JPH1021332A - Non-linear normalizing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract a curved character like a handwritten character as straight character lines. SOLUTION: In the case of finding out distances between respective character lines constituting a character to be discriminated and distances between respective sides of a circumscribed rectangular frame 11 and respective character lines adjacent to the frame 11, setting up the inverses of respective distances as line density and executing non-linear normalizing processing in accordance with the line density, the line density of a part adjacent to each character line is set up to zero. Thereby a slightly curved character is extracted as straight character lines. A circumscribed rectangular frame 12 is formed on the outside of the frame 11, distances between respective character lines adjacent to the frame 11 and respective sides of the frame 12 are found out and the inverses of the distances are set up as line density. Consequently a character picture obtained by the non-linear normalizing processing is prevented from being reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition technique for recognizing a character to be identified such as a handwritten character, and more particularly to a non-linear normalization method for performing non-linear normalization processing on an image of a character to be identified.

[0002]

2. Description of the Related Art In general, character recognition is performed according to a procedure according to a recognition processing flow shown in FIG. That is, first, in a preprocessing step S1, a noise component is removed from an original image of a character to be recognized. Next, in a direction pattern creation step S2, a contour image is obtained from the original image of the recognized character from which the noise component has been removed, and the contour image is divided into pixels, and a black pixel having character information or character information is stored. It is configured as a binary pixel of a white pixel which is not used.
Then, the direction pattern of the contour image is extracted from the binary image including the black pixels or the white pixels.

After the direction pattern is extracted in this way,
In the next non-linear normalization step S3, the outline image is enlarged or reduced at different magnifications for each point of the image as described later. In the next feature vector creation step S4, each direction pattern of the image subjected to the nonlinear normalization processing is equally divided into 8 × 8 small areas (square areas).
By using the number of black pixels of each small area as a feature amount, a feature vector of each direction pattern is created.

[0004] Next, in a large classification step S5, a candidate character having a vector having a short distance among vectors in a dictionary provided in advance is compared with the directional feature vector of the input character (character to be recognized) obtained in step S4. A plurality is extracted in order from the closest one. This distance calculation is performed using a general calculation method called 4-neighbor-distance shown in Expression (1).

That is, d4 ((i, j), (h, k)) = | i−h | + | j−k | (1) (where (i, j) is the coordinate value of the input character, (h, k) are the coordinate values of the characters in the dictionary.) Then, in the detailed classification step S6, a detailed comparison is performed using a fine pattern (for example, a separate 32 × 32-dimensional feature vector is prepared) to make the final comparison. Perform character identification.

FIG. 8 is a diagram showing an outline of the nonlinear normalization processing. As described above, the direction pattern creation step S2
In FIG. 8B, horizontal, vertical, downward-sloping and upward-sloping directional patterns shown in FIG. 8B are extracted from the original figure (contour image) shown in FIG. 8A. In the normalization step S3, first, a line interval is obtained from the original figure in FIG. 8A to obtain line density information shown in FIG. 8C.

Here, the original figure is divided into 8 × 8 regions. At this time, the position of the dividing line is determined so that the line density of each region is the same as the line density information. I do. FIG. 8D is a diagram showing a result of the nonlinear normalization processing. Then, non-linear mapping is performed on each direction pattern in FIG. 8B based on the non-linear normalization information indicating the position of the dividing line, and the processing result for each direction pattern as shown in FIG. obtain. In each of the directional patterns of the image subjected to the non-linear normalization processing, a feature vector is extracted in the next feature vector creation step S4.

[0008]

In the conventional non-linear normalization processing, when obtaining line density information from an original figure, a line interval (line width) between character lines is counted and obtained. For this reason, in the case of a slightly curved straight character line such as a handwritten character, a line density that does not occur in the case of a purely straight character line occurs in the curved portion, and the character line is curved accordingly. Since it is extracted as it is, there is a problem that the subsequent character recognition is adversely affected, and it is desired to extract as a straight character line. Accordingly, an object of the present invention is to extract a linear character line having individual differences such as handwritten characters as a straight character line.

[0009]

