JPS63153682A - Method and apparatus for processing contrast picture - Google Patents

Abstract

PURPOSE:To extract an object picture such as a character or the like sharply at a high speed even from a complicated picture or even under the environment where uneven lightness is fluctuated for 1min to the next by applying arithmetic calculation between a background picture and a picture being object of processing. CONSTITUTION:Figure gA is a figure indicating only one line A-A' of figure fg showing an input picture, with the position as the X axis and the lightness as the Y axis. A filter is provided for three picture elements f1-f3 to process the relation of mix (fi)i=1-3 and to write the result to the position of the element f2. In applying the processing above to all picture elements, Figure gA<(>A<)> is obtained in which the lightness of the data is increased. Let the character width be 5 picture elements, then the dark portion of a character is replaced completely into the surrounding background picture data in repeating the said processing thrice and the background picture is generated easily. In taking a difference between the picture shown in figure gA<(3)> and the Figure gA, the character part only is made sharp and the object picture is extracted.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to grayscale image processing, and in particular to a grayscale image processing method and apparatus that can be applied to images with complex backgrounds and uneven brightness. Regarding.

[Conventional technology]

When recognizing characters, etc. using an image processing device, it is generally necessary to compress image data taken with a television camera or the like into binary data of '0' and '1' at a certain threshold level and process it. For example, the character is '1' and the background is i.
The data is binarized to a level of 0+, and recognition processing is performed on the '1' data.

By the way, when the target is a fairly clear image, such as when recognizing characters written in black on a blank sheet of paper, the above-mentioned threshold level can be easily determined in advance (for example, by calculating the average density). However, in order to deal with more difficult applications, the simple binarization process described above often does not provide good results.

Examples of this application include the following:

(1) Extract characters from a patterned cardboard box.

(2) Extract characters from outside signboards, etc.

(3) Extract characters from the printed circuit board.

Simple binarization cannot be applied to such objects because the background of the characters is complex and the brightness fluctuates rapidly, so
It is necessary to devise a binarization method to extract good characters etc. from these objects.

As a conventional example, for example, "Study of character extraction method from scene images" (Information Processing Society of Japan National Conference Proceedings):
A method for extracting characters from complex or low-contrast images has been proposed, as described in '86.33 by Otani).

As shown in FIG.
The optimal threshold level θha for binarization within this block is determined (FIG. 2).

The threshold value θi at this time is regarded as a two-class problem in which the sub-block 141 is separated into two classes, black and white, and is set to a value that maximizes the inter-class variance. Furthermore,
In order to maintain continuity between sub-blocks 141, interpolation is performed for each pixel using the θ-th as shown in FIG. . , y are determined in advance and the input image is binarized.

Furthermore, in JP-A-61-7406, when detecting defects in shape, a reference image is stored in advance, and the reference image and the object to be inspected are compared (for example, by difference) to detect defects. A method for detecting this has also been proposed.

[Problem that the invention seeks to solve]

The problem with the above conventional technology is that in the former, a density histogram (the frequency of each density level is determined within a subblock) is used to determine the threshold value θN within a subblock, and two-dimensional image data is is converted into one-dimensional data, the positional information of brightness cannot be taken into account, and the optimal threshold value cannot be determined.

Also, for each pixel, the threshold value θ is determined by interpolation using θi−.
Ill? , which differs from the true threshold.Furthermore, each of the above processes takes processing time, making it impossible to extract characters in real time.
” On the other hand, the latter method, which compares both the dark and dark images themselves with a reference image stored in advance, can remove the effects of shading and is fast if the brightness, position, size, etc. are always constant. However, in reality, fluctuations in brightness cannot be removed, and this has the disadvantage that it cannot be applied to complex images or objects whose brightness changes in real time.

In view of the above-mentioned shortcomings of the prior art, it is an object of the present invention to quickly and clearly print out characters from images to be processed, even in complex images or in environments where uneven brightness changes over time. The goal is to extract the target image.

[Means for solving problems]

By the way, the reason why it is difficult to extract a target image from a complex target image or a target image with changes in brightness is that it is difficult to find a threshold value for extracting the target image over a predetermined area, and it is difficult to extract the target image from a complex target image or a target image with changes in brightness. This is thought to be because the threshold value is also affected.

However, 1. For example, the meaning that the letters, etc. are black means that the letters, etc. are darker than RH.

