JPH02204880A - Picture recognizing device - Google Patents

Abstract

PURPOSE:To easily decides a subject part of a subject matter by applying the Fourier transformation to the picture data after adding and compressing them for each line and obtaining a power spectrum signal in case the picture data on the subject matter is static. CONSTITUTION:An image pickup means 10 is prepared to photograph a subject matter and to acquire the picture signals, together with an addition/compression means 11 which adds and compresses the picture data, a power spectrum calculation means 12 which calculates the power spectrum of the picture data, a deciding circuit 13 which compares the power spectrum signal with the reference value, and a data output means 14 which displays the deciding result of the circuit 13. The picture data are added together in the single direction via the means 11 in case the picture data are static. The added picture data undergo the Fourier transformation for calculation of the power spectrum. This calculated spectrum is compared with the reference value for recognition of pictures. Therefore the picture data are treated in a space frequency area. Thus it is possible to surely grasp a subject part even in such a case where no difference of density is substantially produced between the subject matter and the subject part.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention relates to a device for recognizing a desired image signal, and in particular, to a device for recognizing a desired image signal, and in particular for extracting a desired signal from an acquired still object image or a moving object image. Related to a device for recognizing objects (prior art) Generally, this type of device is equipped with an imaging section such as a COD camera that images a target object, and processes the image output obtained by the imaging section before performing gradation conversion, etc. After the processing, a binarization process is performed at a predetermined level.

The binarized output from this process corresponds to, for example, characters written on the target object or scratches present on the target object. This 2
The digitized output is compared with pre-stored reference data,
It is determined whether the binarized output corresponds to a desired image signal, and the determination result is output to the data output means.

However, with such devices, when the target object is a metal plate with characters printed on it, subtle shading may occur in the image due to the finish of the metal plate surface, the arrangement of the imaging unit and the lighting, etc. By doing this, when the density profile on the line that crosses the print is binarized at a predetermined level, the print profile and the background metal plate profile have approximately the same brightness (density), so the print alone This poses a problem in that it becomes difficult to extract and cannot be recognized. This situation is the same when the target to be extracted is not limited to printed characters, but also when various types of printed matter or various target parts such as scratches have approximately the same brightness as the target object.

Furthermore, if this method is applied to a case where a target object is being transported, such as on a production line in a manufacturing factory, the image of the target part in the input image will deteriorate, so 2. If the image is binarized by the digitization means, there is also the problem that it becomes difficult to recognize the target part.

(Problems to be Solved by the Invention) As described above, in conventional image recognition devices, when the density difference between the target object and the target part disappears, or when the target object is being transported, the input This poses a problem in that it becomes difficult to extract the target object from the image, making reliable recognition impossible.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems, and to provide an image recognition device that can more reliably extract a desired image.

Another object of the present invention is to provide a device for recognizing images that can reliably extract a desired image regardless of whether the target object is stationary or moving.

[Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the image recognition device of the present invention includes an imaging means for imaging a target object and obtaining an image signal, and an additive compression method for adding image data. A power spectrum calculation means for calculating a power spectrum of image data, a determination circuit for comparing the power spectrum signal with a reference value, and a data output means for displaying a determination result of the determination circuit.

Further, when the captured image is a still image, the addition compression means adds the image signals line by line on the screen, and then the power spectrum calculation means calculates a power spectrum signal and outputs it to the determination circuit.

When the captured image is a moving image, the power spectrum calculation means calculates the power spectrum for each line of the image signal, and then the addition compression means adds these power spectrum signals line by line, and the added output is determined by the judgment circuit. Output to.

(Function) According to the above configuration, when the image data is stationary, it is added in one direction via the addition compression means, then Fourier transformed to calculate the power spectrum, and this calculated value and the reference value are The images are compared and recognized.

Therefore, since the image data is handled in the spatial frequency domain, the target part can be reliably grasped even when there is almost no density difference between the target object and the target part.

In addition, if the image data is moving, it is possible to first perform Fourier transform on the image data, treat the image as a spatial frequency, and add and compress this to emphasize the signal of the target area. You can get it.

