JPH1063779A - Character recognition method, and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recognize character from a picture including an inverted character in a character recognizing device with high precision. SOLUTION: A binarized picture 3 is generated by extracting and binarizing the contour of a picture from an original picture 1 including an inverted character 2 by a binarized picture generating means 30 and an inner part of the contour is painted out by setting pixel values of pixels in a closed area inside the contour of the character whose contour is extracted to be equal to the pixel value of the contour by a painting-out means 35. Then, the character is recognized by performing pattern matching using a template 7 for a binarized picture 3' in which inner part of its contour is painted out by a matching means 40.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for recognizing characters in an image including characters, and more particularly to a method and an apparatus for recognizing characters by pattern matching using a template. It is.

[0002]

2. Description of the Related Art Conventionally, a subject such as a human body is subjected to CT, MR, or the like.
The image data of a medical image obtained by photographing with I or the like is subjected to appropriate image processing for each image data, and then the image is reproduced and recorded. Based on these image data, the image of the subject is Is a recording material such as a photosensitive material, C
An image recording / reproducing system for outputting a visible image to an RT or the like has already been put to practical use.

Normally, medical images are provided with ID information such as a patient name, an ID number, an imaging date, a date of birth, and the like. These ID information generally include English characters, numerals, symbols, and the like (hereinafter, “characters”). ). Therefore, a pattern and a character are mixed in the medical image. There is a case where it is desired to automatically identify only characters from a medical image in which pictures and characters are mixed. This is the case, for example, as follows.

Normally, in an image recording / reproducing system,
The captured image data is stored in the filing device in association with ID information serving as search information. The association between the image data and the ID information at the time of filing the image data to the filing device is performed by an external input, and is a burden on the operator side. Therefore, it is desired to be able to automatically recognize the ID information added to the image at the time of shooting.

Conventionally, as a method for recognizing a character from an image including the character, there is known a method called template matching in which a template of a character to be recognized is prepared and compared with an image to perform recognition. In this method, the recognition target area on the image and all templates are sequentially matched, and the template with the best matching result is determined as the recognition result. If no recognition target area is specified, the recognition on the image is performed. It is necessary to match the entire image by shifting the target area sequentially.

Hereinafter, a general recognition method using template matching will be described. When the image pattern f and the template pattern g have a size and a positional relationship as shown in FIG. 4, respectively, the similarity between the image pattern f and the template pattern g at the point (m, n) of the image is evaluated as follows. Expressed by a function.

[0007]

(Equation 1)

Here, usually, R (m, n) is called a cross-correlation coefficient, and a region D has a range of (i, j) satisfying | im | <M / 2, | jn | <N / 2. It is. By moving the template on the image, the similarity between the image and the template is sequentially determined based on the cross-correlation coefficient R described above. A larger value of the cross-correlation coefficient R indicates that a figure closer to the template exists around the image (m, n), that is, the image and the template have a stronger correlation, and the image and the template are completely matched. Takes a value of 1 and a value of 0 when there is no match. Normally, the one having the largest cross-correlation coefficient R is regarded as the recognition result.

However, the above-described template matching method requires the calculation of the square or the square root as shown in the above equation, and usually requires matching for a multi-tone image. However, there is a drawback that recognition takes a long time.

Therefore, for high-speed recognition processing, two images are used.
A character recognition device that performs binarization and performs matching on a binarized image. Furthermore, in order to accurately recognize a portion where a character overlaps a picture, a binarized image extracted from the outline of the image is used. A character recognition device that performs matching has already been proposed by the present applicant (Japanese Patent Application No. 8-167302).

[0011]

SUMMARY OF THE INVENTION
The accuracy of the recognition is improved to some extent by the matching in the binarized image. However, when the image 1 including the inverted character 2 as shown in FIG. 2 is subjected to outline extraction binarization, the edge 4 remains inside the outline of the character in the binarized image 3. For this reason, there is a drawback that the template and the matching are not perfect, and the evaluation value of the above-mentioned evaluation function is lowered.
This decrease in the evaluation value can cause erroneous recognition.

