JPH0327488A - Character recognizing device - Google Patents

Abstract

PURPOSE:To read even a business form in which characters of different character styles are mixed at high speed and at high precision by detecting an italic character by extracting feature point coordinate, and deciding the character style, and selecting a dictionary mask for collation based on a decided result. CONSTITUTION:An input character pattern is represented by X-Y coordinates system by giving X-coordinate and Y-coordinate to the picture element of the picture data of the input character pattern segmented by every one character. Then, the maximum and the minimum calculation values alphaX+betaY are detected by using X and Y coordinates and a specified values of alpha and beta, and the X and Y coordinates of the picture element of the input character pattern to give these maximum and minimum values are detected. Geometrical feature quantity is calculated based on the detected X and Y coordinates. The character style is decided based on the calculated feature quantity. A recognition diction ary mask corresponding to the decided character style is selected. The collation of the input character pattern is executed by using the selected dictionary mask, and an input character is identified. Thus, the business form containing plural character styles can be read at high speed.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a character recognition device that can read documents containing characters in a plurality of fonts at high speed and with high accuracy.

(Prior Art) Conventionally, in character recognition devices, for example,
There is one disclosed in Japanese Patent No. 38756, which is configured with the following components (a) to (f).

(a) By photoelectrically converting and quantizing character figures, an original pattern of a digital signal represented by black bits and white bits is created.

(b) Next, calculate the line width of the original pattern.

(c) Next, the original pattern is scanned in a plurality of directions to detect the number of consecutive black bits for each scanning line, and the number of consecutive black bits and the number of consecutive black bits are detected, and the number of consecutive black bits and the line A plurality of sub-patterns corresponding to each of the plurality of scanning directions are extracted based on the width.

(d) Next, the region within the character frame of the original pattern is divided into (NXM) regions (N, 'M are constants) for the subpattern, and black points are counted in units of cells within the divided region. The feature amount is calculated based on the result and the line width of each sub-pattern.

(e) Next, a feature matrix is created by normalizing the feature amount using the character size.

(f) Then, the feature matrix is compared with a standard character mask of a character/figure pattern prepared in advance to recognize the character/figure.

When such a character recognition device recognizes a form that includes multiple fonts, for example, English text in which only specific words are printed in italics to emphasize names or place names, as shown in Figure 9, Prepare standard character masks for all fonts to be recognized as dictionary masks in advance,
A widely used method is to perform recognition by comparing all of the dictionary masks with input character figures.

(Problems to be Solved by the Invention) However, in the conventional method described above, the number of dictionary matrices increases in proportion to the number of fonts, and the number of matching increases, resulting in a significant reduction in recognition speed, and furthermore, recognition accuracy. There was a problem in that it caused a decrease in .

The present invention has been made to solve these problems, and an object of the present invention is to provide a character recognition device that can read forms containing a plurality of fonts at high speed and with high accuracy.

(Means for Solving the Invention) In order to solve the above-mentioned problems, the present invention photoelectrically converts the feature extraction target on the medium to obtain quantized image data, and from the image data, character patterns are created one by one. , determine the font of the extracted input character pattern, select a recognition dictionary mask based on the determination result,
In a character recognition device that recognizes characters by collating an input character pattern using a selected dictionary mask, , a Y-coordinate generation means for assigning a Y-coordinate to a pixel of image data of an input character pattern cut out one character at a time; coordinate detection means for detecting the maximum and minimum calculated values αX+βY for pixels of the input character pattern having values, and detecting the X, Y coordinates of the pixels of the input character pattern that give these maximum and minimum calculated values; , a feature amount calculation means for calculating a geometric feature amount based on the Y coordinate, and a font determination unit that determines the font based on the feature amount calculated by the feature amount calculation means; The present invention is characterized in that it includes a dictionary section that selects a recognition dictionary mask corresponding to a font, and an identification section that uses the selected dictionary mask to compare input character patterns and identify characters.

(Operation) According to the present invention having the above configuration, X is applied to pixels of image data of an input character pattern cut out character by character.
The input character pattern is expressed in the X-Y coordinate system by assigning coordinates and Y coordinates. Then, using the X, Y coordinates and at least two specific sets of α and β values, calculate maximum and minimum values α for pixels of the input character pattern having a predetermined pixel value.
X and βY are detected, and the X and Y coordinates of the pixels of the input character pattern that give these maximum and minimum calculated values are detected. Geometric features are calculated based on the detected X and Y coordinates. The font is determined based on the calculated feature amount. A recognition dictionary mask corresponding to the determined font is selected.