(Example) Hereinafter, one example of the present invention will be described with reference to the drawings.

FIG. 1 shows an image recognition apparatus when the target object for image recognition is stationary, and the target object will be explained. In FIG. 1, the image input means 10 is an imaging device such as a CCD camera, and an image signal obtained by imaging a target object is sent to an A/D converter.
The data bus 10a is converted into a digital grayscale image.
It is output to the addition compression means 11 via. Additive compression means 1
1 is composed of, for example, a digital image signal processing device (DSP), which performs addition and compression processing on image signals input via a data bus 10a, and then outputs an output signal to the power spectrum calculation means 12 via a data bus 11a. supply That is, the addition and compression means 11 adds the image signals from the data bus 10a in one direction on the screen, and outputs the addition result as one line of data to the power spectrum calculation means 12 via the data bus lla.

The power spectrum calculating means 12 is comprised of, for example, a digital signal processing device (DSP), and calculates the power spectrum of the image signal in the frequency domain by performing, for example, one-dimensional Fourier transform on the image signal from the data bus lla. The power spectrum signal from the power spectrum calculation means 12 is output to the determination circuit 13 via the data bus 12a.

The determination circuit 13 stores a reference level in advance, and compares the power spectrum signal level from the data bus 12a with the reference level, and if there is a power spectrum signal that exceeds the reference level, it determines whether characters or the like are on the target object. It is determined that the flag is attached, and a determination signal is output to the data output means 14 via the data bus 13a.

The data output means 14 displays whether or not there are characters or the like on the target object according to the determination signal from the determination circuit 13 using an appropriate display method. This data output means is a visual display means, for example an LCD.

LED, CRT, etc. may also be used, and printing means,
For example, a thermal printer or the like may be used.

Note that the image input means 10. The addition compression means 11, the power spectrum calculation means 121, the determination circuit 13, and the data output means 14 have their control signal input terminals connected to a control means 15 such as a general-purpose computer via a control bus 15a, and each of them The drive is controlled via the means 15 as described above.

FIG. 2 is a block diagram of the additive compression means 11 of the apparatus for recognizing images shown in FIG. The image memory 21 stores image data given from the data bus 10a, for example, for one screen (for example, 512 x 51
It can store data for 2 pixels) and has a capacity of 2 pixels. Of the image data stored in the image memory 21, data addressed by the address generator 22 via the address line 25a is stored in the buffer memory 23, 24, 25 via the data buses 21a, 21b, 21c. . The addressing of the buffer memory 23, 24, 25 takes place via the address line 25b of the address generator 22.

The adder 26 has data buses 23a and 24a.

The data in the buffer memories 23, 24, and 25 are supplied through the buffer memory 25a, the data in the two buffer memories are added, and the addition result is stored in the buffer memory 23 or 25 in an updated manner. Thereby, data corresponding to multiple lines of the screen of data stored in the image memory 21 can be added. Image memory 21. Address generator 22. Buffer memories 23, 24, 25. A control signal from the control means 15 shown in FIG. 1 is supplied to the adder 26 via the control bus 15a, and the addition operation of image data is controlled.

FIG. 3 is a diagram schematically showing the addition operation of the addition compression means 11 shown in FIG. 2. The beginning of the addition operation is as shown in Figure 3 (a
), two lines of image data stored in the image memory 21 are stored in the buffer memories 23 and 2, respectively.
After being added by the adder 26, the addition result is stored and held in the buffer memory 25. In the next addition operation, as shown in FIG. 3(b), the next line of image data is taken from the image memory 21 into the buffer memory 24, and the adder 26 adds the data stored in the buffer memories 24 and 25. After that, the addition result is output to the buffer memory 23 and stored therein.

Furthermore, in the next addition operation, as shown in FIG. 3(C), the next line of image data from the image memory 21 is taken into the buffer memory 24, and is stored in the buffer memories 23 and 24 by the adder 26. After adding the data, the addition result is output to the buffer memory 25 and stored therein. Thereafter, by repeating this operation, image data for a predetermined line is cumulatively added, and the final result is output to the power spectrum calculation means 12 via the data bus 11a.

