JP2000048120A - Method for extracting character area of gray level image and recording medium having recorded the program thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve precision in reading and recognizing characters by extracting and specifying only a part including a character from the image of a processing object so as to execute image processing and character collation only in a partial area. SOLUTION: Based on the density value of each divided partial area, a distributed quantity is calculated to judge the area to be a character block or a background block (S201). Each partial area is rotated to calculate a distribution value from the distribution of the projecting values of each direction to judge the area to be the character block or the background block (S202). Whether the partial area is the character block or the background block is totally judged by these discriminated result to finally establish the character area (S203). Only the character area is extracted by limiting to the partial area including a character specified from the image of the processing object. Otherwise, the shape of a character boundary where variation of density is steep is extracted (S204). Thus, character reading and recognition hardly influenced by density variation and a noise pattern existing in the background area of an original image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention is regarded as one of the character recognition techniques based on image processing, and separates a character area from a background area by paying attention to the density distribution of each divided partial area and its directionality. The present invention relates to a method for extracting a character area of a grayscale image for extracting only a character existing area and a recording medium on which a program for the method is recorded.

[0002] The present invention can be used as a preprocessing (automatic generation of a mask, etc.) of recognition and collation of engraved characters imprinted on metal parts or castings with rough surface processing accuracy.

[0003]

2. Description of the Related Art In the manufacturing process of industrial products, reading and recognizing characters is an indispensable technology for automating product identification and classification. Hereinafter, a conventional technique of reading a gray-scale image character will be described.

[0004] As a typical method, a template image serving as a reference is prepared in advance by the number of types of characters, and
There is template matching performed by specifying the best matching reference image as compared with these reference image groups. This method can be recognized with high accuracy if the character image to be processed is clear, but image differences due to noise in the background part prevent matching with the template image,
There is a problem that the recognition accuracy of the character is reduced.

In order to solve such a problem, a method of extracting a character boundary having a sharp change in density has conventionally been used in the past by a technique for removing background noise by binarization processing or various edge extraction techniques. .

[0006]

However, in the former binarization processing method, it is difficult to determine a threshold value for binarization unless the contrast between the character portion and the background portion is sufficient. In addition, the latter edge extraction method has a problem that a similar image of a character boundary is generated in a background portion other than a character boundary where the density change is steep due to the differential processing. Regardless of the method, the main cause that the accuracy of later recognition / matching process is reduced is that the noise image other than the character remains and the background noise is included in the matching target region in the matching with the template image. Has become.

FIG. 8 shows an example of a character image pattern of the inspection object. FIG. 8A shows an example of an ideal character pattern, but the actually obtained image is caused by subtle irregularities on the component surface and the lighting conditions as shown in FIGS. 8B and 8C. Depending on the processing condition of the part surface and the stamping position, even if it is a component line of the same character, the character part is brighter than the background area depending on the density unevenness and the background noise. An event occurs in which an image becomes mixed when it becomes dark and when it becomes dark, which makes it difficult to extract a character area based on a density change from a background area.

In the appearance inspection of industrial products, high-precision recognition of engraved characters is required even under such circumstances.
The density distribution and outline of the character image tend to be unstable, and it is not easy to actually obtain a clear image under the influence of the noise pattern in the background.

In view of the above-mentioned problems, the present invention specifies and extracts only a partial region including a line segment constituting a character from a grayscale image including a character, and improves the separation accuracy, thereby improving the separation accuracy. It is an object of the invention to improve the accuracy of character reading / recognition by removing a region in advance and performing various image processing and character matching on a limited partial region including a character.

[0010]

In order to achieve the above object, according to the present invention, a gray image to be processed is divided into small areas of a fixed size, and the density value of the partial area is determined for each partial area. A first determination process for specifying a character candidate partial region based on the variance calculated based on the above, and a rotation angle dependency of a projection value distribution indicating the directionality of density change for each rotation direction in the partial region A second determination process for specifying a character candidate partial region; a comprehensive determination process for determining a character region from both a determination result based on the first determination process and a determination result based on the second determination process. It is characterized by having.

One feature of the present invention is to focus on the image property that a character is composed of line segments, and to extract the directionality of line segments from an image by calculating the projection value. The method for separating and extracting a character area from an image to be processed is greatly different from the technique.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described in detail with reference to the drawings, taking as an example a case where the present invention is applied to extraction of a character area.

FIG. 1 shows an embodiment of a process for extracting only the existing region of the character line from the engraved character grayscale image of the industrial part using the character region extraction of the grayscale image according to the present invention. This embodiment has first to fourth processing steps. Here, a partial area including a character line is referred to as a character block, and a partial area that is regarded as uniform ignoring background and density unevenness is referred to as a background block.