In view of the above circumstances, it is an object of the present invention to provide a character recognition method and apparatus for recognizing characters with high accuracy even from an image including inverted characters.

[0013]

A character recognition method according to the present invention is a character recognition method for recognizing a character from an image including the character, and generates a binary image by extracting an outline of the image from the image. Then, a closed region surrounded by the outline is set to the same value as the pixel value of the outline, a dictionary having a template of a binarized character is prepared, and the character of the template is used for the binarized image. And performs pattern matching to recognize characters.

A character recognition apparatus according to the present invention is a character recognition apparatus for recognizing a character from an image including the character, and generates a binary image by extracting a contour of the image from the image. Means, means for making the pixel value of a closed region surrounded by the contour the same as the pixel value of the contour, a dictionary having a binarized character template, and A matching unit that recognizes characters by performing pattern matching using a template.

By generating a binarized image by extracting the outline of the image, characters in the image are extracted as characters generally called hollow characters or bag characters. Therefore, the “closed region surrounded by the outline” mainly refers to a hollow portion of a hollow character. However, for example, a contour (inner contour) having a further closed area inside a closed area surrounded by an outer contour, such as hollow characters “A” and “B” shown in the binarized image 3 in FIG. In the case of a certain character, the closed area surrounded by the internal outline also corresponds to the “closed area surrounded by the outline”. Therefore, for these characters, it is desirable to perform matching using a template having the same value as the pixel value of the outline not only of the hollow portion of the normal hollow character but also of a closed region surrounded by the internal outline.

The phrase "the closed area has the same value as the pixel value of the contour" means, for example, that the image is binarized into "0" and "1" and the pixel value of the contour is "1". In this case, it means that the pixel value of the closed area is “1”.

That is, in the character recognition method and apparatus according to the present invention, the outline character is extracted at the time of binarizing the outline of the image by setting the outline portion to the same pixel value as the pixel value of the outline portion. This eliminates a matching error due to an edge that may occur inside the symbol and improves the matching accuracy.

As the method of "binarization by extracting the outline of an image", for example, the binarization method disclosed in Japanese Patent Application No. 8-167302 may be used. Other examples include a method using a first-order differential filter (Roberts filter, Brevit filter, Sobel filter, or the like), a Laplacian filter, or the like.

Here, the image means a mixture of characters and pictures.

[0020]

According to the character recognition method and apparatus of the present invention,
In a binarized image obtained by extracting and binarizing an outline of an image, a closed region surrounded by the extracted outline is subjected to pattern matching using a character template as the same pixel value as the pixel value of the outline. When a character or the like is subjected to contour extraction binarization, it is possible to prevent a decrease in a matching evaluation value due to an edge that may occur in a hollow portion of a hollow character, thereby performing highly accurate character recognition with reduced erroneous recognition.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, specific embodiments of the character recognition method and apparatus of the present invention will be described with reference to the drawings.

As shown in FIG. 1, the character recognition device of the present invention
Reference numeral 10 denotes a character area recognizing means 20 for recognizing a character area of the original image based on the original image data S input from the various photographing devices 9.
A binarized image generating means 30 for extracting the outline of the image of the character area recognized by the character area recognizing means 20 and binarizing the image to generate a binarized image;
A filling means 35 for setting the pixel value of the closed region surrounded by the contour extracted in the binarized image generated by the pixel 30 to the same value as the pixel value of the edge portion; Matching means 40 for recognizing characters by performing pattern matching using template 7
Consisting of

More specifically, the character area recognizing means 20 includes an image structure determining means 22. The image structure determining means 22 divides the original image into a plurality of areas and generates data representing the image structure of each area. It comprises a creating means 24 and a judging means 26 for determining whether each area is a character area based on the data created by the data creating means 24.

