JPH05165965A - Edge detection device - Google Patents

Abstract

PURPOSE:To securely detect only the edge part of an object image in an image which includes a background by comparing a statistical quantity ratio with a threshold value and deciding whether or not image data is edge data. CONSTITUTION:The square mu<2> of the mean value mu of image data in a local area which is obtained by passing the image data through a mean value calculating means 3, a variance value V found by a variance value calculating means 4, and their statistical quantity ratio V/mu<2> are used as materials to decide the edge. The statistical quantity ratio V/mu<2> does not depend upon lighting conditions and increases at the image edge part. For the purpose, the statistical quantity ratio V/mu<2> is compared with the threshold value for edge decision making which is previously set by statistical experiments, etc., to judge the edge by an edge decision means 9. Consequently, only the edge part of the object image in the image which includes the background can securely be detected without depending upon the lighting conditions.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an edge detecting device for detecting edges from an image.

[0002]

2. Description of the Related Art Generally, in image processing such as image feature extraction, image edge detection is one of the most basic processes. It should be noted that the edge mentioned here means a boundary portion between the target object and the background of the target object, and generally means a portion where the brightness of the image largely changes.

FIG. 5 shows a schematic block diagram of a conventional edge detecting apparatus. In FIG. 5, 11 is an image memory in which image data is stored in advance. Reference numeral 12 is a spatial filter for finding the brightness difference in the local region. Reference numeral 13 is a comparator, which compares the data obtained by the spatial filter 12 with a threshold value. 14 is a fixed value, which is used as a threshold value.

In the operation of the conventional edge detecting apparatus configured as described above, first, the image data stored in the image memory 11 is passed through the spatial filter 12 so that the pixel in the place where the luminance of the image greatly changes. Only the brightness of is emphasized. A differential operator is often used for the spatial filter 12. This is within a local region (eg m ×
Image data I (1,1) to I (m, n) and weighting factors w (1,1) to w (m, n) in n pixels)

[0005]

[Equation 1]

Is a method for obtaining the brightness difference ΔI in the local area by. This matrix of weighting factors is called an operation mask. For example, when the image data in the local region is as shown in FIG. 6A, if weighting is performed using an operation mask of 3 × 3 pixels as shown in FIG. The luminance difference ΔI in (2, 2) is given by

[0007]

[Expression 2] ΔI = | −I (1,2) −I (2,1) + 4I (2,2) −I (2,3) −I (3,2)
| A spatial filter using such an operation mask is called a Laplacian filter. Further, when different operation masks are used in the horizontal direction and the vertical direction as shown in FIG. 7, the brightness difference ΔI _x between the horizontal direction and the vertical direction,
ΔI _y is given by Equation 3.

[0008]

## EQU3 ## ΔI _x = I (1,3) + 2I (2,3) + I (3,3)-(I (1,1) + 2I (2,1) + I (3,1)) ΔI _y = I (3,1) ＋ 2I (3,2) ＋ I (3,3) − (I (1,1) ＋ 2I (1,2) ＋ I (1,3)) and the pixel I (2,2) in the local area Difference in brightness ΔI
Is obtained by the following equation 4.

[0009]

[Equation 4]

A spatial filter using such a mask is called a Sobel filter. The spatial filtering processing by these difference type operators is described, for example, on pages 280 to 283 of the first edition of the third edition of the image processing handbook (March, 1991).

The pixel brightness difference thus obtained is compared with the comparator of FIG.
The threshold value of 13 is compared with the threshold value of the fixed value 14. If it is larger than the threshold value, the comparator 13 outputs "1", and if it is smaller than "0", the pixel of interest is an edge. It is determined whether or not, and as shown in FIG. 8, only the edge portion of the image is output and the edge is detected.

Now, the principle of detecting an edge in an image by using the brightness difference will be described. The difference between the target object and the background generally appears in the difference in surface reflectance. Now, assuming that the incident light is i, the reflected light of the target object is I, the reflected light of the background is I ′, the surface reflectance of the target object is r, and the surface reflectance of the background is r ′, the reflected light I, I ′ is several. Represented in 5.

[0013]

## EQU00005 ## I = r.times.i I '= r'.times.i Also, the difference .DELTA.I in the reflected light between the target object and the background is given by

[0014]

ΔI = | I−I ′ | = | r−r ′ | × i The difference in surface reflectance appears as a difference in reflected light. The reflected light is the brightness of the image data, and the difference ΔI in the reflected light is the brightness difference between the target object and the background. If the difference in the reflectance between the target object and the background is large, the brightness difference ΔI becomes large, and therefore the edge can be detected.