In a first processing step 201, the variance is calculated based on the density value of each divided partial area, and the value of the variance is compared with 0 to determine whether the partial area is a character block or a background block. Is determined.

In a second processing step 202, rotation is performed for each partial area, and the above-mentioned variance value is calculated from the projection value distribution in each direction. Depending on the magnitude of the value, whether the partial area is a character block or a background block is determined. Is determined.

In a third processing step 203, a logical product or a logical sum of the determination result based on the first determination process and the determination result based on the second determination process is calculated, so that the partial region is a character region. Or the background area is comprehensively determined, and the character area is finally determined.

In the fourth processing step 204, only the character area is extracted by binarizing processing such as discriminant analysis, for example, by limiting to the partial area including the character specified from the image to be processed. Alternatively, a character boundary shape having a sharp density change is extracted using a so-called edge extraction technique.

FIG. 2 shows a detailed processing procedure of the present embodiment. 301 is a region division process, 302 is a variance amount calculation and a first block determination process, 303 is a rotation image generation process, 304 is a projection distribution calculation and a variance data calculation determination process, 305 is a second block determination process, and 306 is a character An area / background area comprehensive determination process, and 307, an edge extraction process. Hereinafter, each process according to the above embodiment will be specifically described.

First, the input gray-scale image G (i, j) is divided into partial regions of a predetermined size (for example, 16 × 16 pixels) in a region dividing process 301. FIG. 3 is a diagram illustrating a division mode. For the following description, the m-th partial area number in the horizontal direction and the n-th partial area number in the vertical direction of the image are described as (m, n). This dividing process is not necessarily performed by using FIG.
It is not necessary to perform equal division as shown in FIG.
As shown in (b) and (c), the regions may overlap each other or may be separated from each other.

Calculation of variance and first block determination processing 30
In 2, the average μ _mn and the standard deviation σ _mn of the density values of the pixels in each of the divided partial areas (m, n) are calculated, and the character block and the background are calculated from the average and the standard deviation σ _mn. Set a background separation straight line for discriminating between blocks. Using this separation line, it is determined whether the partial region belongs to a character block or a background block. That is, the variance amount v _mn is given by v _mn = σ _mn −c ₁ μ _mn −c ₂ where c ₁ is the slope of the separation line and c ₂ is the intercept, and the determination rule is as follows.

If v _mn > 0, the partial area (m, n) is a character block. If v _mn <0, the partial area (m, n) is a background block. Average point concentration, typical point sequence data plotting the standard deviation of concentration on the vertical axis,
In such a case, the character block and the background block can be distinguished from each other by the illustrated background separation straight line.

The rotated image generating process 303 is a process for creating an image rotated by a predetermined angle (for example, 10 degrees) by affine transformation in each partial area. This is pre-processing for obtaining the projection distribution, and from that purpose,
Since the rotation direction of 30 degrees and the rotation direction of 210 degrees are the same direction, for example, in the case of increments of 10 degrees, the direction of rotation from 0 degree is 1 degree.
It is sufficient to generate a total of 18 images up to the 70-degree direction.

Projection distribution calculation and distributed data calculation determination processing 3
In step 04, in each partial region, a vertical projection value (an average value obtained by integrating the density values of pixels on each scanning line) in each pixel column is calculated for each rotation direction, and the projection in that direction is calculated. Calculate the variance of the value data. In other words, the size of the partial region is P horizontal pixels × Q vertical pixels, and 180 degrees is D pixels.
When divided, the variance d of the projection value data in the rotation direction k
_k (k = 0, ..., D-1) is

[0024]

(Equation 1)

Is given by Here, p _i (i = 0,
.., P-1) represent projection values in the pixel row i. In the second block determination processing 305, the shared data d
_{From k} (k = 0,..., D-1), the variance value var ₁ or the difference variance (difference between the variance maximum value and the variance minimum value) var ₂ is calculated. It is determined whether there is linearity. In other words, var ₁ or var
_By comparing the value of ₂ with a predetermined threshold value, it is determined whether the partial area is a character block including a character to be detected or a background block that is not. Here, the variance value var ₁ and the difference variance var ₂ are given by the following equations, respectively.