In order to solve the above-mentioned problems, the present invention provides a first circumscribed rectangular frame circumscribing a character to be identified, the distance between each character line constituting the character to be identified,
And determining a distance between each side of the first circumscribed rectangular frame and each character line adjacent to the first circumscribed rectangular frame, and using a reciprocal of each distance as a line density in accordance with the line density information, the first circumscribed rectangular frame. Is divided into small regions, and the linear density near each character line is reduced to zero in a non-linear normalization method in which each information of each divided small region is mapped to each of the equally divided regions of a prescribed size. This is the method that was used. Therefore, in the case of a slightly curved character line such as a handwritten character, the line density generated in the vicinity thereof is removed, and such a curved character line can be extracted as a regular straight character line. Further, a second circumscribed rectangular frame having a constant multiple of the first circumscribed rectangular frame is provided outside the first circumscribed rectangular frame, and each character line adjacent to the first circumscribed rectangular frame is connected to the second circumscribed rectangular frame. The distance to each side is determined, and the reciprocal of this distance is defined as the linear density. As a result, the line density of a character image in which each character line adjacent to each side of the first circumscribed rectangular frame is close to the corresponding side can be reduced, and therefore, the reduction of the character image caused by nonlinear normalization Therefore, it is possible to avoid a problem that it becomes difficult to distinguish similar characters such as “Ta” and “Yu” after nonlinear normalization.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an apparatus to which the present invention is applied. Referring to FIG.
1, a touch panel 2, an LCD 3 formed on the lower surface of the touch panel, a memory 4, and a dictionary memory 5. Here, when the touch panel 2 is moved while being pressed by an input pen (not shown) between points A and B on the touch panel 2, for example, the CPU 1 sets the movement locus of the input pen during this time as a plurality of coordinate values via the touch panel 2. input. Then, the line segment from the point A to the point B is stored in the memory 4 and the line segment is displayed on the corresponding portion of the LCD 3. In this manner, handwritten characters based on pen input can be displayed on the LCD 3.

The data stored in the memory 4 and L
The CPU 1 performs the following processes on the handwritten characters displayed on the CD 3, and compares the processing result with each of the character types stored in the dictionary memory 5. Recognize whether it corresponds to. That is, in this case, the CPU 1 first performs preprocessing to remove noise and the like from the input characters (handwritten characters), which are the characters to be identified, to make the characters smooth. Thereafter, a contour image as shown in FIG. 2B is obtained from the original image shown in FIG. 2A, and the contour image is divided into 64 × 64 pixels.

Then, with respect to all the pixels (black pixels) constituting the outline, the direction component of each pixel is detected. That is, an arbitrary black pixel constituting the contour is focused on, and a directional component of the pixel of interest is detected from a connection pattern between the pixel of interest and black pixels before and after the pixel of interest.
FIG. 3 is a diagram showing a detection state of such a directional component of a black pixel.

Here, in FIG. 3, a pixel a in a shaded area
Ｃc indicate black pixels constituting the contour line, and pixels other than the hatched portions are assumed to be white pixels. If the pixel of interest is b and the black pixels before and after it are a and c, respectively, the outline is
The pixel of interest b enters the target pixel b horizontally from the black pixel a, and further reaches the black pixel c in the upward right direction through the pixel of interest b. Therefore, the target pixel b has two directional components of a horizontal component and a right upward component. It can be seen that Then, these two directional components are detected as the directional components of the pixel of interest b. Note that, in addition to the above-described two directional components, there are a vertical component and a rightward downward component as the directional components of the black pixels forming the contour line.
Two directional components of the one directional component are allocated.
When extracting the directional pattern of the contour, the CPU 1 sequentially detects the directional component for each black pixel for all the black pixels on the contour, and based on the detected directional components of the black pixels, detects the horizontal and horizontal components of the contour. Vertical, upward and downward directional patterns are extracted.

Next, the CPU 1 performs a non-linear normalization process of the contour image. That is, as shown in FIG. 4 (a), the distance (for example, A, B, C, D in the figure) between each character line (character stroke) of the character image "" is obtained, and this reciprocal is represented by a line. Density. Here, the distance between the character lines is counted separately in the horizontal direction and the vertical direction, and the reciprocals thereof are defined as the line density in the horizontal direction and the line density in the vertical direction, respectively. In this way, the line density between character lines (that is, white points other than the character lines (black portions) in the figure) is obtained. Note that character lines (that is, black dots)
Is "0". In the figure, reference numeral 11 denotes a circumscribed rectangular frame that circumscribes a character to be processed.

Next, the circumscribed rectangular frame 11 is replaced with coordinates, and as shown in FIG. 4B, the horizontal line densities of all the white points are in the x-axis direction and the vertical line densities are in the y-axis direction. Collect. Thereby, a line density histogram in the horizontal direction and the vertical direction is created. Note that 21 in FIG.
Indicates an arbitrary white point. Next, the CPU 1 divides the line density histogram into eight. At this time, as shown in FIG. 4C, the division positions are determined so that the sum of each of the eight divided histograms has the same value. Then, the area of the circumscribed rectangular frame 11 is divided into eight in each of the horizontal and vertical directions by each division line determined by the division position, to be an 8 × 8 small area.

The 8 × shown in FIG.
Each small area of 8 is mapped to a square rectangle of FIG. 4E equally divided into 8 × 8 with a specified size. Although the original image is shown in FIG. 4 for convenience of explanation, the image handled here is actually a contour line of the original image. In this case, the size of the mapping source and the size of the corresponding mapping destination are different in each region, but the individual regions of the mapping source are enlarged or reduced to map in correspondence with the region of the mapping destination.