The input character pattern is matched using the selected dictionary mask, and the input character is identified.

Therefore, the present invention can solve the above-mentioned problems, and can provide a character recognition device that can read forms containing a plurality of fonts at high speed and can also read documents with high accuracy.

(Example) Hereinafter, one embodiment of the present invention will be described based on the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. In the figure, IO is a character recognition device, 11 is a photoelectric conversion unit,
12 is a line buffer, l3 is a character cutting section, l4 is a pattern register, 15 is a font determining section, l6 is a dictionary section,
17 is an identification section, and 18 is an output terminal.

The output terminal 18 is connected to, for example, a data input terminal of an external device such as a computer, and is used to output a character name (for example, a JIS character code) for which character recognition has been completed.

Figure 2 (a) is a diagram showing binary image data in the character line area, Figure 2 (b) is a diagram showing the peripheral distribution of binary image data, and Figure 2 (c) is a diagram showing the characteristic values of the input character pattern. FIG. 2(d) is a diagram showing recognition results for input character patterns.

FIG. 3 is a block diagram showing the font determining section 15 of FIG. 1. In the figure, 30 is an X coordinate generating means, 32 is a Y coordinate generating means, 34. 36 is a coordinate detection means, and the coordinate detection means 34 includes an x+y calculation means 34l, a maximum value detection means 342, a maximum value coordinate storage means 343, and a minimum value detection means 3.
44 and minimum value coordinate storage means 345, and the minimum value detection means 36 includes an X-Y calculation means 36l, a maximum value detection means 362, a maximum value coordinate storage means 363, and a minimum value detection means 3.
64, minimum value coordinate storage means 365 is provided. 38 is a feature calculation means.

FIG. 4 is a flowchart showing the operation of the maximum value coordinate detection means in this embodiment.

FIG. 5 is a flowchart showing the operation of the minimum value coordinate detection means in this embodiment.

FIG. 6 is a block diagram showing the configuration of dictionary l6 in FIG. 1. In the figure, 60 is a dictionary selection section, 61 is a first dictionary matrix, and 62 is a second dictionary matrix.

FIG. 7 is a diagram explaining the principle of feature point coordinate detection in this embodiment. FIG. 7(a) shows Roman's standard type, and FIG. 7(b) shows Roman's italic type.

FIG. 8 is an explanatory diagram of standard character patterns of Roman's standard type and Roman's italic type, and a dictionary matrix of the characters.

FIG. 9 is a diagram showing a form in which characters used in this embodiment are written.

Figure lO(a) is a diagram showing an example of an input format table,
Figure O (b) is a diagram illustrating the input format table.

Hereinafter, this embodiment will be explained in detail using FIGS. 1 to 10.

First, obtaining an input character pattern of characters to be recognized will be explained. The photoelectric conversion unit l1 in FIG. 1 detects a character line area from an optical signal (indicated by S in FIG. 1) from a medium such as a form on which characters, figures, etc. (hereinafter referred to as characters) are written, The character line area is photoelectrically converted, and the character line part is set to a pixel value of "1".
The row image data expressed by a binary digital signal for each pixel is obtained by using the black bit and the background part as the white bit of the pixel value rOJ, and is stored in the line buffer 12. Here, the character line area is an area for one line in which characters are written on a form.

The line buffer l2 stores the signals of each pixel in the line image data of the input character pattern in a format that can reproduce the two-dimensional coordinates of this area, and has a size of 128×4096 pixels.

Next, the character cutting section l3 reads the line image data from the line buffer 12, performs scanning in the vertical direction, and creates a distribution of black dots. The area from the point where the distribution of black dots changes from O to 1 or more to the point where it changes from 1 or more to O is defined as one character expected area, and is stored in the pattern register 14 as an input character pattern.

The pattern register 14 stores the signals of each pixel in the character area of the input character pattern in a format that can reproduce the two-dimensional coordinates of this area, and has a size of 128×128 pixels.

