JP2960468B2 - Method and apparatus for binarizing grayscale image - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to grayscale image processing, and more particularly to a grayscale image processing method and apparatus applicable to an image having a complicated background or uneven brightness. About.

[Conventional technology]

In the case of recognizing characters and the like using an image processing device, generally, image data taken by a television camera or the like is compressed at a certain threshold level into binary data of '0' and '1' for processing. Many. For example, the character is binarized to a level of “1” and the background is binarized to a level of “0”, and recognition processing is performed on the data of the “1”.

By the way, when a fairly clear image is to be processed, such as when recognizing a character written in black on white paper, the above-described threshold level can be easily determined in advance (for example, by determining the average density). Threshold). However, in order to cope with more difficult applications, good results are often not obtained by simple binarization processing as described above. As an example of this application, the following are fried.

(1) Characters are extracted from a patterned cardboard box.

(2) Characters are extracted from an outdoor signboard or the like.

(3) Characters are extracted from the printed circuit board.

Such a target cannot extract a character by simple binarization processing because the background of the character is complicated and the brightness varies greatly. Therefore, it is necessary to devise a binarization method for extracting characters and the like well in these objects. As a conventional example, for example, "Examination of a character extraction method from a scene image" (Information Processing Society of Japan (86.3; by Otani)).

As shown in FIG. 2, one screen of a gray-scale image 140 input from a television camera or the like is divided into a plurality of sub-blocks 141, and an optimum binarization threshold level θ _ij is determined within the blocks (see FIG. 2). FIG. 2 (2)). The threshold θ at this time
_ij is regarded as a two-class problem in which the black and white are separated into two classes in the subblock 141, and is set to a value that maximizes the inter-class variance. Further, in order to maintain continuity between the sub-blocks 141, interpolation is performed for each pixel using θ _ij as shown in FIG. 2 (3), and the resulting threshold values θ _{x, y} ,
That is, the threshold value θ _{x, y} for each pixel is determined, and the input image is binarized.

[Problems to be solved by the invention]

The problem with the prior art described above is that the former uses a density histogram (determining the frequency of each density level within the sub-block) to determine the threshold θ _ij within the sub-block.
Since the two-dimensional image data is converted to one-dimensional data, the position information of the brightness cannot be considered, and the optimum threshold cannot be determined.

Another problem is that the processing time for obtaining θ _{x, y} is enormous, and character extraction cannot be performed in real time.

By the way, in the case where the input image is differentiated, a portion which is not an edge is almost '0', and a portion of the edge is a grayscale image having a value corresponding to the edge strength. In this case, the above-described method does not allow optimum edge extraction.

The present invention has been made in view of such conventional drawbacks, and has an object to create a high-speed and clear binary image even for an image having complexity or uneven brightness.

[Means for solving the problem]

The basic processing of the present invention is a binarization method of an image in which the density is “0” in an area other than the object, such as the above-described differential image.
The value conversion method will be described later.

FIG. 1 (1) shows A- in order to simplify the explanation of the input image.
A one-dimensional image connecting A 'is shown in (2). An image obtained by differentiating this image (obtaining the absolute value of the differential value) is shown in (3).

A non-edge portion shows a value of almost '0', and only a portion of the edge shows a certain density value corresponding to the edge strength. At this time, if the brightness is uneven or the like, the edge strength varies in the partial portion, so that even if a certain fixed threshold value is set, all edges cannot be satisfactorily extracted. For example, if the threshold value is too high, a portion having a small edge strength cannot be extracted. Conversely, if the threshold value is too high, a portion having a high edge intensity will be extracted to be considerably thick.

For this reason, the present invention considers a curved surface passing through the intermediate density of a region (pixel) to be extracted as an ideal threshold value, and provides means for creating this surface, and extracts a region having a density value larger than the threshold value surface. By doing so, a good binary image is obtained.

[Action]

The threshold plane is a plane that passes through the intermediate value of the density of the region (pixel) to be extracted. However, since this is a two-dimensional plane, it is difficult to quickly and accurately find it.