[0015]

However, in the conventional edge detecting apparatus as described above, the brightness difference ΔI depends on the incident light i, and is dependent on the illumination condition at the time of image input, as is clear from the equation (6). Understand that That is, in the case of an image input under dark illumination, even if the difference in reflectance between the target object and the background is large, if the incident light i is small, the brightness difference ΔI will be small, and it will be difficult to detect it as an edge. ..

On the contrary, when the illumination at the time of image input is very bright, the incident light i is very large and the luminance difference ΔI is large even if the difference in reflectance between the portion other than the target object and the background is small. Therefore, there is a possibility that the edge will be erroneously detected.

As described above, the conventional technique has a problem that the edge of the target portion to be detected cannot be detected or the error is detected depending on the illumination condition at the time of image input.

In order to solve the above problems, the present invention aims to provide an edge detecting apparatus capable of reliably detecting only the edge portion of an image of interest from an image including a background without depending on illumination conditions. And

[0019]

According to the present invention, an average value calculating means for calculating an average value in a local area from image data, a dispersion value calculating means for calculating a dispersion value in the local area, and the average value calculation. Means for squaring the average value obtained by means,
The image data is obtained by comparing the output of the statistical value ratio calculating means with a threshold value, and the statistical value ratio calculating means for calculating a ratio between the square of the average value and the distributed value calculated by the distributed value calculating means. Edge determination means for determining whether or not the edge is an edge.

[0020]

According to the present invention, the mean value μ squared μ ² and the variance value V of the image data in the local area of the image data are obtained, and the statistical value ratio V / μ ² is used as the edge judgment data. , The statistic ratio V / μ ² does not depend on the lighting condition,
Also, since there is a property that it becomes large at the image edge part,
By comparing the above-mentioned statistic ratio V / μ ² with a threshold value for judging an edge set in advance by a statistical experiment or the like, and judging whether it is an edge or not, an edge portion of an image including a background It is possible to reliably detect the light without depending on the illumination.

[0021]

[Practical Example] FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention. Reference numeral 1 is an image memory in which image data prepared in advance is stored. 2 is a delay means, which is the image memory 1
The image data sent from is divided into three horizontal line data. Reference numeral 3 denotes an average value calculating means, which calculates an average value in the local area. Reference numeral 4 denotes a variance value calculating means, which obtains a variance value in the local area. Reference numeral 5 denotes a multiplier, which outputs the output of the average value calculation means 3 to 2
Get on. Buffers 6 and 7 are for sequentially storing the data sent through the average value calculating means 3 and the variance value calculating means 4. 8 is a statistic ratio calculation means,
A statistical amount ratio between the square of the average value in the local area stored in the buffer 6 and the variance value in the local area stored in the buffer 7 is obtained. Reference numeral 9 denotes an edge determination means, which compares the statistic ratio with a threshold value to determine whether it is an edge, 10 is a control unit, which is an output of the average value calculation means 3 stored in the buffer 6. And the output of the variance value calculation means 4 stored in the buffer 7,
By controlling the buffer 6 and the buffer 7, the data of the same pixel position (address) is synchronously input to the input of the statistic ratio calculating means 8.

The operation of the present invention configured as above is as follows.
First, the control unit 10 controls the output of the image memory 1 so that among the horizontal lines of the stored image data, only the data of three consecutive lines in the vertical direction is dispersed through the delay unit 2 with the average value calculation unit 3. It is sent to the value calculation means 4, respectively. In the average value calculation means 3, the local area (mask) is shifted pixel by pixel from the sent data for three lines, and
The average value (in the mask) is sequentially obtained and sent to the multiplier 5.

Here, the average value μ in the local area by the average value calculating means 3 is obtained by calculating the image data in the mask shown in FIG.

[0024]

[Equation 7]

This average value μ is squared by the multiplier 5 to become μ ² , which is sequentially sent to the buffer 6, and μ ^{2 for} one line is sent.
Are stored in the buffer 6. In the variance value calculating means 4, the average value μ in the mask is calculated from the received three lines of data in the local area (in the mask) as shown in FIG. 6A. It is obtained by performing the same processing as that of the calculating means 3, and then the variance value V in the mask is obtained by performing the operation of Eq.