Such a non-linear normalization process is performed for each of the vertical, horizontal, upward-sloping and downward-sloping contour patterns, and after the process is completed, the CPU 1 determines the number of black pixels of each equally divided small area. Counting is performed, and a count value for each small area is obtained as a directional feature vector of the directional pattern. Then, with respect to the obtained direction feature vector, a plurality of candidate characters having a vector having a short distance among the vectors of the dictionary in the dictionary memory 5 are extracted in order from the one with the shortest distance, and the extracted candidate character and the input character are extracted. By comparing in more detail,
Perform final character identification.

When the line density is obtained from the original figure, the distance between the character lines (line width) is counted and obtained as its reciprocal. However, the curve is slightly curved as shown in FIG. In the case of the straight character line 31, a line density that does not occur in a pure straight character line is generated due to the distance h of the curved portion, and it cannot be extracted as a normal straight character. In addition, characters such as “garden” and “field” have line spacings A, B, and B between the circumscribed rectangular frame 11 and the character line as shown in FIG.
C is small. In such a case, if this character image is subjected to nonlinear normalization processing, the value of the line density, which is the reciprocal of the line interval, becomes large at the line intervals A, B, and C. Therefore, the outline of the character after nonlinear normalization Is separated from the circumscribed rectangular frame 11, and is considerably reduced from the original outer size as shown in FIG. 7D. For this reason, it is difficult to distinguish similar characters such as “ta” and “yu”.

Therefore, when obtaining the line density, attention is paid to the position of a certain white point, the direction of the character line around the white point is examined, and the distance to the character line is determined based on the information. To ask. Then, the linear density is obtained from the reciprocal of the obtained distance. First, as shown in FIG.
Outside a circumscribed rectangular frame 12 which is a constant multiple of this circumscribed rectangular frame 11
Is provided. Then, as shown in FIG. 5A, when the character line 31 exists at any one of the upper, lower, left, and right of the white point 21, regarding the direction in which the character line 31 exists, the character line 31 and the circumscribed rectangular frame 12 The distance h between the sides where the white point 21 exists is counted, and the reciprocal thereof is defined as the linear density. In the direction in which the character line 31 does not exist, the reciprocal of the distance w between the sides of the circumscribed rectangular frame 12 is defined as the line density.

Next, as shown in FIG. 5B, when the character lines 31A and 31B exist above and below the white point 21 (or when each character line exists to the left and right of the white point 21), the character line 31
Regarding the direction in which A and 31B exist, the reciprocal of the distance h between the character lines is defined as the line density. In the direction in which no character line exists, the reciprocal of the distance w between the sides of the circumscribed rectangular frame 12 is defined as the line density. Next, as shown in FIG. 5C, when the character lines 31A and 31B exist above and to the left of the white point 21 (or any of the upper and right sides, the lower and left sides, and the lower and right sides of the white point 21). May be used), each character line 31A, 31B and the circumscribed rectangular frame 12
Of each side of the side where the white point 21 exists, h,
w is counted, and the reciprocal thereof is defined as the linear density.

Next, as shown in FIG. 5D, when character lines 31A, 31B and 31C are present on the left and right of and above the white point 21 (or on the left and right and below the white point 21, up and down and left and up and down, respectively). (It may be any one on the right).
2 is a distance h between one side of the side where the white point 21 exists.
And the reciprocal thereof is defined as the linear density. In addition, each character line 3
The distance w between 1B and 31C is counted, and the reciprocal thereof is defined as the linear density. Next, as shown in FIG. 5E, when the character lines 31A to 31D exist at all of the upper, lower, left, and right sides of the white point 21, the distances h and w between the opposing character lines are counted, and the reciprocal thereof is calculated. Each is defined as a linear density.

Thus, the linear density of the character image can be obtained. Therefore, as shown in FIG. 6 (f), the distances A and D between the circumscribed rectangular frame 11 and the character line are obtained, and the reciprocal thereof is defined as the line density. When the distances A and D are short, the character is obtained by nonlinear normalization. The problem that the image is reduced can be avoided. That is, as shown in FIG.
1, a circumscribed rectangular frame 12 is provided outside, and the distances A and D are set as distances A ′ and D ′ between the circumscribed rectangular frame 12 to increase the distance value, thereby reducing the line density. As a result, the size of the character outer shape after the non-linear normalization processing can be maintained without being reduced, and therefore, it is difficult to distinguish similar characters such as “ta” and “yu”, for example. Can be avoided.