Therefore, means for simulating the threshold plane at high speed is adopted. That is, in order to create a surface passing through the intermediate value of the density of the pixel of the edge to be extracted, an image H in which the density of the binarization target, here, the density of the differential image is 1 / n (normally n is 2). Create Since the peak density value of this image indicates the intermediate value of the density of each edge (when n is 2), this H image is further subjected to density expansion processing several times and further smoothed. An image similar to a typical threshold plane can be created. Then, a good binary image can be created by comparing the original differential image and the threshold image and extracting a portion where the density value is larger than the threshold image.

〔Example〕

 Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 shows the flow of processing, in which (1) shows an input image F, in which information of one line AA 'is shown as a position on the horizontal axis and brightness (density) on the vertical axis. Of the (2)
F _A. When the input image is differentiated (the output is an absolute value), the G image of (3) is created. This image shows a value of almost '0' other than the edge, and a value corresponding to the edge intensity at the edge position. But not to create a binary threshold value for the image, ideally, you may be creating a curve T _C through an intermediate concentration of the intensity of the edge (contrast). However, in practice since the input image is a two-dimensional plane, it is necessary to create a curved surface rather than a curve T _C.
This is because it is rather difficult to create a threshold curve similar to curve T _C In terms of one dimension.

Specifically, an image _{GH in} which the density of each pixel of the differential image G is halved is created as shown in (4). The peak value of the contrast of the image _GH indicates an intermediate value of the contrast of the original differential image. If the density is further expanded several times, the _GHX image of (5) is obtained. On the other hand, if the smoothing process is further performed, an image C similar to an ideal threshold can be created as shown in (6). Then, a comparison is made between the original differential image G and the created threshold image C, and binarization is performed to extract only pixels whose density of the differential image is larger than the threshold image, thereby forming a clear binary image. Can be created.

Here, the density expansion processing is what is called a local maximum value filter. On a two-dimensional plane, the density f ₁ , f _2, ... F ₉ of 3 × 3 pixels of the input image f as shown in FIG. The process of taking the maximum value and outputting that value as F is shown below.

F (i, j) = max [f (i + u, j + v)] (u, v) = − 1,0, + 1 In the smoothing processing, for example, an average value of the density of 3 × 3 pixels is output. There is.

FIG. 4 shows one configuration of the present invention. TV camera 10, A / D
Converter 11, image memory 12, differentiation circuit 13, image scaling circuit 14, local maximum value filter circuit 15, smoothing circuit 16, inter-image operation circuit 17, binarization circuit 18, CPU 19, D / A converter 20, Monitor 2
It's from one.

In such a configuration, an image signal photographed by the television camera 10 or the like is, for example, 128
The image data is converted into class density data and stored in the image memory 12. Here, the image memory 12 has, for example, k gray-scale memories G _{1 to} G _{k of} 256 × 256 pixels, and _one binary image memory B _{1 to} B _{1 as} needed.

The input image as stored in the image memory of the G _1, described in detail in accordance with the following procedure.

First, differential processing by the differential circuit 13 the image of G _1, and stores the G ₂ image memory. This image is the target image for the binarization processing. For this image G _2, to create an image where the brightness in the image scaling circuit 14 by 1 / n to store the G _3. Normally, this n is 2, but it may be 3 or 4 depending on the properties of the target image (of course, it may be a real value such as 1.5). For this G ₃ image, the local maximum value filter circuit 15
In stores to G ₄ performs expansion processing of gray. Since the maximum value of the local maximum filter circuit usually 3 × 3 pixel density which is less the amount of expansion in a single process, again stored in the G ₄ repeated several times expansion processing on the image of the G ₄ I do. The number of repetitions is such that the width of the edge is widened (about 5 times).
To Furthermore, the image G ₄ inflated, smoothed by the smoothing circuit 16, to create a smooth surface. It also performs a number of times, and stores the G _5.

