JP2580976B2 - Character extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character extracting device for a character recognizing device, and more particularly to a character extracting device for optically read free handwritten characters.

[0002]

2. Description of the Related Art Conventionally, as a character extracting method of this kind, as shown in FIGS.
This image is considered to exist on a two-dimensional space represented by two coordinates, i.e., an axis and a j-axis. By irradiating this image with parallel rays perpendicular to the i-axis, shadows I0 and I1 are obtained on the i-axis. I was

[0003] Of the end points of the shadows I0 and I1, the one closer to the coordinate origin is denoted by s, and the far end is denoted by l. If the end points of the shadows I0 and I1 are Is, respectively.
0, I10 and Is1, I11 (see FIG. 10).

When the image is irradiated with parallel rays perpendicular to the j-axis from the end points of the shadows I0 and I1, shadows J0 and J1 are obtained as shown in FIG. Of these shadows, the one closer to the origin is denoted by s and the farther one is denoted by l, as in the previous case, the shadows J0 and J1
Are respectively Js0, J10 and Js1, Js
11 (see FIG. 11).

[0005] The thus obtained (Is0, Js
0), (I10, J10), (Is1, Js1), (I
11 and Jl1) represent the diagonal coordinates of the rectangle circumscribing each character as shown in FIG. 12. As a conventional technique, it is general to cut out each character image from the character string image in this way. there were.

In the following, a method of irradiating the coordinate axes with parallel rays perpendicular to the respective coordinate axes as described above and obtaining a circumscribed rectangle from the shadow thereof is referred to as a simple projection method.

[0007]

In such a conventional character extracting apparatus, a simple projection method for extracting an area having a projection distribution of 0 (that is, a portion where no shadow appears) as a boundary between characters is adopted. Because of the basics, even if the characters are in contact with each other or the characters are not in contact with each other, if the shadow is in contact, accurate character extraction cannot be performed, which causes a reduction in the recognition rate.

[0008]

In order to solve the above-mentioned problems, a character extracting apparatus according to the present invention provides a character extracting apparatus for optically reading and identifying each character of a character string on a paper sheet. In the device, an image memory for storing a binary image input from the image input device, and a block circumscribing each character or character string from the projection data obtained by projecting the binary pixels stored in the image memory. A circumscribing block extracting unit that extracts information (circumscribing block information), a standard character size calculating unit that calculates a standard character block size in the binary image from the circumscribing block information extracted by the circumscribing block extracting unit, By comparing the standard character block size calculated by the standard character size calculation unit with the circumscribed block information, a plurality of character image A continuation character block detection unit that detects a continuation character block in which is stored, a projection distribution calculation unit that calculates a projection distribution of a one-line character string binary image stored in the image memory and a frequency distribution with respect to the projection distribution. A threshold value for calculating an optimal threshold value for separating a continuous character block detected by the continuous character block detection unit into a circumscribed block of one character size from the frequency distribution calculated by the projection distribution calculation unit; A calculation unit; a separation point determination unit that determines a separation point for separating the continuous character block into a circumscribed block of one character size based on the optimal threshold value calculated by the threshold value calculation unit; A final block determining unit that determines final block information of the circumscribed block based on the separation point coordinates determined by the determining unit.

[0009]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a character extracting apparatus according to the present invention.

FIG. 1 is a block diagram showing one embodiment of the present invention. The character string described on the paper sheet is input by the binary image input device 11, further converted into a binary image, and stored in the image memory 12. The stored binary image is passed through an image bus to a circumscribed block extraction unit 1.
3. Projection distribution calculation unit 17 and final block determination unit 21
Can be referenced from The circumscribed block extracting unit 13 obtains a circumscribed rectangle by using the simple projection method described in the section of the related art so that a character or a character string is accommodated in the frame. Hereinafter, the circumscribed rectangle is referred to as a primary circumscribed block, and the diagonal position coordinates of the primary circumscribed block with respect to the coordinate axes and the length in the horizontal axis direction or the vertical axis direction are referred to as primary circumscribed block information. The primary circumscribed block information is stored in the circumscribed block memory 14. The standard character size calculation unit 15 calculates the standard character size from the arithmetic mean of the size of the primary circumscribed block based on the primary circumscribed block information. Continuation character block detection unit 16
Detects one circumscribed block considered to include a plurality of character images based on the standard character size and the primary circumscribed block information. The projection distribution calculating unit 17 calculates the projection distribution by projecting the binary image in the direction perpendicular to the character string direction, and stores the result in the projection distribution memory 18.
The frequency distribution of this projection distribution is calculated. Threshold calculator 1
9 calculates a line width frequency value from the frequency distribution, and
Calculate the optimal threshold. The separation point determination unit 20 separates the primary circumscribing block of the continuous character portion based on the optimum threshold. The final block determination unit 21 finally determines the separation determination unit 2
The last block information separated by 0 is transferred to the last block memory 22.