FIG. 4 is a block diagram of the power spectrum calculation means. The image data from data bus lla is
After being stored in the single-person memory 41, the signal is supplied to the FFT processor 42 and subjected to Fourier transformation. The FFT processor 42 has points equal to the number of pixels in one line, and outputs the conversion result in the form of a complex signal to the output memory 43. Therefore, real number data R and imaginary number data I are stored in the output memory 43 as the FFT results.

The sum of squares calculator 44 is composed of, for example, a multiplier and an adder,
The real number data R1 and the imaginary number data (2) stored in the output memory 43 are each squared to obtain the sum (R2+12), which becomes the power spectrum. Sum of squares calculator 44
may be configured with a digital signal processor (DSP). The power spectrum obtained by the sum-of-squares calculator 44 is stored and held in the output memory 45, and is transmitted to the data bus 12.
The signal is supplied to the determination circuit 13 via a.

FIG. 5 shows the concept of addition compression of image data and power spectrum calculation operation when the target object is stationary. It is assumed that the characters ABC are written on the target object, and the captured image 51 is as shown in FIG. 5(a).

The image memory 21 shown in FIG. 2 stores, for example, 512 lines of data on the captured screen, and the 512 lines of data are added in the manner described in FIG. 3. The data to be added need not be for 512 lines. That is,
The respective data of the corresponding pixels of each line are added in order in the direction of the arrow 51R, and finally the added compressed data 52 for -line is obtained.

Of this -line data, the data 52a portion corresponds to the character portion 51a of the captured image 51, and similarly, the data 52b portion corresponds to the background portion 51b. This addition compressed data 52 is subjected to one-dimensional Fourier transform in the direction of arrow 52R to calculate a frequency spectrum, and by taking the sum of squares, a power spectrum is obtained as shown in FIG. 5(b).

When the target object is stationary, the image data is first added and compressed, and then the power spectrum is calculated. In this power spectrum, as shown in FIG. 5(b), a frequency component with a characteristic part 53, appear above. This is because by adding and compressing the image data, the changes in the density profile of the random data in the background part are canceled out and become smaller, and the changes in the density profile of the data in the text part are at almost similar positions on one line. Due to the fact that it was more emphasized. Therefore, when the level of the power spectrum exceeds the reference level at a certain frequency, it can be determined that a characteristic portion, ie, a character, exists on the image capture screen. The addition and compression operation before calculating the power spectrum may not necessarily be performed when a characteristic portion appears even in one line of data of the captured image.

Alternatively, the presence or absence of a characteristic part may be detected by adding the level data of the power spectrum in the range of band B near the frequency where the characteristic part appears in the power spectrum, and comparing this addition result with a reference value. good.

Further, FIG. 6 is a block diagram of the determination circuit 13. Power spectrum memory 71 stores power spectrum data supplied from data bus 12a. This stored power spectrum data is
This corresponds to FIG. 5(b). In the parameter memory 72,
A reference value for determination is stored. This reference value is shown in Figure 5 (b
) corresponds to the reference level L shown in . Data held in power spectrum memory 71 and parameter memory 72 are supplied to subtracter 73 via data buses 71a and 72a, respectively. The subtracter 73 subtracts the reference value data from the power spectrum data and outputs the result to the determiner 74. The data output from the subtracter 73a may be only the code data of the subtraction result.

If the sign of the subtraction result is positive, the determiner 74 determines that there is a characteristic portion in the captured image. That is, it is determined that characters are attached to the target object, and the data of the determination result is outputted to the data output means 14 via the data bus 13a. If the sign of the subtraction result is negative, it is determined that there is no characteristic part in the captured image, and either no signal is output, or a signal indicating that there is no characteristic part is output to the data output means 14.

Next, we will take an example of detecting characters attached to a target object when the target object to be image recognized moves, that is, the relative positional relationship between the target object and the image recognition device changes, and the image recognition device is shown in Figure 7. Refer to and explain. As shown in FIG. 7, the image signal obtained by the image input means 110 is transmitted to the data bus 1
The signal is supplied to the addition compression means 111 via the data bus 111a via the data bus 111a, and then to the power spectrum calculation means 112. Furthermore, the output of the power spectrum calculation means 112 is again supplied to the addition compression means 111 via the data bus 112a.