[0026]

[Equation 8]

The variance value V in the local area thus obtained
Is stored in the buffer 7 for one line while shifting the mask by one pixel at a time. After all, the output of the average value calculating means 3 is sent to the buffer 6 through the multiplier 5, and the output of the variance value calculating means 4 is sent to the buffer 7.

Next, the control unit 10 sends the data for one pixel from the buffer 6 to the statistic amount calculating means 8, and at the same time, the buffer 7 outputs the data at the same address as the output of the buffer 6 to the statistic amount ratio. It is sent to the calculation means 8. By repeating this while shifting one line at a time over all the lines, the square of the average value in the local regions of all the one screen and the variance value in the local regions are sequentially synchronized and sequentially sent to the statistic ratio calculating means 8. Will be sent.

2 sent to the statistical quantity ratio calculating means 8 in this way
From the two data, the mean squared μ ² and the statistic ratio V / μ ^{2 of the} variances in the local region are obtained. The obtained statistic ratio and the threshold value are compared by the edge determination means 9, and if larger than the threshold value, the edge determination means 9 outputs "1", and if smaller, outputs "0" to output the center of the local area. It is determined whether the pixel of is an edge.

The nature of the statistical ratio will be described below. If there is image data as shown in Fig. 2 (a),
The histogram of the local area of this image is as shown in FIG. 2 (b), and it can be seen that the variance value V is large when the local area includes the edge portion of the image and is small when it does not.

On the other hand, since the average value μ in the local area can be considered as the background brightness in the local area, the statistic ratio V / μ ² is calculated as follows: It can be seen that this is a measure for examining how large the brightness change is. Therefore, the statistic ratio V
/ Μ ² changes depending on the intensity of the edge in the image, and the edge can be detected by comparing this with a threshold value.

Further, the property that the statistic ratio is not influenced by the illuminance will be described. Now, the incident light is i again, and the image data in the local region (N = m × n pixels) is I (1,1)-
I (m, n), and the surface reflectance in this region is r
(1,1) to r (m, n), the background brightness is defined as I ', and the surface reflectance of the background is defined as r'. At this time, the incident light i can be considered to be substantially constant in the local region. Further, if it is considered that the average value μ in the local area obtained by passing the image data through the average value calculating means 3 is approximately replaced by the background luminance I ′, then the above-mentioned expression 5 holds and the expression 9 is established.

[0033]

When μ = I ′ = i × r ′, the variance value V in the local area obtained by passing the image data through the variance value calculating means 4 is given by the following equations 8 and 9: ..

[0034]

[Equation 10]

Therefore, the statistic ratio V in the local region is
/ Μ ² is calculated as in Eq.

[0036]

[Equation 11]

That is, it can be seen that the statistic ratio V / μ ² is determined only by the values of the reflectances r and r'and does not depend on the incident light i.

Therefore, according to equation 11, even if the illumination is dark as shown in FIG. 3A and the difference (ΔI) between the luminance level of the object to be detected and the luminance level of the background is small, the reflection of the object part is reflected. If the change in the rate r is more significant than the background reflectance r ', the statistic ratio V / μ ² becomes relatively large and can be detected as an edge.

On the contrary, as shown in FIG. 3 (b), even if the illumination is bright and the difference in brightness (ΔI) between the non-target luminance and the background is large, the change in the reflectance r of the non-target portion causes the background Reflectance r '
If it is not remarkable as compared with the above, the statistic ratio V / μ ² becomes relatively small and is not erroneously detected as an edge.

By the way, the statistics ratio V / μ ² in the embodiment
And the logarithmically converted luminance ratio logarithmic value log (V / μ ² ) is
It looks like 12.

[0041]

[Equation 12]

Therefore, the statistical value ratio calculating means 8 in FIG. 1 is configured as shown in FIG. 4 by using the difference value after logarithmic conversion of the variance value V and the square μ ² of the average value. Is feasible.

That is, in FIG. 4, logarithmic conversion means 81 and 82 are prepared to perform logarithmic conversion, and the difference value calculation means 83 is used to obtain the difference value after logarithmic conversion. In this way, the average value of 2 in the local area of the buffer 6 in FIG.
The output of the power μ ² and the variance value V in the local area output by the buffer 7 are logarithmically converted by the logarithmic conversion means 81 and 82 of FIG. 4, respectively, to obtain log (μ ² ) and log (V). Be done. Next, the difference value calculating means 83 calculates the statistical ratio logarithmic value = log (V).
−log (μ ² ) is obtained, and the statistic ratio calculating means can be realized. This has the advantage that it is not necessary to use a divider with a complicated structure.