The input pattern stored in the pattern register 14 is output to the identification section 17 and the font determination section 15.

Next, the output of the font control signal in the font determining section l5 of FIG. 1 will be explained.

The font determining section 15 determines the font of the input character pattern read from the pattern register l4, and outputs a font determining signal to the dictionary section l6. Note that M in FIG. 3 indicates quantized image data containing an input character pattern to be recognized, which is read from the pattern register l4. The font determining unit 15 includes an X coordinate generating means 30 for assigning an X coordinate to a pixel of the image data M, a Y coordinate generating means 32 for assigning a Y coordinate to a pixel of the image data M, and the X, Y coordinates. and at least two sets of specific α and β values to detect the maximum and minimum calculated values αX + βY for the pixels of the input character pattern, and calculate the Coordinate detection means 34 for outputting coordinates as feature point coordinates. 36, and feature amount calculating means 38 for calculating a geometric feature amount for determining the font of an input character pattern based on the feature point coordinates.

Further, the coordinate detection means 34 includes a calculation means 341 for calculating the calculated value aX+βY, a maximum value detection means 342 for detecting the maximum calculated value, and a storage for storing the X and Y coordinates of the pixel giving the maximum calculated value. Maximum value coordinate storage means 343 for
and a minimum value detection means 34 for detecting the re-acceptance calculated value.
4, and minimum value coordinate storage means 345 for storing the X, Y coordinates of the pixel that gives the minimum calculated value. The coordinate detection means 36 also includes a coordinate detection means 362, a maximum value coordinate storage means 363, a reacknowledgement detection means 364, and a minimum value coordinate storage means 365.

In this example, (α=β=1) and (α=1.β=-1
), the inclination is detected using the pixel coordinates that give the maximum and minimum calculated values for X+Y and X-Y, and a signal for dictionary selection is output to the dictionary section l6, so two Coordinate detection means 34. 36. The coordinate detecting means 34 detects the coordinates of the pixel giving the maximum and minimum calculated values regarding the calculated value X+Y, and the coordinate detecting means 36 regarding the calculated value X-Y.

The font determination unit 15 in FIG. 1 will be explained below in FIGS. 4 and 5.
This will be explained in detail using figures.

[Explanation focusing on the coordinate detection means 34] (Explanation of the maximum value and minimum value detection method of x+y) The input character pattern read from the pattern register l4 in FIG. and is input to the minimum value detection means 344. Along with this, the X coordinate generating means 30 and the Y coordinate generating means 32
generates X and Y coordinates associated with each pixel of the data M in synchronization with the output of the image data M. As a result, these generating means 30. 32, X and Y coordinates are assigned to the image data M. The output X, Y coordinates are calculated by the calculation means 34l and the maximum value coordinate storage means 3.
43 and the minimum value coordinate storage means 345 (step 402). When the calculation means 341 receives the X and Y coordinates, it calculates a calculated value X+Y from these X and Y coordinates, and outputs the calculated value to the maximum value detection means 342 and the minimum value detection means 344 (step 403).

Through these steps 402 and 403, the maximum value detection stage 342 transfers the image data M and the calculated value to the minimum value detection means 3.
44 stores image data M and calculated values, maximum value coordinate storage means 343 stores X and Y coordinates, and minimum value coordinate storage means 345 stores X and Y coordinates.
, Y coordinates are input for each l pixel. Then, the coordinate detection means repeatedly performs the determination in steps 404, 405, 405, or 407, which will be described later, for each pixel, and performs an operation according to the determination result.

In particular, in step 404, the maximum value detection means 342 and the minimum value detection means 344 determine whether the pixels of the input image data M are the pixels of the input character pattern M2.

This determination is made by determining whether the pixel value of the input pixel is a predetermined pixel value (pixel value "1" in this embodiment) indicating the input character pattern M2.

When it has a predetermined pixel value, the maximum value detection means 342
In step 404, the comparison value and the calculated value are compared in step 405, and the minimum value detection means 344 is in step 4.
After step 04, the comparison value and the calculated value are compared in step 501.