The target image can be recognized not by whether it is bright or dark in absolute value, but by the contrast between the brightness of the target image and the brightness of surrounding images, and in this case, it is thought that it is less affected by changes in brightness. .

The present invention focuses on this point, and has the above object.

An image to be processed with a brightness level of several levels is input, a background image is created from the inputted image to be processed, and a target image is extracted by an inter-image operation, for example, a comparison operation, between this background image and the above-mentioned image to be processed. By applying this idea, even more advanced processing can be achieved.

Here, the target image is an image to be extracted.

Furthermore, the background image is an image obtained by removing the target image from the processing target image, and includes not only an image in which the influence of the target image is completely removed, but also an image in which the influence of the target image is reduced.

[Effect]

In the present invention, a background image is created based on the input image to be processed. Therefore, this background image also reflects the contents of the target image at that time or changes in brightness.

Therefore, if a calculation is performed between this background image and the input processing target image, and the background image is removed from the target image,
A target image can be extracted. In this case, the background image is created from the image to be processed, so there is no need to set it in advance, and it also reflects the brightness etc. at that time.
This means that images can be extracted quickly and clearly.

Note that a specific method for creating this background image will be explained in detail in the following explanation by giving various examples.

〔Example〕

First, an overview of the method for creating a background image and the method for creating a contrast image will be briefly explained using an example of a -dimensional image.

FIG. 3(1) shows the input image fs, of which A-A'
g^ in Fig. 2 represents only one line of , with the position on the horizontal axis and the brightness on the vertical axis. Here, if you want to extract a character, a preset threshold value TH is set, and parts brighter than this are set to '0'.

It is also possible to perform binarization in which the dark portion is set to 11'. However, since this threshold value TH is an absolute value, it becomes difficult to determine Ts if the image to be processed, that is, the input image, has uneven brightness or a complex background.

Therefore, a method of creating a background image and taking the difference is used. For the input image, for example, 3 pixels f1r as shown in FIG.
Provide fz and fs filters, and use a+a with this filter.
The process of finding x(ft) i = 1 to 3 and writing it to the fz position is performed by rask scanning and is now shown in Figure 3 (
Assuming that the character width in 1) is 5 pixels, repeating the above process three times will completely replace the dark part of the character with the surrounding background image data, as shown in FIG. In other words, if you perform processing according to the character width.

Background images can be easily created. If we take the difference between this background image (δ) g^ and the input image g^, only the character part becomes clear with the brightness at the 10' level as a reference, as shown in Fig. 5. This difference image, that is, the target image, is obtained from the contrast of the region to be extracted, and remains almost constant even if the brightness of the entire screen changes. Therefore, the binarization threshold when extracting a character, for example, from such an input image can be easily set without using complicated calculations.

The above describes the extraction of dark characters, but in the case of bright character extraction, the above filter calculation is performed using win(f+)i
= 1 to 3, and the background image may be subtracted from the input image.

By the way, in an actual image, not only a clear input image as shown in Fig. 3, but also a gradient (unevenness) of brightness as shown in Fig.
If only the above-mentioned processing is performed on such an image, in which there are many objects with a large number of objects, even the slanted background portion (hatched area in the figure) as shown in FIG. 1 (4) will be extracted. Therefore, in such a case, the image (wax(ft)i=
The image obtained by executing the results of steps 1 to 3 n times) is further updated with wi
It is also possible to perform a process of repeating n(ft)i=1 to 3 the same number of times. Through this process, the brightness of the concave area that was once filled remains as it is, and only the slope area returns to the same brightness as the input image. Therefore, by taking the difference between this image and the input image, only the text portion can be extracted. Such a method of creating a background image, calculating the difference from the input image, and binarizing it is hereinafter referred to as a background image difference method.

Here, the process of giving the values of wax (ft) i = 1 to n to the center of the filter of fl is called local maximum filter processing, while the case of win (ft) i = l to n is called local minimum filter processing. It is called.

Although FIG. 1 shows an image with shading, shading correction is generally performed by photographing a white sheet of paper in advance, calculating the unevenness of the obtained image, and storing a correction value for each pixel. When online, the density of each pixel of the image to be processed that is obtained one after another is corrected using the correction value stored above. However, in this correction method, the shading state is always constant;
Furthermore, it is not effective unless the overall brightness is constant, but in the background image subtraction method of the present invention described above, the background image is created directly from the obtained input image, so it is not effective against shading conditions and brightness fluctuations. can also be effectively dealt with.

