JPH07101464B2 - Pattern recognition device for coins - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an apparatus capable of recognizing and selecting a pattern on a surface of a circular object such as a coin.

(Prior Art) A conventionally known pattern recognition apparatus recognizes a pattern by inputting an image of a subject and then processing the image, and there are the following types depending on an image processing method.

A method in which the input image data is binarized, the area feature, the center of gravity, the perimeter, and other geometrical feature quantities are determined, and this is compared with a reference value.

A method of recognizing an input pattern by pattern matching of a binary image or a multivalued image. For example, a method in which an input character pattern of an object and a standard character pattern prepared in advance are superimposed and the input character pattern is recognized based on the degree of similarity.

A method of performing density projection on XY coordinates or r-θ coordinates of an input image and comparing the characteristics of the intensity distribution. For example, 9th
The figure shows an example of recognizing a character pattern by performing density projection on X-Y coordinates. The density projection 21 of the input character pattern 20 on the X coordinate axis and the density projection 22 on the Y coordinate axis are obtained, and The pattern is recognized by using the waveform of the intensity distribution as a feature. The method of recognizing the pattern by obtaining the density projection at the r-θ coordinate has the same principle, and is different only in that the density projection in the rotation direction is obtained.

(Problems to be Solved by the Invention) The above-described conventional pattern recognition device uses any method,
The positional relationship in the rotation direction is not constant, which is not rational for recognizing an object that is free in the rotation direction, such as a hard object. That is, when trying to recognize coins or the like by the above method, after binarization, it is called labeling processing or area division,
For example, when multiple characters are input at one time, the area is divided into each character, and geometrical characteristics such as center coordinates, area, outer diameter, and perimeter of each area are calculated. However, there is a problem in that it takes a long time to perform pattern recognition because it requires complicated processing.

When trying to recognize a coin or the like by the above method, it is necessary to normalize the position including the direction of the coin or the like, and there is a problem that it takes time to find the direction.

According to the method of density projection on the r-θ coordinate among the above methods, it is possible to recognize a coin or the like. According to this method, the center of the coin or the like is obtained from the input image data,
Next, it is necessary to perform density projection to obtain the correlation with a reference coin or the like for one rotation, which causes a problem that pattern recognition takes a long time.

As described above, in short, the conventional pattern recognition device has a problem that the image processing time for recognizing a pattern such as a coin becomes long.

The present invention has been made in order to solve the problems of the prior art, and can perform pattern recognition of a target object that is free to rotate a coin or the like by high-speed processing, and further,
An object of the present invention is to provide a pattern recognition device for coins or the like that can keep the recognition rate high.

(Means for Solving the Problem) The present invention is an improvement of the method of density projection on the r-θ coordinate, and includes an image input unit for inputting a pattern such as coins,
An image memory unit that processes and stores the input pattern of coins, etc. into a multi-valued image, and a ring formation that determines the center of the pattern of coins, etc. stored in the image memory unit and forms a plurality of rings concentric with this center And a ring data capturing unit that captures data from the image memory unit along a plurality of formed rings and stores the data,
Fourier transform calculation unit for Fourier transforming the data along the plurality of rings stored in the ring data capturing unit, and data along the plurality of Fourier transformed data along the plurality of rings and coins stored in advance by Fourier transform And a comparison unit for comparing the reference pattern data.

(Operation) The brightness of the image data stored in the ring data capturing unit along a plurality of concentric rings changes according to the pattern of the portion where the rings are located. When this image data is Fourier transformed by the Fourier transform calculation unit, the frequency intensity of the image data is obtained. Since this frequency intensity differs depending on the pattern of the coin or the like as the object, it is possible to perform pattern recognition by comparing it with a reference pattern which is Fourier-transformed and stored in advance. Further, the frequency intensity is irrelevant to the rotational position of the coin or the like as the object.

(Embodiment) An embodiment of a pattern recognition device for coins and the like according to the present invention will be described below with reference to FIGS. 1 to 8.