Here, the image data supplied from the image input means 110 to the addition compression means 111 is directly outputted to the power spectrum calculation means 112 via the data bus 111a without performing addition compression processing in the addition compression 111. That is, FIG. 8 shows a block diagram of the addition compression means 111, and each component is the same as the image recognition device for recognizing a stationary target object shown in FIG. 2, but the signal flow is slightly different. There is. Image memory 12 of addition compression means 111
The image data stored in 1 is buffered. Furthermore, the output of the power spectrum calculation means 112 is again supplied to the addition compression means 111, and this time addition and compression processing is performed in the same manner as the operation shown in FIG.

Furthermore, this added and compressed data is transferred to the data bus 111.
b to the determining means 113, which compares it with a pre-stored reference level and outputs a determination signal. This judgment signal is transmitted to the data output means 1 via the data bus 113a.
14 to display whether there are characters or the like on the target object.

In addition, in FIG. 7, the power spectrum calculation means 112,
The determining means 113 and the data output means 114 are similar to those shown in FIG.

Next, the concept of addition compression operation and power spectrum calculation operation when the target object is moving is shown in Figure 9 (a) ~
This will be explained in detail with reference to (d).

Image data of a moving object stored in the image memory 121 in FIG. 8 is supplied from the buffer memory 123 or 125 to the power spectrum calculation means 112 via the data bus 111a. The power spectrum calculation means 112 performs a Fourier transform on each line of input data, calculates the sum of squares, and outputs the resultant data via the data bus 112a.
The power spectrum for each line is sent to the addition compression means 111.

The power spectrum calculation means 112 calculates the power spectrum as shown in FIG. 9(a).
Fourier transform is performed on the captured image 61 in the direction of the arrow 61R, and the sum of squares of the complex data is calculated to derive the power spectrum. This power spectrum is schematically shown in Figure 9 (
As shown in 62 of b), frequencies 0 to +f and -f to -0 are symmetrical, and this power spectrum corresponds to a collection of data for each line of the captured image,
For example, as shown in FIG. 9(c), a number corresponding to the number of lines n can be obtained. In the case of a moving image, characteristic parts do not clearly appear in each power spectrum of each line shown in FIG. 9(C). Therefore, these power spectra 631~
When 63n is added by the addition compression means 111, the result shown in FIG.
The power spectrum becomes as shown in d), and the characteristic portion 64 clearly appears. Therefore, by comparing the level of this power spectrum with a reference value, the presence or absence of a characteristic portion can be detected.

Even in this case, the presence or absence of the characteristic part may be detected by adding the level data of the power spectrum in the range of band B near the frequency where the characteristic part appears, and comparing this addition result with the reference value. good. In the embodiment shown in FIG. 7, the output of the image input means 110 is added to the addition compression means
However, this addition compression means 111 only stores the image input data in the buffer memories 123 and 125 and does not perform any addition compression operation, so the image input data is directly input to the power source as shown in FIG. A configuration may also be used in which the signal is supplied to the spectrum calculation means 112 and the output thereof is supplied to the addition compression means 111.

Next, an image recognition device that can recognize characters and the like attached to a target object regardless of whether the target object is stationary or moving will be described. FIG. 11 shows a block diagram of an image recognition device capable of image recognition regardless of the state of the target object.

The image signal obtained by the image input means 210 is supplied to the addition compression means 211 via the data bus 210a.

In addition compression means 211, when the image input data is stationary, after performing addition compression processing, the switch 227 is switched and the data is supplied to power spectrum calculation means 212 via data bus 211a, and is then moved. In this case, the switch 227 is switched and the image input data is directly supplied to the power spectrum calculation means 212 via the data bus 211a.