Hereinafter, a specific embodiment of the present invention will be described using FIG. FIG. 5 is a diagram showing one configuration of the present invention, in which a television camera 10. A/D converter 119 Image memory 129 Local maximum value filter circuit 139 Local minimum value filter circuit 1419
Inter-image calculation circuit 15° binarization circuit 16. CPU17. D
/A conversionill Consists of 19 8° monitors.

In such a configuration, an image signal photographed by a television camera 10 or the like is converted into density data of, for example, 128 gradations by an A/D converter 11 and stored in an image memory 12.
is memorized. Here, the image memory 12 includes, for example, 25
The 6x256 pixel grayscale image memory has 01~Gkm.

Q pieces of binary image memory 81 to Bm are provided as required.

By the way, in the explanation of the local maximum value and minimum value filter processing mentioned above, a -dimensional filter was used, but the actual image is
Since the filtering process is also 3× as shown in FIG.
The maximum value and minimum value are extracted from the three pixels ft, f2, .

Now, assuming that the size of the above filter is a commonly used 3x3 pixel filter, the text is darker than the background.

The processing procedure for extracting characters is shown below.

As shown in the procedure in FIG. 7, the input grayscale image Gl is filtered by the local maximum value filter processing circuit 13 and stored in the grayscale image memory of G2 (91), and the image of Gz is similarly subjected to the maximum value filtering process. and store G2^ again (
93), this maximum value filtering process is repeated until one character width is filled. Here, it is applied a total of m times (92
), this 02 image is further processed by local minimum value filter processing circuit 1
4 to filter and store it again in 02 image (95
). This process is repeated n times (94)-.

This n is usually n=m.

The finally obtained aS image is 1m relative to the input image G1.
This is an image that has been subjected to local maximum value filtering twice and local minimum value filtering n times, and this image G2 becomes the background image. Here, in order to extract characters that are darker than the background, the background image is filled with brightness depressions. If, for example, a pixel-by-pixel difference is calculated between this image Gt, the input image G1, and the inter-image calculation circuit (96), in this example, only the dark character area becomes brighter, and the surrounding background becomes almost 10 lvl (Ga
image).

When extracting characters by binarizing this image G8, the binarization circuit 16 performs binarization to extract pixels brighter than a predetermined threshold level, or
Based on the density level of this difference image G8, for example, the average value of the maximum density and minimum density or the overall average density may be binarized using the threshold level. In any case, the threshold level is It is easily found (97).

In addition, if you calculate the difference between the input image and the image obtained by adding a constant value to the two background images in addition to the above-mentioned inversion, and perform binarization to extract 0 or more of this contrast image, it is possible to convert the contrast image into a fixed binary value. is similarly possible, and if the difference image between the input image and the background image is divided by a scale of 1/2 or 1/3 and this image is binarized, noise generation can be reduced. It is also easy to deform.

Here, the number of times m or n is determined uniquely depending on the width of the character you want to extract. For example, if the character width is about 5 pixels, m or n
is about 3 times. An example of the processing results is shown in Fig. 8.
101 is the input image G1. 102 is the result of local maximum filter processing, 103 is the result of further local minimum filter processing of 102, and 104 is the image obtained by subtracting the input image G1 of 101 from 103. In this figure, the number of times m and n are set so that the processing is performed to extract a thin character area, which has the advantage that objects existing around the character are not extracted even though they are the same dark area.

When the above processing is applied to a target image with shading (unevenness in brightness) as shown in Figure 9, it is generally difficult to determine the binarization threshold level for extracting text (second (see figure).

However, if the above-mentioned method is used, the unevenness in brightness can be removed by extracting the contrast of the text portion as shown in FIG. 4, and the binarization for extracting the text becomes easy.

FIG. 10 shows three-dimensional displays of the input image, ■ the result of creating a background image, and ■ the result of subtracting the background image (envelope image) from the input image. Note that in this example, the characters are brighter than the background, and the background image is created using a local minimum value filter and local maximum value filter procedure, and a background image with convex parts of brightness removed is created.

Note that such a background image subtraction method is also effective for detecting a chip (a) or a protrusion in an object as shown in FIG.

Figure 1 is the input image, Figure 2 is the local minimum filter n times 1
The background image that has been subjected to the local minimum filter m times, and FIG. 3, is a difference image between FIGS. 1 and 2.