In FIG. 1, the image input unit 1 is composed of a television camera or the like, and an image signal obtained by capturing an image of a pattern of coins or the like is captured. The image input unit 1 uses a one-dimensional sensor or a two-dimensional sensor. In any case, it can be processed as a two-dimensional image. Therefore, in the case of a one-dimensional sensor, one screen is input by moving it relative to the object. The image signal captured by the image input unit 1 is converted into a multi-valued digital image signal by the A / D conversion unit 2 and input to the image processing unit 3.

The image processing unit 3 includes an image memory unit 4, a ring formation unit 5, a ring data acquisition unit 6, a fast Fourier transform calculation unit (hereinafter referred to as “FFT calculation unit”) 7, a reference pattern storage unit 8 and a comparison unit 9.

The image memory unit 4 stores the multivalued digital image signal. The ring forming unit 5 determines the center of the pattern of coins or the like stored in the image memory unit 4 and forms a plurality of rings concentric with the center.

The ring data fetching unit 6 is the ring forming unit 5
The data is fetched from the image memory unit 4 along the ring formed in 1 and stored.

The FFT calculation unit 7 performs a fast Fourier transform on the data stored in the ring data acquisition unit 6 along the plurality of rings.

The reference pattern storage unit 8 stores reference pattern data such as coins in a Fourier-transformed form in advance.

The comparison unit 9 compares the data Fourier-transformed by the FFT calculation unit 7 with the reference pattern data stored in the reference pattern storage unit 8 and determines whether the data such as coins is similar to each other. Recognize the pattern and output the result.

Next, the details of the pattern recognition operation according to the above embodiment will be described in the second
The description will be made with reference to FIGS.

First, an image signal of a coin is input from the image input unit 1. FIG. 3 (a) shows an example of the image 10 of a coin. The input image signal is converted by the A / D conversion unit 2 into a multivalued, for example, 256-gradation digital image signal, which is stored in the image memory unit 4.

Next, the ring forming section 5 calculates the image signal to obtain the center position. Here, for example, a plurality of points are taken on the edge of the image signal, and the least square method or the like is used to obtain the points as accurately as possible from the plurality of points. FIG. 3B shows a state in which the center point O is obtained from the edge 11 of the image signal as described above. The obtained central point O of the image signal is stored in the image memory unit 4. Further, in the ring forming portion 5, the obtained center point O
Forming a plurality of rings concentric with each other. Fig. 3 (c)
Shows an example of the plurality of rings 13, 14, 15 obtained as described above.

Next, the ring data fetching unit 6 fetches the image data from the image memory unit 4 along the plurality of rings and stores the image data for each ring. The image data taken along each ring becomes data whose density changes according to the pattern along each ring.
The left half of FIG. 3 (d) shows various examples of changes in density along the ring, and the frequency components included differ depending on the change in the pattern along each ring.

Next, the FFT calculation unit 7 performs FFT calculation for each data along each ring. By performing the FFT calculation, the frequency intensity of each ring-shaped data can be obtained. Frequency intensity is the change in concentration during one rotation of the ring, 1Hz, 2Hz ...
When expressed in terms of frequency, it means the frequency at which the same frequency component is included. The right half of FIG. 3 (d) shows an example of data obtained by FFT calculation of the ring-shaped density change data, where the horizontal axis is frequency and the vertical axis is amplitude (intensity). As can be seen from the right half of FIG. 3 (d), when the frequency component of the ring-shaped density is relatively small,
The FFT amplitude is large in the low frequency range and is large in the high frequency range when the frequency component of the ring-shaped density is relatively large. In other words, the FFT amplitude waveform will change depending on the pattern along the ring.