As shown in FIGS. 6A to 6D,
The horizontal position of the white point 21 is set to a position substantially corresponding to the leading end position of a vertical character line (for example, the character line 31B), and the vertical distance h from the white point 21 to the horizontal character line 31A is counted. The horizontal distance w from the character line 31B to the position of the other end of the horizontal character line 31A through the white point 21 (or to another vertical character line 31C) is counted, and the counted vertical distance h and horizontal distance w By comparing the sizes of the horizontal character lines 31A, only the horizontal character lines 31A can be extracted as straight-line characters.

That is, in the case where the vertical distance h is extremely small as compared with the horizontal distance w as shown in FIGS. 6B and 6D, the distance h is set to "0", and the linear density is set to "0". It is set to “0”. Therefore, in the case of FIGS. 6B and 6D,
Considering one character line 31A, only the straight character line 31 can be extracted. 6 (a) and 6 (c) that the vertical distance h is not extremely small compared to the horizontal distance w.
In such a case, the distance h is not set to “0” and stored, and the reciprocal thereof is set as the linear density.

In this way, when the vertical distance h is extremely small compared to the horizontal distance w, the distance h is "0".
Similarly, when the horizontal distance w is extremely smaller than the vertical distance h, the distance w is “0”. Accordingly, the line spacing is “0” around the character line of the handwritten character “upper” in which the straight line portion is slightly curved as shown in FIG.
(That is, the linear density is “0”).
As described above, when one distance is extremely smaller than the other distance by comparing the distances in the vertical and horizontal directions, one distance is set to “0”. The line density becomes “0” around the slightly curved straight character line 31 as shown, and thus can be extracted as a straight character line. Note that it is not always necessary to provide the circumscribed rectangular frame 12 at the time of this straight-line character extraction. In this embodiment, an example of handwritten characters has been described, but any character data may be used as long as it is bitmap data, such as character data read by OCR.

[0026]

As described above, according to the present invention, the first circumscribed rectangular frame circumscribing the character to be identified is provided, the distance between each character line constituting the character to be identified, and the first circumscribed rectangle. The distance between each side of the frame and each character line adjacent to the first circumscribed rectangular frame is determined, and the first circumscribed rectangular frame is divided into small areas in accordance with the line density information using the reciprocal of each distance as the line density. Along with
When each information of each divided small area is mapped to each of the equally divided areas with the specified size, the line density near each character line is set to zero, so it is slightly different from handwritten characters. In the case of a curved character line, the line density occurring in the vicinity thereof is removed, and thus it can be extracted as a normal straight character line.
Further, a second circumscribed rectangular frame having a constant multiple of the first circumscribed rectangular frame is provided outside the first circumscribed rectangular frame, and each character line adjacent to the first circumscribed rectangular frame is connected to the second circumscribed rectangular frame. Since the distance to each side of was calculated and the reciprocal of this distance was used as the linear density,
The line density of a character image in which each character line adjacent to each side of the first circumscribed rectangular frame is close to each corresponding side can be reduced,
Therefore, since it is possible to prevent the character image from being reduced when the nonlinear normalization is performed, it is possible to avoid a problem that it is difficult to determine similar characters such as “ta” and “yu” after the nonlinear normalization.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an apparatus to which the present invention is applied.

FIG. 2 is a diagram illustrating a character image to be recognized by the apparatus and a contour line of the character image.

FIG. 3 is a diagram illustrating a detection state of a directional component of a pixel in the present apparatus.

FIG. 4 is a diagram illustrating a state of nonlinear normalization processing of a contour line.

FIG. 5 is a diagram illustrating a calculation state of a distance (line interval) between character lines during nonlinear normalization.

FIG. 6 is a diagram illustrating a main part of a process when calculating a distance between character lines.

FIG. 7 is a diagram illustrating an example of a problem that occurs during conventional nonlinear normalization.

FIG. 8 is a diagram showing a processing procedure of conventional nonlinear normalization.

FIG. 9 is a recognition processing flow showing a process of character recognition.

[Explanation of symbols]

1 CPU, 2 touch panel, 3 LCD, 4 memory, 5 dictionary memory, 11 and 12 circumscribed rectangular frame.

Claims (2)

[Claims] A first circumscribing rectangular frame circumscribing the character to be identified; a distance between character lines constituting the character to be identified;
And determining a distance between each side of the first circumscribed rectangular frame and each character line adjacent to the first circumscribed rectangular frame, and defining a reciprocal of each distance as a line density in accordance with the line density information. A non-linear normalization method that divides a frame into each small area and maps each piece of information of each divided small area into each area equally divided into a prescribed size, wherein a line density near each of the character lines is provided. Is set to zero. 2. The character according to claim 1, wherein a second circumscribed rectangular frame having a constant multiple of the first circumscribed rectangular frame is provided outside the first circumscribed rectangular frame, and each character adjacent to the first circumscribed rectangular frame is provided. A non-linear normalization method, wherein a distance between a line and each side of a second circumscribed rectangular frame is obtained, and a reciprocal of the distance is used as a line density.