIS can scan the reflection image reflected from the upper surface of the banknote by the reflection light source and the transmission image transmitted from the lower part of the banknote by the transmission light source. Accordingly, the present invention can discriminate whether a bill is counterfeit or not by using reflection images and transmission images of a unique infrared ray reaction element provided on a bill showing different characteristics from the reflection image of the visible light region together with the reflection image of the visible light region.

In addition, the second CIS scans the reflection image of the banknote bottom surface. The first sensing module senses a magnetic component provided on the banknote. For example, the first sensing module may include first to third magnetoresistive sensors disposed in parallel on a line extending in a direction perpendicular to the conveying path of the banknote. The first to third magnetoresistive sensors are installed at preset intervals to sense the left side, the center and the right side of the banknote, respectively, so that the dead zone can be removed by sensing the entire width of the banknote. The second sensing module includes first and second ultraviolet ray sensors installed on the upper and lower sides of the banknote conveying path to sense ultraviolet characteristics, for example, fluorescence components of bills, provided on the upper and lower surfaces of the banknote, respectively .

Next, the position of the leading edge and the center point are searched for performing the ring projection. Even if the dead zone is removed by sensing the entire width of the banknote, positional change and rotation may occur during transfer. To do this, pre-processing is used to find the position and center of the paper currency through binarization, expansion / contraction, and outline detection of the image, and calculate the rotation angle.

That is, in step S104, the image obtained through the banknote scanning device is binarized, the binarized image is expanded in step S105, and the outline is extracted from the expanded image in step S106. Then, the rotation angle [theta] of the bill inserted based on the outline extracted in step S107 is estimated. At this time, the rotation angle estimation uses Equation (5) below. FIG. 24 is a sequence of operations relating to the extraction of the rotation angle of the input image.

Next, a ring projection is performed on the image obtained through the bill scanning apparatus as described above to convert the two-dimensional image into a one-dimensional image.

Next, Fig. 25A shows the input banknote images, Fig. 25B shows the image region projected on the ring, and Fig. 25C shows the ring projected image.

Here, the algorithm for the ring projection for converting the two-dimensional image into the one-dimensional image can be summarized as Equation (6) below.

Next, in step S108, the ring projection feature value acquired based on the estimated rotation angle is recognized by using an approximate nearest neighbor (ANN).

In order to recognize banknotes using the ANN, a k-d tree is constructed as shown in FIG. Then, the Euclidean distance between two points between the feature points in the multi-dimensional space of the k-d tree is obtained as a recognition result by using Equation (7) below. In Equation (8), p is a ring projection characteristic value, and q is a ring projection characteristic value of a database loaded in a k-d tree. Here, the specified value is a value obtained by normalizing the result of the ring projection.

According to the present invention, bill recognition is possible without being set to a Thresold value which is robust against rotation. Therefore, while the bill recognition time is shortened, the accuracy of bill recognition can be improved.

27A and 27B show feature extraction results using a ring projection of rupiah.

28 shows ANN-based recognition results using characteristic values using Ring Projection of rupiah. In particular, it can be seen that recognition is good even when there is rotation.

FIG. 29 shows a result of recognizing banknotes of the Indian bills rotated at an arbitrary position based on ANN based on feature values using Ring Projection.

30A and 30B show ring projection characteristic values of USD.

Figs. 31A and 31B show a banknote counterfeiting / pneumatic template characteristic value (banknote through transmitted light) using the ring projection of the Indian banknote (rupee).

32 is a table showing the measurement speed according to the recognition method of the Indian banknote (rupee).

FIG. 33A is a graph of execution speed according to the recognition method of the Indian banknote (rupee), 33b is the execution speed graph by the ring projection of the Hanwha banknote, and FIG. 33C is the execution speed graph by ring projection of the USD banknote.

It can be seen that the bill recognition speed is improved when the bill is recognized by applying the present invention to the existing technology.

Although the invention made by the present inventors has been described concretely with reference to the above embodiments, the present invention is not limited to the above embodiments, and it goes without saying that various changes can be made without departing from the gist of the present invention.

According to the present invention, the present invention is effectively applied to a bill recognition device technology that is robust to a rotation angle and requires an improvement in recognition speed.

10: Feed module
20: Lighting device
30: Image acquisition unit
40:
50:

Claims (7)

As a method of recognizing the world banknotes,
(a) providing illumination for acquiring an image of a banknote to be input;
(b) obtaining an image of the bill provided with the illumination;
(c) estimating a rotation angle of the obtained banknote image; And
(d) extracting a key point image for recognition from a banknote image acquired based on the estimated rotation angle, performing a ring projection on the extracted key point image to calculate characteristic values, and calculating ANN And recognizing the banknote using the banknote identification information.
The method according to claim 1, wherein the step (a) provides illumination using at least one of a white LED, an infrared ray (IR), an ultraviolet ray (UV), and a red LED.
The method according to claim 1, wherein the step (b) acquires a bill image according to a change in illumination through a bill scanning apparatus.
The method of claim 1 or 3, wherein the step (b) acquires a transmission image or an infrared image using illumination as a banknote image.
The method of claim 1, wherein the step (c) comprises: (c1) binarizing the image acquired in step (b); (c2) expanding the binarized image; (c3) extracting an outline from the inflated banknote image; (c4) estimating the rotation angle of the banknote based on the outline extracted in the step (c3).
[2] The method of claim 1, wherein the step (d) comprises converting the acquired key point image into a one-dimensional image through ring projection, and normalizing the result to acquire the feature value of the input banknote image How To Recognize World Banknotes.
[7] The method of claim 6, wherein the step (d) comprises constructing a kd tree and calculating the Euclidean distance between two points between the feature points in the multidimensional space of the kd tree according to the following equation, How To Recognize World Banknotes.
<Formula>

Where p is the ring projection feature and q is the ring projection feature value of the database loaded in the kd tree.

Images