Therefore, next, in the comparison unit 9, the FFT amplitude data obtained for each ring and the FFT of the reference pattern stored in the reference pattern storage unit 8 in advance by Fourier transform are stored.
The amplitude data is compared to recognize which coin or the like the coin or the like as the object corresponds to by the evaluation amount. Here, the evaluation amount Sk is However, Is the ring number, Is the frequency, Wij is the weight, Fij is the FFT amplitude of the input image, R
ijk is represented by the FFT amplitude of the reference coin. The weight Wij weights the portion representing the characteristic pattern of the object as a large value. In the above equation, it is determined that the smaller the evaluation amount Sk, the more similar to the reference pattern. The pattern of coins or the like can be recognized as described above.

In the pattern recognition by the above series of image processing, FFT
The calculation will be further described by taking the case of a 100-yen coin as an example. As shown in FIG. 4, since the pattern 10 on the back surface of the 100-yen coin is a relatively simple pattern, a ring 14 having a predetermined radius is set on the pattern 10 and the brightness level of the image signal along the ring 14 is set. When the key data is obtained, as shown in FIG. 4 (b), the ring 14 makes one rotation and the rotation angle θ becomes 0 to 2
The change period of the brightness gradation until reaching π is relatively long, and the frequency component is relatively low. Therefore, when this gradation data is subjected to FFT processing, as shown in FIG. 4 (c), the amplitude (intensity) is large in a low frequency region of 1 Hz or 2 Hz, and the amplitude (intensity) in a relatively high frequency region ( Strength)
Is small.

On the other hand, as shown in FIG. 5, since the pattern 10A on the surface of the 100-yen coin is a complicated pattern, a ring 14 having a predetermined radius is set on the pattern 10A and the brightness level of the image signal along the ring 14 is set. When the key data is obtained, as shown in FIG. 5 (b), the ring 14 makes one rotation and the rotation angle θ becomes 0 to 2π.
The change cycle of the brightness gradation is short, and the frequency component is relatively high. Therefore, when this gradation data is subjected to FFT processing, as shown in FIG. 5 (c), a peak is seen in the amplitude (intensity) at a relatively high frequency of 5 Hz, which is a characteristic feature.

The amplitude (intensity) for each frequency in FIG. 4 (c) and FIG. 5 (c) represents the FFT amplitude Fij of the input coin or the FFT amplitude Rijk of the reference coin in the formula for obtaining the evaluation amount Sk described above. , FFT amplitude Fij of input coin and FFT amplitude R of reference coin
By evaluating the difference from ijk, it is possible to recognize the type of coin and whether the coin is front or back.

According to the embodiment described above, a concentric ring is set for a circular object such as a coin, the input image data is taken along this ring, and the ring-shaped data is further calculated by FFT calculation. Since the frequency intensity is calculated and the pattern recognition is performed by comparing this with the reference pattern data, the frequency intensity finally used for the pattern recognition becomes data irrelevant to the rotation direction of the object, and is free in the rotation direction. Even for objects such as coins, pattern recognition can be performed in a short time with a relatively simple process, and the recognition rate can be kept high. Further, since the image data is taken in along the concentric ring, the value at the rotational position 0 becomes the value at the rotational position 2π, there is no influence of the data folding, and the window is set like a certain pattern recognition device. It is not necessary to perform the process of matching the value of 0 with the value of 2π.

Furthermore, since multiple concentric rings are set and image data is captured along each ring and processed, it is possible to more accurately recognize the characteristics of the pattern of the target object and improve the recognition rate. Can be improved. Therefore, it is desirable to set each ring in a portion that can best capture the characteristics of the pattern such as coins.

It should be noted that the coins, etc. as the object have different reflection intensities depending on whether they are new or used ones, or due to the difference in the material due to the difference in the type of coins etc. The signal is different,
This may reduce the reliability of pattern recognition. Therefore, other data, for example, the sum of the brightness of each ring is compared, and the reliability is secured by normalizing the data according to the result. FIG. 6 shows an example of data normalization, in which the intensity after FFT calculation is corrected. The curve a shows the case of a new coin, and the curve b shows the case of a circulating coin. The strength of circulating coins at 1 Hz is converted to the strength of new coins at 1 Hz, the strength of circulating coins at other frequencies is shifted according to this converted value, and this is used for evaluation.