The binarized image generating means 30 includes, for each pixel of the original image, a calculating means 32 for calculating a difference between a maximum value and a minimum value of a pixel density in a predetermined neighborhood of the pixel, and the calculating means 32 A binarizing means 34 for binarizing each pixel by comparing the obtained density difference with a predetermined threshold to extract the outline of the image;

The operation of the character recognition device will be described below.

Image data carrying an original image photographed by the various photographing devices 9 is digitized and input to the character recognition device 10 as original image data S. Character recognition device 10
In, first, a character area is recognized by the character area recognition means 20. For example, Japanese Patent Application No. 8
The image determination can be performed using an image determination device described in JP-A-179344.

The image is divided into a plurality of areas, and
The data creating means 24 in the area 22 creates histogram data representing the image configuration of each area.

Generally, the histogram data is a graph in which the frequency of the intermediate density area is large and the frequency of the low density and the high density are small for the picture area, and the frequency of the intermediate density area is small and the frequency of the low density and the high density are large for the character area. Becomes a graph with. Note that an area in which characters and patterns are combined (hereinafter, referred to as a mixed area) is a graph having frequency as a whole. Therefore, the determining means 24 determines whether each area is an area including a character (a character area or a mixed area) based on the histogram data of each area. By previously identifying the area including the character, the matching range can be limited, and the processing can be speeded up.

The binarized image generation unit 30 performs binarization on an image of an area (hereinafter, referred to as a character addition area) determined to be an area including a character by the character area determination unit 20. The image contour extraction binarization is performed using, for example, a method disclosed in Japanese Patent Application No. 8-167302.

For the image data of the character providing area, first, the difference between the maximum value and the minimum value of the density in a predetermined vicinity range of each pixel of the image is calculated by the calculating means 32. For example, when the target pixel is located at the lower left of the 2 × 2 pixel mask, the maximum value and the minimum value of the gradation density (original image data S) at a total of 4 pixels in the 2 × 2 pixel mask (predetermined neighborhood range) , And this is used as the density difference of the pixel of interest. Naturally, the range of 2 × 2 is an example, and it is not necessary to limit to this. Next, in the binarizing means 34, the arithmetic means 32
The target pixel is binarized by comparing the density difference obtained by the above with a predetermined threshold value, for example, setting “1” when the density difference is larger than the threshold value and “0” when the density difference is smaller than the threshold value. By performing this binarization for each pixel in the image, a portion where the density difference between neighboring pixels in the image is large, that is, the outline of the image is extracted, and a binarized image is generated. Usually, since the density difference between the character and the surrounding image is large, the outline of the character is extracted. In the case of including the inverted character, for example, a binary image 3 is generated from the original image character area 1 including the inverted character E as shown in FIG. The inversion boundary line is extracted as the edge 4 in the hollow portion of “E”.

In addition, as a contour extraction binarization process, the signal (density) difference (S (i, j) −
S (i + m, j + n)), and the absolute value of this signal difference (| S (i, j)
−S (i + m, j + n) |) (max | S (i, j) −S (i + m,
j + n) |) is regarded as a signal difference in the target pixel, and when the density difference is larger than a predetermined threshold (Th), the pixel value of the target pixel is “1”, and the density difference is equal to or smaller than the predetermined threshold. If
"0", that is, S '(i, j) = max (| S (i, j) -S (i + m, j + n) |), (m = -1,0,
+ 1, n = -1,0,1) if S '(i, j)> Th, S "(i, j) = 1 else S" (i, j) = 0 Good.

Next, in the filling means 35, the pixel value of each pixel of the portion (closed area) surrounded by the contour extracted in the binarized image 3 of FIG. 1)). That is, the outline character “E” in the binarized image whose outline is extracted and becomes the outline character
Is defined as a binarized image 3 ′ in which the inside of the outline (the hollow portion) is painted out. As a result, the edge 4 inside the contour generated at the time of contour extraction disappears.

Thereafter, pattern matching is performed by the matching means 40 using the template 7 of characters of various sizes and sizes of the dictionary 45.