By the way, the above explanation was performed using a 3 x 3 pixel local filter, so if the width of one character is, for example, 5 pixels, it would be necessary to repeat the local filter process about 3 times, but 1
If the size of the local filter is 5×5 pixels, the local filter processing can be performed twice, and if the size of the local filter is 7×7 pixels, the local filter processing can be performed once. Therefore, the same effect can be obtained by repeatedly executing the process using a small local filter as described above, or by executing the process once using a filter whose size corresponds to the character width to be extracted. In this background image subtraction method, if there is no unevenness in brightness, the background image may be created using only a local maximum value or local minimum value filter, as shown in Figure 3, and accuracy is not required. If so, you can always just use a maximum or minimum filter.

Application to Detection of Scratches and Dirt> When detecting scratches and stains, there are white stains, black stains, etc., and they also come in various sizes.

If you use the above-mentioned method to detect dirt etc. from such an object, although it is possible to remove shading,
It is not possible to detect both white and black stains, and it cannot be applied to extremely large stains that cannot be detected by a preset number of local maximum value filter processing.On the other hand, if you try to detect them, the number of repetitions increases as described above. Processing time becomes enormous.

Therefore, when detecting scratches and dirt, even if the shape cannot be extracted accurately, it is important to know whether they are there or not, so it is possible to detect all small dirt and detect only the outline of large dirt. , processing time is also faster.

The present invention can also be applied to such cases. That is, as shown in FIG. 12, for an object having a small black stain 121, a small white stain 122, and a large black stain 123, for example, a small black stain (with a width of 5 pixels) is detected as a whole. The input image at this time is as shown in FIG. 1. If the local maximum value filtering process is executed three times, the input image becomes as shown in FIG. 2, and if the local minimum value filtering process is executed twice, it becomes as shown in FIG. 3. If we subtract the input image from this background image, we get 5 as shown in Figure 4.
All black dirt below the pixel can be detected, and other dirt can be detected.
Contours can be extracted (5), L.Therefore, by setting the number of repetitions of local maximum value filtering and local minimum value filtering to be different in this way, it is possible to extract stains larger than a predetermined size, white, and black images. It is also possible to detect the presence or absence of dirt on the surface.

Note that the above processing example is local maximum value filtering → local minimum value filtering → subtracting the input image from the background image, but conversely, local minimum value filtering → local maximum value filtering → subtracting the background from the input image A similar effect appears in the image drawing procedure.

Further, regarding the number of repetitions, it is sufficient that the maximum, a, and J = each number are different, and the difference is not limited to one time as described above.

Also, if hole filling processing is performed on the processing results shown in Fig. 5,
The shape of dirt can also be easily recognized.

<Application for extracting only arbitrary widths> In document images, etc., the width of five characters is mixed, including thick and thin widths. In order to extract only bold characters, for example, from such an object, I think it is common to take the following steps.

(1) Binarization: Extract all characters clearly.

(2) Determine the area of each character. (3) Measure the width of the character within the area using a feature amount or the like.

(4) If the line width satisfies a predetermined value, it is extracted.

When such processing is performed, the processing time becomes enormous as the number of characters increases. Therefore, by applying the background image difference method described above, it becomes possible to extract a specific line width relatively easily. In other words, in the above process, the character width extracted is limited depending on the number of times the local filter processes or the size of the local filter, but if you use this to extract only thick characters, first Execute the background image difference method as many times as the local filter can extract everything below the bold characters. Next, the background image subtraction method is executed using the number of local filter processes that allow only thin characters to be extracted, and the thin characters and smaller characters are extracted. By obtaining a difference image between the two, it is possible to extract a contrast image of only bold characters.

Specifically, using FIG. 13, when extracting only the thick characters 132 from an image containing a mixture of thin characters 131 with a line width of 3 and thick characters 132 with a line width of 9 pixels, first,
In order to extract all thick and a-sized characters, we perform the local maximum value filtering process five times and calculate the difference B between the input image g^ and find that the characters are thin.

A contrast image (Figure 3) of bold characters can be created.

On the other hand, in order to extract only thin characters, a station (Fig. 4) is created. By finding the difference C between this image and the input image g^, a contrast image of only thin characters (Fig. 5) can be created. Furthermore, by calculating the difference between these two contrast images B and C, a contrast image of only bold characters like D (FIG. 6) can be created. If this image is binarized, only the bold characters can be clearly extracted.