The image G ₅ is a threshold image, binarization by comparing the concentration of each pixel of the image and the original image derivatives G. Specific processing, inter-image calculation circuit from differential image G the process of subtracting the concentration of the threshold image G ₅ as FIG. 1 (7)
Run at 17. The image obtained here is information of a differential image having a density higher than that of the threshold image.
Is set (about two or three gradations) and binarized by the binarization circuit 18. By the above processing, a binary image with the intermediate brightness value as a threshold value is obtained.

By the way, if there is a bright portion in the differential image as shown in FIG. That is, as shown in FIG.
When the image is expanded by the local maximum value filter, the threshold image becomes a high level as shown in FIG. 5 (3) due to the influence of noise. As a result, it becomes impossible to extract a portion around the noise. Would. Therefore, after creating an image with a brightness of 1/2 (or 1 / n), a smoothing process for reducing noise is performed several times, and then the image is expanded with a local maximum value filter. To By doing so, it is possible to prevent the threshold from fluctuating due to noise.

When the noise is small, the extraction can be performed well without performing the smoothing processing as described above. Therefore, the presence or absence of the smoothing may be determined in consideration of the processing time. Note that this noise countermeasure has the same effect even if the binarization target image (here, the differential image) is smoothed several times and the 1 / n scaling process is applied thereto.

The method described above determines which value of n, whether to create an image of 1 / n of brightness and smooth it as described above, or whether to create an image of 1 / n after smoothing There are various methods.

Of these, the results of all pre-processing are displayed in order to determine the method most suitable for a certain object, and the pre-processing that the user considers most optimal is adopted.

In character recognition or the like, the result of preprocessing is recognized, and the preprocessing with the highest recognition rate is adopted.

The user specifies a region to be extracted in advance, and adopts a pre-processing that results in a pre-processing result most similar to this information.

And so on.

As described above, the present embodiment is a technique capable of favorably determining a binarization threshold value for a target having a certain density value for a pixel to be extracted and a value of approximately '0' for the other pixels. Half-contrast binarization method). Therefore,
It cannot be applied to a target having a certain density value other than the region to be extracted, such as a normal grayscale image. Therefore, a method applicable to these will be described below as another embodiment of the present invention. FIGS. 6 (1) and (2) show a target including a character as a normal gray-scale image.

In order to binarize a normal grayscale image by the method in the above-described embodiment, if the density of the background of the character is converted to the '0' level, only the contrast of the character remains, so that the above-described half-contrast method can be applied. For this reason,
Convert using the method of -153682. This method creates a background image from an input image and subtracts the background image from the input image to create an image containing only density information to be extracted (for example, a character image for a character image).
As a result, it is possible to create an image in which the character part has a density corresponding to the contrast and the other background is almost '0',
Since an image similar to the differential image described above can be created, the binarization method of the present invention is effective. This background removal processing will be briefly described. FIG. 6A shows an input image, and FIG. 6B shows the density information of one line AA '. The figure shows extracting black characters. When the brightness expansion processing is performed several times using the above-described local maximum value filter to create a background image, the concave portion of the character portion is filled with the surrounding brightness as shown in FIG. 6 (3). If this image is further contracted several times by a local minimum value filter in the same manner as the dilation process, a background image can be created as shown in FIG. 6 (4). Here, the local minimum value filter processing is the inverse processing of the local maximum value filter, and the density f of the local area is
The minimum value of (x, y) is defined as the output F (x, y) and is represented by the following equation.

F (i, j) = min [f (i + u, j + v)] (u, v) = − 1,0, + 1 After the background image is created, the difference in brightness between the original input image and the background image is obtained. As shown in FIG. 6 (5), a contrast image in which the background density is set to '0' can be created. This method is called a background subtraction method.

This image can remove the uneven brightness of the background, but cannot correct the difference in the contrast of characters, so that it cannot be binarized with a fixed threshold. Therefore, by applying the half-contrast binarization method to this image, a good binary image can be created.