Next, the operation of the present invention will be described with reference to FIGS. 1, 2, 3, 4 and 5.

In the binary image input device 11, characters and the like on a paper sheet are input by optical scanning, the obtained image is converted into a binary image, and temporarily stored in the image memory 12.

Next, in the circumscribed block extraction unit 13,
One line of the binary image stored in the image memory 12 is subjected to projection processing by the simple projection method, and a rectangle (primary circumscribed block) circumscribing each character or character string is obtained from the obtained projection data. The primary circumscribed block information is stored in the circumscribed block memory 14. FIG. 2 shows an example of the obtained primary circumscribed block. Here, BK7
Block contains a plurality of characters, and it was found that accurate character extraction could not be performed.

Next, the standard character size calculator 15 calculates i for each of the primary circumscribed blocks BKn (n = 1, 2, 3,..., 8) in FIG.
The width Hn in the axial direction and the height Wn in the j-axis direction are obtained, and the arithmetic mean of Hn and Wn is calculated.
A standard character size, which is an average block size of one character, is calculated. Here, when "1" (one) like BK1 and "," (reading point) like BK2 in FIG. 2 are included in the calculation target, the arithmetic mean value becomes small, and a plurality of BK7 like BK7 If the primary circumscribing block containing the characters is included in the calculation target, the arithmetic mean value will increase, and it will not be possible to calculate an accurate standard character size. Blocks to be calculated are excluded from the calculation.

Next, the standard character size calculated by the standard character size calculation unit 15 is transferred to the continuation character block detection unit 16, and the continuation character block detection unit 16 stores the primary circumscription stored in the circumscription block memory 14. The block information is compared with the standard character size, and a block larger than the standard character size in the primary circumscribed block information is a block containing a plurality of character images (for example, B in FIG. 2).
K7).

Next, let the binary image input as shown in FIG. 2 be p (i, j), and p (i, j) be

[0017]

## EQU1 ##

The projection distribution calculation unit 17 calculates a projection distribution h (i) obtained by projecting the binary image p (i, j) stored in the image memory 12 in a direction perpendicular to the row direction. It is stored in the projection distribution memory 18. Here, the projection distribution h (i) is

[0020]

Is defined as Further, the projection distribution calculation unit 17
Calculates the frequency distribution g (k) of the calculated projection distribution h (i). Here, the frequency distribution g (k) is

[0022]

Is defined as Graphs of the projection distribution h (i) and the frequency distribution g (k) calculated by the projection distribution calculation unit 17 are shown in FIGS. 3 and 4, respectively.

Next, the threshold value calculating section 19 calculates the value of k when the value of the frequency distribution g (k) calculated by the projection distribution calculating section 17 has the maximum value. Here,
The value of k is referred to as a line width degree value m. If the thickness of the character stroke is constant, a flat region having the height of the character thickness appears on the projection distribution, and the flat portion on the projection distribution has the maximum value on the frequency distribution. It uses the property that appears.

Next, utilizing the fact that the inherent optimum threshold value θs and the line width degree value m have a correlation as shown in FIG. 5 with respect to the projection distribution of the input image, the threshold value calculating section 19 , Θs = a · m + b (a and b are constants) to find the optimum threshold value θs. Here, FIG. 5 shows the optimal threshold θs and the line width degree value m, which were investigated for input images of character strings of various thicknesses written with various writing tools such as a pencil, a ballpoint pen, a felt pen, and a brush. Shows the correlation. Here, in the present embodiment, a = a =
0.4 and b = 4.1 are used. However, the optimum values of the constants a and b are selected depending on factors such as the resolution of the input image to be applied, the distribution of the character size, the regionality, the vertical writing and the horizontal writing.