Although the present invention has been described above, the present invention passes the value obtained by squaring the average value in the local area obtained by passing the image data through the average value calculating means, and the image data through the variance value calculating means. Since the edge is detected using the ratio of the variance values in the local area obtained by the above, if the brightness difference between the target brightness level to be detected and the background brightness level is small depending on the lighting conditions, Even if there is a large difference in brightness between the background brightness and the background brightness, it is possible to reliably detect only the edge portion of the target image from the image including the background. Further, since the present invention uses the statistic in the local area, that is, the average value and the variance value, even when random noise is included in the image data, the influence is small and the edge can be detected correctly. ..

The variance value used when calculating the statistic ratio may be obtained by squaring each image data in the local area and averaging the squared values, that is, the mean square value may be used instead. good. This is because, as is clear from the right side of Equation 11, the square mean and the statistic ratio by the square of the mean are 1
This is because it only increases by one, and can be implemented by reducing the threshold value by one. Thus, by using the square mean value instead of the variance value, it is possible to omit the preprocessing for calculating the mean value when the variance value is calculated.

The reciprocal of the statistical quantity ratio may be used. In this case, the edge determining means outputs "1" if the reciprocal of the statistic ratio is smaller than the threshold value, and outputs "0" if it is larger.

Furthermore, although the pixel size of the local area (mask) is 3 × 3, it is not limited to this, and 4 × 4, 5 ×
The pixel size may be 5, or any other pixel size.

[0048]

As described above, in the edge detecting apparatus of the present invention, the value obtained by squaring the average value of the image data in the local region obtained through the average value calculating means and the image data variance value calculating means. Since the edge is detected using the ratio of the variance values in the local area obtained by passing through, when the brightness difference between the brightness level of the target to be detected and the brightness level of the background depending on the illumination condition is small, or Even if the difference in brightness between the non-target brightness and the background brightness is large, it is possible to reliably detect only the edge portion of the target image from the image including the background. Further, even when random noise or the like is included in the image, there is an effect that an edge can be correctly detected without being affected by noise.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

FIG. 2 is a diagram of image data for explaining the operation of FIG.

FIG. 3 is a diagram illustrating an operating characteristic that changes according to the illumination of the present invention.

FIG. 4 is a block diagram showing the configuration of another example of the statistic ratio calculating means in FIG.

FIG. 5 is a schematic block diagram of a conventional edge detection device.

FIG. 6 is a layout diagram (a) of image data in a local area for explaining an embodiment of the present invention and a conventional example, and an operation mask configuration diagram (b) of a spatial filter in the conventional example.

FIG. 7 is a horizontal operation mask configuration diagram (a) and a vertical operation mask configuration diagram (b) of a Sobel filter for explaining a conventional example.

FIG. 8 is a diagram illustrating operating characteristics in a conventional example.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Image memory, 2 ... Delay means, 3 ... Average value calculation means, 4 ... Variance value calculation means, 5 ... Multiplier, 6, 7 ...
Buffer, 8 ... Statistics ratio calculation means, 9 ... Edge determination means, 10 ... Control section, 81, 82 ... Logarithmic conversion means, 83 ...
Difference value calculation means.

Claims (2)

[Claims] 1. An average value calculating means for calculating an average value in a local area from image data, a dispersion value calculating means for calculating a dispersion value in the local area, and an average value calculated by the average value calculating means is 2 Means for multiplying, a statistic amount ratio calculating unit for calculating a ratio of the square of the average value and the variance value calculated by the variance value calculating unit, and an output of the statistic amount ratio calculating unit is compared with a threshold value. An edge detection device comprising: an edge determination unit that determines whether or not the image data is an edge. 2. The statistic ratio calculation means logarithmically transforms a value obtained by squaring the average value in the local area obtained by the average value calculation means and a variance value in the local area obtained by the variance value calculation means. 2. The edge detecting apparatus according to claim 1, further comprising: a means and a difference value calculating means for obtaining a difference value of the output after the logarithmic conversion by the logarithmic converting means.