In addition, although the above method was calculated using the difference between two contrast images, another method is to binarize the image in Figure 3 with a certain value, further binarize the image in Figure 5, and calculate the difference between these two contrast images. By performing exclusive OR processing on the value images, only bold characters can be similarly extracted.

Extraction of dark white characters and black characters> Character recognition targets generally have black characters written on white paper, but for example, a car license plate has green characters on a white background, as shown in Figure 14. There are various colors such as (a) and white letters on a green background (b). When using the background image subtraction method for such a target, if the color of the license plate is known in advance, it is determined which local filtering process to perform first. When extracting characters, it is necessary to determine each time whether the characters to be extracted are darker or brighter than the background. As a process for this purpose,
This will be explained using FIG. 16.

The area of the license plate can be
0-168485) (161), the density histogram of the license plate area becomes the pattern shown in FIG. 15 (a) or (b). In other words, since the background of the license plate has a larger area than the letters,
If the license plate is shown in (a) in Figure 14, it is shown in (a) in Figure 15.
The frequency is higher for brighter densities as in a). On the other hand, if the license plate is as shown in Figure 14(b), it is as shown in Figure 15(
This becomes the pattern b). Therefore, by obtaining the density histogram, the character color can be determined (161).

Using this determination result, it is only necessary to decide which filter processing to perform first depending on whether the characters are bright or dark compared to the background.In other words, if the characters are bright compared to the background, the local minimum value filter process as shown in 164 in FIG. If the text is dark, the local maximum value filter is processed first (165). By adding text color determination in this way, it is possible to extract both bright and dark characters. Become. In addition, a simple method other than that shown in Fig. 16 is to perform both processing when it is always bright and when it is dark, and then use the feature quantity (
For example, there is also a method of making a determination using (area per perimeter, etc.) and using the processed image on the side where the characters are present for subsequent processing.

By the way, a method for extracting dark characters and bright characters from an object in which both characters are mixed will be described below.

Figure 17 shows one line component of the input image with g^.
Create image B in the figure. On the other hand, an input image is created, and the image C in FIG. 5 is created from the difference between this and the input image. Inter-image calculation (B+C) that calculates the sum of these two contrast images B and C, inter-image calculation MAX (B, C) that calculates the maximum value of each pixel of the two images, and image that calculates the difference and takes the absolute value. When performing one of the inter-interval operations IB-Cl, if the number of dark characters (j) is set to be different for g^ and f^,
It is also possible to set the width from which dark characters are extracted and the width from which bright characters are extracted to be different. Furthermore, as described in the section on specific line width extraction, there is also a method of first binarizing each contrast image B and C and then performing exclusive OR (XOR) processing. In any case, by executing the background image subtraction method to extract bright or dark characters, and performing inter-image calculations on the images obtained from both, the contrast of the characters can be improved without worrying about white or black. Images can be created.

Furthermore, there is also a method shown in FIG. 18 as another method. In other words, if one line component of the input image is g^, it is processed in the order of local maximum value filter → local minimum value g^, and conversely, the background image in Figure 3 is bright. A contrast image of both text and dark text can be created, and the same effect as the method described above can be obtained.

Processing that takes connectivity into consideration> The method described above uses a two-dimensional 3x3 pixel filter and n
Although the filtering process for xn pixels is simply repeated, in reality, character widths can be extracted more clearly if connectivity is taken into account. Therefore, the following processing is performed. In the case of 3 x 3 pixels, there are 4-connections and 8-connections that are commonly known in binary image processing, so the positions shown in Figure 19 are not used (Do not care). (a) for 4 concatenation, i.e. +aax(f+) is i=1.2,4,6
．． (b) is for 8-connection (using all 3x3 pixels mentioned above), when performing the local filter processing described above, for example, when repeating the local maximum value filter, 4 You can repeat the process by changing the connectivity sequentially, such as connection → 8 connections → 4 connections (or vice versa).Also, with a 5 x 5 pixel filter, a filter that uses only O marks as shown in Figure 20 can be used. You can repeat it.

Generally, if the filter has n×n pixels, a circular filter shape will provide a good image when creating a background image.

A method for creating a background image with a single filter process
Although this method involves repeating each local minimum value filter several times, the same effect can be obtained by executing one raster scan by creating the following filter.