The optimum threshold value θ thus calculated
Based on s, the separation point determination unit 20 performs the following processing. (1) The height h (i) of the projection distribution is binarized using the optimum threshold value θs. (2) When the projection distribution is viewed along the i-axis direction, the i-coordinate of a change point where the height h (i) of the projection distribution shifts from a point higher than θs to below θs is set as a cutout candidate position. (3) Investigate whether the area where the height of the projection distribution is less than θs continues for a certain length or more, and if it does not continue for a certain length, remove it from the candidates. (4) With respect to all new blocks obtained as a result of cutting at the cutout position in the i direction, the inside of the block is scanned in the j-axis direction to check for a black image, and j is checked until a black image is found for the first time.
Decrease the block size in the direction.

The information of the block cut out by the separation point determining unit 20 in this manner is called final block information.
The final block determining unit 21 transfers the final block information to the final block memory 22. Here, FIG. 6 shows an example of the last block information for FIG. 2 which is the input image of the present embodiment.

FIG. 7 is an example of character extraction for a contact character and a continuous character, FIG. 8 is an example of character extraction for a thin character, and FIG. 9 is an example of character extraction for a thick character.

[0029]

As described above, the character extracting apparatus according to the present invention can be used for free-handwriting with no restriction on writing implements, and can be used to convert continuous characters and overlapping characters written by an unspecified number of people according to the thickness of a character stroke. By having the function of automatically calculating the unique threshold value, highly accurate extraction can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention.

FIG. 2 is a diagram showing an input binary image and its primary circumscribed block.

FIG. 3 is a projection distribution diagram for the binary image shown in FIG. 2;

FIG. 4 is a frequency distribution diagram for the projection distribution diagram shown in FIG. 3;

FIG. 5 is a diagram showing a correlation between a line width degree numerical value and an optimum threshold value.

FIG. 6 is a diagram showing a final block with respect to the binary image shown in FIG. 2;

FIG. 7 is a diagram illustrating an example of character cutout for a contact character and a continuation character.

FIG. 8 is a diagram showing an example of character extraction for a thin character.

FIG. 9 is a diagram showing an example of character extraction for thick characters.

FIG. 10 is a diagram illustrating a relationship between a coordinate axis and a binary image.

FIG. 11 is a projection view in the i-axis direction.

FIG. 12 is a projection view in the j-axis direction.

FIG. 13 is a diagram showing an example of character extraction by the simple projection method.

[Explanation of symbols]

 Reference Signs List 11 binary image input device 12 image memory 13 circumscribed block extraction unit 14 circumscribed block memory 15 standard character size calculation unit 16 continuous character block detection unit 17 projection distribution calculation unit 18 projection distribution memory 19 threshold value calculation unit 20 separation point determination unit 21 last block determination unit 22 last block memory

Claims (3)

(57) [Claims] 1. A character cutout device in a character recognition device for optically reading and identifying each character of a character string on a paper sheet, comprising: an image memory for storing a binary image input from an image input device; A circumscribing block extracting unit that projects binary pixels stored in an image memory and extracts information (circumscribing block information) of blocks circumscribing each character or character string from the obtained projection data; A standard character size calculation unit for calculating a standard character block size in the binary image from the circumscribed block information extracted in the above, and a standard character block size calculated by the standard character size calculation unit and the circumscribed block information A continuation character block detection unit for detecting a continuation character block storing a plurality of character images by comparing A projection distribution calculator for calculating a projection distribution of a character string binary image for one line stored in the memory and a frequency distribution for the projection distribution; and a continuation character block detector based on the frequency distribution calculated by the projection distribution calculator. A threshold calculation unit that calculates an optimum threshold value for separating the continuous character block detected in step 1 into circumscribed blocks of one character size; A separation point for determining a separation point for separating the continuous character block into a circumscribed block of one character size based on the separation point coordinates based on the separation point coordinates determined by the separation point determination unit. And a final block determining unit for determining information. 2. The method according to claim 1, wherein the threshold value calculating unit calculates a thickness of the input character image based on a correlation between the thickness of the input character image and an optimum threshold value for separating the continuous character block into circumscribed blocks of one character size. 2. The character extracting device according to claim 1, wherein the character extracting device changes according to the change. 3. The separation point determination unit calculates a height of the projection distribution of the character string binary image calculated by the projection distribution calculation unit using an optimum threshold calculated by the threshold calculation unit. A position where the height of the projection distribution shifts from a point higher than the optimum threshold to a point lower than the optimum threshold is set as a character extraction position candidate, and an area where the height of the projection distribution is lower than the optimum threshold continues for a certain length. 2. The character extracting position candidate corresponding to the position is excluded from the candidates if the position has not been set.
The character extracting device according to 1.