Here, since the matching is performed on the binarized image, a similarity evaluation function R 'based on exclusive product-sum (exclusive or) can be used. The calculation time can be shortened as compared with. Note that R ′ (m, n) of the image pattern f and the template g at the point (m, n) on the image is

[0035]

(Equation 2)

In contrast to the case of the aforementioned cross-correlation coefficient R, the template having the smallest R ′ is determined as the recognition result.

In the above-mentioned binarization, "1",
Instead of “0”, the sum may be binarized to “1” or “−1” and a simple sum of products value may be used for the evaluation function. The evaluation function R "(m, n) in that case is

[0038]

(Equation 3)

And the template having the largest R ″ is the recognition result.

Here, the area D is | im | <M / 2, | jn |
<N / 2 is satisfied (i, j).

As described above, in the character recognition device of the present invention, since the matching is performed after the inside of the outline of the character extracted by the outline extraction binarization of the image is filled, the character is extracted at the time of outline extraction binarization. The matching evaluation value does not decrease due to the edge generated inside the contour.

It should be noted that there are cases where a closed region surrounded by an inner contour exists within a closed region surrounded by an outer contour, such as a hollow character such as A, B, D, etc. FIG. Is shown. As shown in FIG. 3, the original image 1 including the inverted character A is
The binarized image 3 is obtained by binarizing the outline of the image.
At this time, the outline 5a outside the hollow character "A", the outline 5b inside the outline 5a, and the inverted boundary edge 4 are respectively extracted. Thereafter, when the closed area surrounded by the outline is painted out by the painting means 35, the area surrounded by the outer outline 5a and the inner outline 5b, that is, the hollow part of the hollow character, is originally filled only with this part. Ideal. However, if the area surrounded by the internal outline 5b is also recognized as the “closed area surrounded by the outline” by the filling means 35, this part is also filled and becomes the binarized image 3 ′. If matching is performed on such a binarized image 3 'using a normal character template 7, the matching evaluation value becomes low, and the matching accuracy cannot be improved. Therefore, a template of a character in which an area surrounded by an internal contour is filled like a template 7 'is prepared in advance, and matching is performed using the template. This allows
Perfect matching is possible. However, what can be matched by also filling a closed area with such an internal contour is limited to plain characters such as alphabets, numbers, and hiragana, and is not suitable for recognition of complicated characters such as kanji.
However, in general, ID information used in medical images is limited to alphabets and numerals, so that the above method can be used.

As described above, the character recognition is performed from the character providing area of the image, and the recognized character information I such as the ID number is output to the filing device 50 together with the original image data S.

In this manner, the character information I as the recognition result can be automatically filed in association with medical image data such as CT and MRI, and can be used as search information of each image data.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a character recognition device of the present invention.

FIG. 2 is a view for explaining an example of matching in the character recognition device of the present invention.

FIG. 3 is a view for explaining another example of matching in the character recognition device of the present invention.

FIG. 4 is a diagram illustrating a matching method.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Original image 2 Inverted character 3 Outline extraction binarized image 3 'Closed area filling binarized image 4 Edge 5 Outline 6 Closed area filling character 7, 7' Template 9 Imaging device 10 Character recognition device 20 Character region recognition means 22 Image Configuration determining means 30 Binary image generating means 35 Filling means 40 Matching means 45 Dictionary 50 Filing device

Claims (2)

[Claims] 1. A character recognition method for recognizing a character from an image including the character, comprising: creating a binarized image by extracting an outline of the image from the image; A dictionary having a binarized character template is prepared with the same value as the pixel value of the contour, and pattern matching is performed on the binarized image using the character of the template to recognize the character. Character recognition method. 2. A character recognition device for recognizing a character from an image including the character, comprising: a binarized image generating unit configured to generate a binarized image obtained by extracting an outline of the image from the image; Means for making the pixel value of the enclosed closed area the same as the pixel value of the outline; a dictionary having a binarized character template; and pattern matching using the template for the binarized image. And a matching unit for recognizing a character by causing the character to be recognized.