For example, in order to perform local maximum filter processing and local minimum filter processing using a 3×3 pixel filter, the filter shown in FIG. 21 is created. The central 3x3 pixel filter is f^, and the 3x3 pixel filter around it (around center pixel C) is fB.

Assuming fc + fn r fe, by executing the following calculation, a total of two raster scans can be performed in one raster scan: local maximum value filter → local minimum value filter processing, once each.

That is, a=MAX(f^(+))i=1~9b
=MAX(f B(+)) i = '-~9c=
MAX(fc(*))i=1~9 d=MAX(f-(t))i=1~9 a=MAX(f e(t)) i=1~9f
=Min (b, C1d, e) g=Min (a+f) If this g is used as the filter output in FIG. 21 and is scanned by rask scanning, it becomes equivalent to the process of performing local maximum filter → local minimum filter with a 3×3 pixel filter.

If the local maximum value filter is processed twice 9 If the local minimum value filter is processed twice, the basic filter is set to 5 x 5 pixels, and the second
All you need to do is create a filter like the one shown in Figure 1 and perform the same calculations as above. In this way, the background image difference method can be done in one go by changing the size of the filter within the range allowed by hardware constraints. Can be processed by scanning.

Addition of smoothing processing> By the way, when a background image is created and the difference from the input image is calculated, a minute noise component is actually generated. Such noise may not be a problem depending on the processing target, but it may become a problem when inspecting minute scratches. The cause of this noise is that the local maximum value filter and local minimum value filter processing lose the continuity of brightness with adjacent pixels, so when the difference is taken from the input image, the position of the pixel that has lost the continuity This is because bright values occur. Such a situation can be reduced by processing that takes connectivity such as the 4-connection or 8-connection mentioned above into account, but in order to further remove noise, smoothing processing can be added, that is, the first As shown in FIG. 22, a product-sum operation circuit 120 for performing smoothing processing is provided (the symbol in FIG.
(The same symbols as in the figure indicate the same processing circuits), the background image described above is subjected to smoothing processing (for example, using the filter shown in FIG. 23) to smooth out the discontinuous pixels. If the difference between this image and the input image is taken, the minute noise is less likely to occur.

In addition, Fig. 24 and Fig. 25 show processing when the input image itself has a lot of noise, but when the background image subtraction method described above is executed to extract the brightness concavity, the local maximum value filter If the image is run one time, the result will be as shown in FIG. 24 (2), if it is executed n times, the result will be as shown in FIG. 3, and if the difference between this image and the input image is taken, the result will be as shown in FIG. 4. Note that in this example, there is no gradient in brightness, so the local minimum value filter is not executed.As shown in Figure 1, in contrast images, the noise components of the input image are also extracted as bright values. The threshold value needs to be set higher than this noise component. In order to perform this without being aware of this, first perform local minimum filter processing several times on the input image (1) in Figure 24, as shown in (1) in Figure 25, and then If a local maximum value filter is executed for
Contrast images can be created. On the other hand, if you want to extract a convex part of brightness, you can first perform the local maximum value filtering process several times and then repeat the same process, or you can perform the above-mentioned smoothing process on the input image and then apply the background image subtraction method. You can get the same effect by running it.

〔Effect of the invention〕

According to the present invention, a target image can be extracted quickly and clearly even in a complex image or in an environment where brightness unevenness changes from moment to moment.

Therefore, it is effective as a pre-processing for images that are converted from grayscale images to binary images, and can also be processed without making any special adjustments to lighting, etc., so it can be used for character recognition, shape identification, and even scratch inspection. It can be applied not only to this field but also to many other fields.

[Brief explanation of the drawing]

1 and 3 are image processing diagrams for explaining the present invention in detail, FIG. 2 is a diagram for explaining the prior art, FIG. 4 is a diagram for explaining local maximum value and minimum value filters, and FIG. 5 is a diagram for explaining the present invention. A diagram showing the configuration of an embodiment, FIG. 6 is an explanatory diagram of a two-dimensional local maximum value filter, FIG. 7 is a flowchart showing one processing procedure of the present invention, and FIG. 8 shows the processing results of the present invention. Figure, Figure 9, 1st
Figure 0. FIG. 11 is a diagram showing the effects of the present invention, FIG. 12 is a diagram showing an application example of dirt detection, FIG. 13 is a diagram showing an application example of document images, and FIG. 14 is a diagram showing the types of license plates. 1st
Figure 5 is a diagram showing the density histogram of a license plate.
Figure 16 is a flowchart for extracting characters from a license plate, Figures 17 and 18 are image processing diagrams showing an overview of extracting white and black artists, and Figures 19 and 20 are filters that take connectivity into account. 21 is a diagram showing the shape of a filter that creates a background image with one filter scan. FIG. 22 is a diagram showing the configuration of another embodiment of the present invention.
FIG. 3 is a diagram showing smoothing, and FIGS. 24 and 25 are image processing diagrams according to another embodiment of the present invention. 12... Image memory, 13... Local maximum value filter circuit, 14... Local minimum value filter circuit, 15...
Inter-image calculation circuit, 16...binarization circuit, 17...C
P'U, 19...Monitor, 120...Smoothing circuit.