Although the above processing has been described for black characters, in the case of white, the inversion processing of the input image is performed and then the above-described background difference method is applied, or the convex part of the character part is cut in the order of the local minimum value filter and the local maximum value filter. The same effect can be obtained by performing the processing.

Note that the background difference method can be easily realized by adding a local minimum value filter circuit to the configuration of FIG.

〔The invention's effect〕

As in the present invention, a threshold image is created from an image in which the brightness is reduced to 1 / n for an image in which the density of an area to be extracted has a certain value and the density of the other area is almost '0' This makes it possible to set an optimal binarization threshold value for each pixel, so that a clear binary image can be created.

[Brief description of the drawings]

FIG. 1 is a diagram showing a flow of an embodiment of the present invention, FIG. 2 is an explanatory diagram of a conventional example, FIG. 3 is a diagram for explaining a local maximum value filter, and FIG. FIG. 5 is a diagram showing a problem when noise occurs, and FIG. 6 is a diagram showing a flow of creating a contrast image from an input image. 13 …… Differentiation circuit, 14 …… Image scaling circuit, 15 ……
Local maximum value filter circuit, 16: smoothing circuit, 17: inter-image calculation circuit, 18: binarization circuit, 19: CPU.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-62-119896 (JP, A) JP-A-61-214078 (JP, A) JP-A-63-292284 (JP, A) JP-A-1- 244587 (JP, A) JP-A 1-287783 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06K 9/00-9/82 G06T 1/00 G06T 5/00

Claims (5)

(57) [Claims] 1. A method for binarizing a grayscale image obtained by scanning a two-dimensional image, wherein the grayscale image to be binarized has a brightness of 1 / n (n is an integer of 2 or more). And performing an expansion process on the image having the brightness of 1 / n (n is an integer of 2 or more), performing a smoothing process on the image subjected to the expansion process, creating a threshold image, and performing the binarization. Calculating a difference in brightness of each pixel between the gray image and the threshold image, which are the object of the above, and binarizing the obtained difference image with a predetermined threshold value Method. 2. A method for binarizing a grayscale image obtained by scanning a two-dimensional image, wherein the grayscale image to be binarized is subjected to a smoothing process, and the smoothed image is processed. Is 1 / n (n is an integer of 2 or more)
An image having a brightness of 1 / n (n is an integer of 2 or more) is subjected to dilation processing, a smoothing process is performed on the dilated image to generate a threshold image, A gray-scale image obtained by calculating a difference in brightness of each pixel between a gray-scale image to be binarized and the threshold image, and binarizing the obtained difference image with a predetermined threshold value; Binarization method. 3. A method for binarizing a grayscale image obtained by scanning a two-dimensional image, wherein the grayscale image to be binarized has a brightness of 1 / n (n is an integer of 2 or more). Performing a smoothing process on the image having a brightness of 1 / n (n is an integer of 2 or more), performing an expansion process on the smoothed image, and performing a smoothing process on the expanded image. To generate a threshold image, determine a difference in brightness between each pixel between the grayscale image to be binarized and the threshold image, and determine the determined difference image by a predetermined threshold. A binarization method for a gray-scale image, wherein the binarization is performed using values. 4. The method according to claim 1, wherein the grayscale image to be binarized is a background difference image obtained by calculating a pixel-by-pixel difference between the grayscale image and the background image. Binarization method of the gray image to be processed. 5. A gray-scale image memory for storing a gray-scale image input from a TV camera or the like, and a brightness of the gray-scale image stored in the gray-scale image memory is 1 / n (n is an integer of 2 or more). An image scaling circuit that performs an expansion process on an image obtained by the scaling circuit; a smoothing circuit that performs a smoothing process on an image expanded by the expansion circuit; An inter-image operation circuit for performing a difference operation between the obtained image and the grayscale image, and a binarization circuit for converting the difference image obtained by the inter-image operation circuit into a binary value with a predetermined threshold value An apparatus for binarizing a grayscale image.