Further, the power spectrum calculation means 212 calculates the power spectrum of the image input data, and if the image input data has been subjected to addition compression processing, the switch 246
and the determination circuit 21 via the data bus 212a.
If the image input data has not been subjected to addition compression processing, the switch 246 is switched to output data to data bus 2.
12b to the addition compression means 211. In this case, the addition compression means 211 performs addition compression of the power spectrum signal, switches the switch 227, and outputs the result to the determination circuit 213 via the data bus 211b.゛The determination circuit 213 compares the power spectrum signal level from the data bus 212a or the data bus 211b with a pre-stored reference level, and based on the comparison result, determines whether or not there are characters, etc. on the target object. A determination signal is output to the data output means 214. Furthermore, the data output means 214 displays the determination result in accordance with the determination signal. In addition, the switch 227° 24 is used when processing when the target object is stationary or moving.
6, the output of the determination circuit 213 is supplied to the detector 216, and the detector 216 supplies a control signal for switching between a stationary state and a moving state to each switch 227, 246 depending on the presence or absence of an output signal. and will be carried out in a timely manner.

Further, addition compression means 211. Power spectrum calculation means 212 and determination circuit 213. Each operation of the data output means 214 is similar to that shown in FIG. 1 or FIG. 7 described above.

Note that the control of the switching devices 227 and 246 depending on whether the target object is stationary or moving may be done manually by specifying 0, or by determining the compression level and reference level of the power spectrum signal and switching based on the results. It is possible.

This device can be used to recognize prints on stationary objects or even on moving objects with a high-speed camera (shack - fast speed, e.g. 1/10
Of course, this method can also be applied to cases in which images can be captured in an almost still state by using a camera with a camera of 0.00 seconds or the like.

The device of the present invention adds characters in one direction on the screen when characters appear relatively clearly in the input image, and
The image data corresponds to the line. There are some unevenness at the positions corresponding to the characters in the density profile of this one line data. Therefore, when one-dimensional Fourier transform is applied to this one-line data, a characteristic peak appears in the power spectrum, and by determining the level of this peak, it is possible to determine whether there is printing on the target object.

If a characteristic peak does not appear in the -dimensional Fourier transform output and it is dispersed, add a predetermined band of the -dimensional Fourier transform output (for example, band B in FIG. 5(b)) and level the added output. It would be useful if it could be determined. Of course, if the background noise is thick or the characters engraved on the target object are tilted, etc., if you perform additive compression on the data in the first spatial area, the feature will be the addition of the data on each line. As a result, the characteristic spectrum may not appear due to cancellation. In such a case, the method shown in FIG. 9, in which the data of each line is subjected to Fourier transformation and then added, is useful. In this case, although the power spectra shown in FIG. 9(b) are not necessarily constant, the spatial frequencies of the horizontal stripes 61b corresponding to the characters shown in FIG. 9(a) are included in common. Therefore, FIG. 9(b)
When the respective power spectra shown in FIG. 9 are added, a characteristic peak appears in the addition result as shown in FIG. 9(C).

Moreover, even in this addition result, the illumination state.

A characteristic peak may not appear depending on the running condition of the object, etc. In such a case, add the spectra of a certain band 1 centered around the frequency estimated to show the above peak value in the frequency domain, for example, band B shown in FIG. level to be compared with the reference level. Such means can be appropriately set according to the characteristics of the target image, and thereby it is possible to provide an apparatus that can reliably recognize printing on the target object.

In this way, the image recognition device adds the input image data directly or unidirectionally via the addition compression means, performs Fourier transformation, calculates the power spectrum, and further,
This is added in a certain band to obtain a stable spectrum, and this calculated value is compared with a reference value to perform image recognition. This allows image data to be handled in the spatial frequency domain. Even when there is almost no density difference between the object and the target part, the target part can be grasped reliably. Further, according to this, the image data is processed directly or after being added in one direction, thereby achieving substantially constant high-speed processing according to the amount of data.

Furthermore, according to this, even in cases where there is almost no density difference between the target object and the target part and the power spectrum is dispersed in a predetermined band, the extracted image can be reliably understood. Recognition becomes possible. Even when the target object has an unclearly dispersed power spectrum, such as when the target object is being transported, the target part can be grasped reliably, and recognition as reliable as possible is possible.