Claims (1)

[Claims] In an apparatus for inputting a processing target image having one or several gradations of brightness and extracting a target image included in the processing target image, a background image is extracted from the input processing target image. A method for processing a grayscale image, characterized in that the target image is extracted by performing an inter-image calculation between the background image and the image to be processed. 2. In claim 1, the background image is
A method for processing a grayscale image, characterized in that, when the image to be processed is input, a concavity of brightness included in the target image is created by filling it. 3. In claim 1, the background image is
A method for processing a grayscale image, characterized in that when the image to be processed is input, a convex portion of brightness included in the image to be processed is removed. 4. In claim 1, the background image is
A method for processing a gray scale image, characterized in that when the image to be processed is input, the brightness of the target image is smoothed and created. 5. In claim 1, the background image is created by performing a local maximum value filtering process that outputs the maximum value among m×n pixels to fill in the concavities of brightness in the target image. , a local minimum value filtering process that outputs the minimum value among m×n pixels to remove convex portions of brightness in the target image. 6. In claim 5, after performing the local maximum value filtering M times, minimum value filtering is performed on the processed image N times, and the minimum value filtering is performed M times. A method for processing a grayscale image, characterized in that the processed image is then subjected to maximum value filtering N times. 7. A method for processing a grayscale image according to claim 6, wherein the M and N repetitions are executed with connectivity. 8. In claim 5, a first intermediate image is created by an inter-image calculation between an image in which at least the concave portions of the brightness are filled and the image to be processed, and at least the convex portions of the brightness are removed. A second intermediate image is created by an inter-image calculation between the image and the processing target image, and the target image is extracted by an inter-image calculation between the first intermediate image and the second intermediate image. How to process grayscale images. 9. In a method for inputting a processing target image having several gradations of brightness and extracting a target image included in this processing target image, the first and second processing target images are extracted from the input processing target image.
A background image is created, a first target image is extracted by an inter-image calculation between the first background image and the processing target image, and a first target image is extracted by an inter-image calculation between the second background image and the processing target image. A method for processing a grayscale image, characterized in that a second target image is extracted, and a third target image is extracted by inter-image calculation of the first and second target images. 10. In claim 9, the first background image is created by executing a local maximum value filtering process that outputs the maximum value among at least m×n pixels, and the second background image is , a method for processing a grayscale image, characterized in that the image is created by executing local minimum value filter processing that outputs the minimum value among at least m×n pixels. 11. Input the image to be processed having several gradations of brightness,
In the method for extracting the target image included in this processing target image, the brightness of the target image area and other areas is compared, and if the target image area is darker, the input processing target Create a first background image from the image,
When the target image area is brighter, a second background image is created from the input image to be processed, and an inter-image calculation is performed between the created first or second background image and the image to be processed, A method for processing a grayscale image, characterized by extracting the target image. 12. Claim 11, wherein the first background image is created by filling in dark areas in the image to be processed, and the second background image is created by filling in dark areas in the image to be processed. A method for processing a grayscale image, characterized in that it is created by removing the image. 13. In an apparatus for extracting an image of a target from a predetermined area, a camera for capturing an image of the predetermined area, a conversion means for converting an image signal from the camera into grayscale image data, and inputting the grayscale image data; It is characterized by comprising a background image creation means for creating background image data of the predetermined area, and an inter-image calculation means for comparing and calculating the background image data and the grayscale image data to extract an image of the object. A processing device for gray scale images. 14. In claim 1, the background image extraction means includes a local maximum value filter circuit that outputs the maximum value among m×n pixels, and a local minimum value filter circuit that outputs the minimum value among m×n pixels. 1. A grayscale image processing device comprising a filter circuit.