[0026]

(Equation 2)

FIG. 5 conceptually illustrates the flow of the above-described processing 304 and processing 305 with D = 18. For the grayscale image shown in the upper right of the figure, a “partial region original image” as shown is extracted. Then, for the extracted partial area original image, as shown in FIG.
The image is rotated by 0 degrees at 0 degree, 10 degrees... 90 degrees... 160 degrees, and 170 degrees, and each is projected in the vertical direction to obtain a projection distribution. Then, the variances d ₀ and d of the projection values for each of the projection distributions
₁ ... D ₉ ... D ₁₆ and d ₁₇ are obtained. Needless to say, the variances d ₀ , d ₁ ... Are the above-described variance data d _k (k =
0,..., D-1).

Comprehensive judgment processing 306 of character area / background area
Calculates the logical product or logical sum of the result of the determination by the first block determination process and the result of the determination by the second block determination process, and finally divides the entire partial region into the character block region and the background block region. This is the process of separating into

The last edge extraction processing 307 is performed only for the partial area determined as the character block area by the above-mentioned comprehensive judgment processing. Here, for example, a typical Sobel operator is applied to the original image to calculate the differential intensity of the density value of the pixel, and the threshold value is processed to extract an edge serving as a character boundary. Other edge extraction techniques include the Prewitt operator and Canny's method of calculating the zero cross point of the second derivative of the density cross section, and these may be used.

An example of processing obtained by applying the present invention will be described below. FIG. 6A is an original image (size of 500 × 120 pixels) of an engraved character stamped on a casting, and FIG. 6B is an original image of the outer frame for all partial regions determined to be character blocks. 6C is a diagram superimposed and displayed, and FIG. 6C shows a threshold value of the differential intensity image calculated by applying the Sobel operator (threshold value is 25) for the character area, and extracts an edge. It is an image. Here, 12 × 12 pixels were given as the size of the partial region, and the variance value (var ₁ ) was used for each rotation angle of the projection value variance data in the second determination processing. As a comparison, FIG. 7A shows an image subjected to binarization processing by a discriminant analysis method widely used in the conventional representative technology, and FIG. 7A shows a differential intensity image calculated by applying a Sobel operator obtained by threshold processing. FIG. 7B shows the extracted edge extracted image. As is apparent from these figures, the method according to the present invention can extract a clear character shape with a limited character area.

The input image is divided into rectangular regions of a fixed size, and the projection value in the vertical direction is calculated by integrating the density values while changing the rotation angle in each partial region, and the projection values are compared with respect to the rotation direction angle. Then, it is determined whether or not the partial region includes a line segment image from the angle dependence of the projection value. By removing areas other than the area including the character from the original image based on the determination result of each partial area, the character image to be processed is influenced by the background pattern, density unevenness, and noise image from the character image to be processed. And only a partial area including a character can be separated and extracted.

In the above, the method for extracting the character area of the gray image has been described. However, the method for extracting the character area of the gray image can be stored in the form of a program that can be executed by the data processing apparatus. From this, the invention of the present application also targets a recording medium on which the program is recorded, and
It is set forth in the claims herein.

[0033]

As described above, according to the present invention,
By preserving the shape information of the character itself, extracting and specifying only the partial region including the character from the image to be processed, and performing image processing and character matching only on the partial region,
Character reading / recognition that is less affected by uneven density and noise patterns existing in the background area of the original image can be performed.

[Brief description of the drawings]

FIG. 1 is a diagram showing an embodiment of extracting a character region from a gray image of a stamp character according to the present invention.

FIG. 2 is a flowchart illustrating an example of a detailed processing procedure according to the embodiment;

FIG. 3 is a diagram showing an example of area division of an input image.

FIG. 4 is an example of typical point sequence data obtained by plotting the average of the density of each partial region on the horizontal axis and the standard deviation of the density on the horizontal axis for a certain character image.

FIG. 5 is a diagram showing a rough flow of the processing 304 and the processing 305 in 18 directions of 10 degrees.

FIG. 6A is an original image of an engraved character, and FIG.
Is a diagram showing an outer frame of a partial region determined as a character block, and FIG. (C) is an image of a processing result obtained by extracting edges by performing threshold processing on a differential intensity image calculated by applying a Sobel operator. is there.

FIG. 7A is a binary image of a processing result obtained by applying a widely used discriminant analysis method, and FIG. 7B is a diagram illustrating a Sobel operator as in the above-described processing 307. It is the image of the processing result which applied and performed the edge extraction.

FIG. 8 is a diagram showing variations of a target character image pattern.