Furthermore, although each of the image recognition devices described above is configured to directly lead the image data from the image input means to the addition compression means, the present invention is not limited to this, and depending on the type of image signal, It is also possible to arrange a pre-processing means such as gradation conversion of image data integrally or separately.

Further, in the above embodiment, a case has been described in which the analog signal is A/D converted and handled, but the present invention is not limited to this, and it is also possible to handle the analog signal as it is.

Note that the determining means 13, 113, and 213 may be provided with adders 75 and 76 for adding compression levels, as shown in FIG.

[Effects of the Invention] As described above, according to the present invention, when the image data of a target object is stationary, the image data is added and compressed line by line, and then Fourier transformed to obtain a power spectrum signal. In addition, when the image data is moving, the image data is Fourier transformed line by line to calculate the power spectrum signal, and these calculated outputs are added and compressed. The target part of the target object can be easily distinguished by the emphasis section in these additive compressions.

Therefore, the influence of imaging conditions, illumination conditions, and running conditions can be reduced, and the target object or the target part within the target object can be stably detected.

4. Detailed Description of the Invention FIG. 1 is a block diagram showing an embodiment of the image recognition device according to the present invention when the image is somewhat still, and FIG. 2 is a block diagram showing the addition compression means of FIG. 1. Figure 3 is a block diagram showing the addition compression operation in Figure 1, Figure 4 is a block diagram showing the power spectrum calculation means in Figure 1, and Figure 5 is an image when the target image is stationary. FIG. 6 is a block diagram showing the determination means of FIG. 1, and FIG. 7 is a diagram showing the operation of the recognition device.
10 and 10 are block diagrams showing an embodiment of the image recognition device when the target image is moving, FIG. 8 is a block diagram showing the addition compression means of FIG. 7, and FIG. FIG. 11 is a block diagram showing an image recognition device capable of image recognition regardless of the operation of the target image, and FIG. 12 is a block diagram in which an adder is provided as a determination means. be.

10 110.210... Image input means.

11.111,211... Addition compression means.

12.112,212...Power spectrum calculation means.

13.113, 21.3... Determination means.

14.114,214...Data output means.

15.115,215... Control means agent Patent attorney Nori Chika Ken Yudo Yamashita - 1st 30th 4th 5th Kasumi 6th En

Claims (3)

[Claims] (1) an imaging means for imaging a stationary target object and outputting image data; an addition compression means for inputting the imaging output and adding image data of the target object in one direction on the screen;
A power spectrum calculation means for calculating a power spectrum by Fourier transforming the output signal of the addition compression means, a circuit for determining whether or not the output of the power spectrum calculation means exceeds a reference value, and an output signal as a result of this determination. An image recognition apparatus, comprising: a display means for displaying whether or not a desired image is present during imaging of the target object. (2) an imaging means for imaging a moving target object and outputting image data; a power spectrum calculation means for inputting the imaging output and calculating a power spectrum by Fourier transforming the image data of the target object; an addition compression means for adding the output of the power spectrum calculation means in one direction on the screen; a circuit for determining whether the output of the addition compression means exceeds a reference value; and an output signal of the determination result is supplied to the circuit. An image recognition device comprising: a display means for displaying whether or not a desired image is present during imaging of a target object. (3) an imaging means that images a target object and outputs an image signal; an addition compression means that adds image data of the target object in one direction on the screen; and an output signal of the imaging means or an output of the addition compression means. A power spectrum calculation means for calculating a power spectrum by Fourier transforming a signal, and a level of a power spectrum signal outputted from the addition compression means or the power spectrum calculation means is compared with a reference value to determine whether the level exceeds the reference value. a determination circuit that determines whether or not a desired image is present in the captured images of the target object to which the output signal of the determination circuit is supplied; The output image signal is added line by line on the screen by the addition compression means, and the power spectrum of the signal obtained as a result of this addition is calculated by the power spectrum calculation means,
Furthermore, when the output image signal of the imaging means is a moving image,
An image recognition apparatus comprising: a control means for controlling the addition compression means and the power spectrum calculation means so that the power spectrum calculation means calculates a power spectrum of an output image signal of the imaging means.