Even if the width of the ring to be formed is one pixel of the image input unit, the reliability of pattern recognition can be improved by forming a plurality of rings as described above. By doing so, the reliability of pattern recognition can be improved even if there is only one ring. 7th
Figures and 8 show examples thereof.

In FIG. 7, the width w of the ring 17 formed concentrically with respect to the input image 10 such as coins is set to 3 pixels. FIG. 7C is a development view of the ring 17 set to have a width of 3 pixels, and three pixel rows are arranged from one row to n rows per one row. Therefore, as shown in FIG. 7 (d), the sum of the intensities of the pixels in each column is calculated and used for the FFT calculation. Here, when it is assumed that the intensity signals a ₁ , a ₂ , and a ₃ are obtained from the three pixels forming each pixel row as shown in FIG. 8, it is preferable to weight each pixel.
For example, if the weights of a ₁ and a ₃ are 1 and the weight of a ₂ is 2, then the intensity A of one column is A = a ₁ + 2a ₂ + a ₃ . By performing such a process, the feature of the pattern of the object can be made strong, and even if the center of the ring is deviated, the recognition rate can be increased.

The FFT calculation may not be performed on the pixels of the entire formed ring, but a predetermined number of pixels may be extracted and only the extracted pixels may be calculated. In this case, the number of data to be extracted and used for FFT calculation is generally 2 ^N (N is an integer).

(Effect of the Invention) According to the present invention, a concentric ring is set for a circular object such as a coin, the input image data is taken along this ring, and this is further Fourier transformed to form a ring. Since the frequency intensity of the data is obtained and the pattern is recognized by comparing it with the reference pattern data which has been Fourier transformed in advance, the frequency intensity finally used for the pattern recognition is independent of the rotation direction of the object. It becomes data, even for objects such as coins that can freely rotate,
Pattern recognition can be performed in a short time with a relatively simple process, and the recognition rate can be kept high.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a pattern recognition device for coins and the like according to the present invention, FIG. 2 is a flow chart showing the operation of the same embodiment as above, and FIG. 3 is a concept showing a signal processing process according to the above embodiment. FIGS. 4 and 5 are explanatory views showing the principle of recognizing the back side of coins according to the above-described embodiment stepwise, FIG. 5 is an explanatory view showing the principle of recognizing the back side of coins according to the embodiment above, and FIG. A diagram showing the concept of normalization performed in response to changes in the condition of the object, FIG. 7 is an explanatory diagram showing stepwise another example of pattern recognition processing of coins, etc., and FIG. 8 is the same as above. 9 is a cross-sectional view showing a state of one row of pixels used for
The figure is a conceptual diagram for explaining one of the conventional pattern recognition methods. 1 ... Image input unit, 4 ... Image memory unit, 5 ... Ring forming unit, 6 ... Ring data importing unit, 7 ... Fourier transform calculation unit, 9 ... Comparison unit, 10, 10A ... Pattern, 1
3,14,15 …… Ring, O …… Center of the pattern.

Claims (1)

[Claims] 1. An image input unit for inputting a pattern of coins or the like, an image memory unit for processing the inputted pattern of coins or the like into a multi-valued image and storing it, and a center of the pattern of coins or the like stored in the image memory unit. A ring forming unit that forms a plurality of rings concentric with the center, and a ring data capturing unit that captures data from the image memory unit along the plurality of formed rings and stores the data. A Fourier transform calculation unit that performs a Fourier transform on the data along the plurality of rings stored in the data acquisition unit, and a reference to coins and the like that have been Fourier transformed and stored in advance along with the data along the plurality of rings that have been Fourier transformed. A pattern recognition device for a coin or the like, comprising a comparison unit for comparing pattern data.