[Explanation of symbols]

 201 first determination processing 202 second determination processing 203 comprehensive determination processing 204 character area / boundary shape extraction processing 301 area division processing 302 variance amount calculation and first block determination processing 303 rotated image generation processing 304 projection distribution calculation and variance Data calculation determination processing 305 Second block determination processing 306 Character area / background area comprehensive determination processing 307 Edge extraction processing

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Koichi Tanaka 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Japan Telegraph and Telephone Corporation (72) Inventor Masashi Okuhira 3-192-1, Nishi-Shinjuku, Shinjuku-ku, Tokyo No. Nippon Telegraph and Telephone Corporation F term (reference) 5B029 AA01 BB02 CC15 CC29 5C077 LL01 MP06 PP22 PP25 PP27 PP43 PP46 PP47 PP48 PP58 PP68 PQ22 5L096 AA06 BA17 FA32 FA33 FA44 GA19 GA22 GA23

Claims (6)

[Claims] 1. A grayscale image including a character image to be processed is divided into partial regions, and for each partial region, whether the partial region is a region including a character or a background region not including a character is determined. This is a method for extracting a gray-scale image character area by removing the background area from the entire image and extracting only the character area. The gray-scale image to be processed is divided into partial areas, and the density values of the pixels belonging to the partial area are calculated. A first determination process for identifying a character candidate partial region based on the average value and the standard deviation calculated for the partial change region; A second determination process for identifying a character candidate partial region by focusing on rotation angle dependency; and a character region extraction method for a shaded image. 2. The method for extracting a character area of a shaded image according to claim 1, wherein the determination result based on the first determination processing and / or the determination result based on the second determination processing are determined. A character area extracting method for a shaded image, comprising: at least a comprehensive determination process for determining a character area by taking a logical product or a logical sum. 3. The method according to claim 1, wherein the second determination process generates an image rotated by a predetermined angle for each partial region. Then, at each rotation angle, an average value is calculated by integrating the density values of the pixel rows in one direction to obtain a projection value.
A second step of calculating the variance of the projection value of each pixel row in the other direction distribution for each rotation angle in each partial area; and a second step of calculating the variance of each partial area in the second step. A third process of calculating the variance of the projected value variance values of the individual rotation angles, which is the difference between the variance value of the rotation angle or the difference between the variance maximum value and the variance minimum value, and the third process. The given variance value or difference variance is compared with a threshold value given in advance,
A fourth step of determining whether or not the partial area is a partial area candidate that includes a character to be detected. 4. A grayscale image including a character image to be processed is divided into partial regions, and for each partial region, whether the partial region is a region including a character or a background region not including a character is determined. A grayscale image to be processed is divided into partial areas on a recording medium which is recorded in the form of a character area extraction program for a grayscale image that removes a background area from the entire image and extracts only a character area. A first determination process for specifying a character candidate partial area based on the average value and the standard deviation calculated for the density values of pixels belonging to And a second determination process for specifying a character candidate region by focusing on the rotation angle dependency of the value with respect to the angle change, and a program for executing the data processing device by combining A recording medium on which a character area extraction program is recorded. 5. A shaded image including a character image to be processed is divided into partial regions, and for each partial region, whether the partial region is a region including characters or a background region not including characters is determined. A grayscale image to be processed is divided into partial areas on a recording medium that is recorded in the form of a grayscale image character area extraction program that extracts only a character area by removing a background area from the entire image. A first determination process for specifying a character candidate partial area based on the average value and the standard deviation calculated for the density values of the pixels belonging to A second determination process for specifying a character candidate area by focusing on the rotation angle dependency of the value on the angle change, a determination result based on the first determination process, and a determination result based on the second determination process Nozomi Or a logical product or logical sum of both to determine a character area, and a character area / boundary shape extraction processing for extracting a character area or a character boundary shape based on the result of the comprehensive determination processing. A recording medium on which is recorded a program for extracting a character area of a gray-scale image, the program being recorded with a program which can be executed by a data processing apparatus by combining the above. 6. The recording medium according to claim 4, wherein the second determination process generates an image rotated by a predetermined angle for each partial region, and In the first method, an average value is calculated by integrating density values of pixel rows in one direction to obtain a projection value.
A second step of calculating the variance of the projection value of each pixel row in the other direction distribution for each rotation angle in each partial area; and a second step of calculating the variance of each partial area in the second step. A third process of calculating the variance of the projected value variance values of the individual rotation angles, which is the difference between the variance value of the rotation angle or the difference between the variance maximum value and the variance minimum value, and the third process. The given variance value or difference variance is compared with a threshold value given in advance,
A fourth step of determining whether or not the partial area is a partial area candidate including a character to be detected; and a recording medium storing a gray-scale image character area extraction program.