Here, in step 405, ■ When the calculated value is larger than the comparison value, the maximum value detection means 342 replaces the previously stored comparison value and stores the calculated value larger than the comparison value as a new comparison value (comparison (value rewriting), and together with this, a set pulse is output to the maximum value coordinate storage means 343. The maximum value coordinate storage means 343 to which the set pulse has been input stores the X, Y coordinates of the pixel that gives a calculated value larger than the comparison value, instead of the stored X, Y coordinates. rewriting) (step 406).

■When the calculated value is smaller than or equal to the comparison value, the maximum value detection means 342 stores the previously stored comparison value without rewriting it, and the maximum value coordinate storage means 3
43 does not rewrite the stored X, Y coordinates.

The maximum value detection means 342 performs the determination in step 407 following step 405 in both cases (1) and (2) above.

l 5 1 6 Note that as the initial value of the comparison value stored in the maximum value detection means 342, for example, a value smaller than the value that can be taken as the calculated value α may be used. For example, α=β=1, and the image data M is divided into pixels of β rows and m columns (therefore, O≦X≦m
−1, and O≦Y≦Shinichi 1), for example, −
1 can be set as the initial value of the comparison value. Alternatively, the first input calculation value αX+βY may be used.

Further, the maximum value detection means 342 rewrites the comparison value and the X, Y coordinates when the calculated value αX + βY is larger than the comparison value and when the calculated value is equal to the comparison value. When it is smaller than the comparison value and X,
The Y coordinate may not be rewritten.

Furthermore, any suitable numerical values may be used as the X and Y coordinates of the maximum value coordinate storage means 343.

Here, in step 501, ■ When the calculated value is smaller than the comparison value, the minimum value detection means 344 replaces the previously stored comparison value with a new calculation value smaller than the comparison value and stores it as a comparison value (comparison value (rewriting), and outputs a set pulse to the minimum value coordinate storage means 345. When the minimum value coordinate storage means 345 receives a set pulse, it rewrites the stored X, Y coordinates to the X, Y coordinates of a pixel that gives a calculated value smaller than the comparison value (rewriting the x, y coordinates).
Step 406).

■When the calculated value is smaller than or equal to the comparison value, the minimum value detection means 344 stores the previously stored comparison value without rewriting it, and the minimum value coordinate storage means 344 stores the previously stored comparison value as it is without rewriting it.
45 does not rewrite the stored X, Y coordinates.

The minimum value detecting means 344 performs the determination in step 407 after step 501 in either of the cases (1) and (2) above.

Note that as the initial value of the comparison value stored in the minimum value detection means 344, for example, a value larger than the value that can be taken as the calculated value αX+βY may be used. For example, α = β = 1, and the image data M is divided into β rows and m columns of pixels (therefore, 0
≦X≦m−1 and O≦Y≦12−1), for example, m+n−1 can be set as the initial value of the comparison value. Alternatively, the minimum value detection means 34 is used as the initial value of the comparison value.
The calculated value αX+βY inputted first in step 4 may be used.

Further, the minimum value detection means 344 detects the comparison value and X, when the calculated value is smaller than the comparison value and when the calculated value is equal to the comparison value.
The Y coordinate may be rewritten, and at the same time, when the calculated value is larger than the comparison value, the comparison value and the X and T coordinates may not be rewritten.

Furthermore, any suitable numerical values may be used as the X and Y coordinates of the minimum value coordinate storage means 345.

Furthermore, in step 407, ■ If the scanning of the image data M is not completed, the coordinate detection means 3
4, the scanning of the image data M is not completed, and therefore the data M
If processing has not been completed for all pixels of image data M, steps 404, 405,
It makes the judgment 501 or 407 and operates according to the judgment result.

■When the scanning of the image data M is completed, the maximum value detection means 342 and the minimum value detection means 344 output the output signals of It is output to the maximum value coordinate storage means 343 and the minimum value coordinate storage means 345. Coordinate storage means 343 into which this output signal is input
, 345 outputs the stored X, Y coordinates as feature point coordinates. In both cases, the detection means 342 and 344 initialize the comparison values (step 408).

When processing is completed for all pixels, the X, Y coordinates of the pixel whose X, Y coordinates stored in the maximum value coordinate storage means 343 and the minimum value coordinate storage means 345 give the maximum and minimum calculated values, that is, the feature point It becomes the coordinates.

In the coordinate detection means 34, α=β=1, so at the end of all processing, the input character pattern M shown in FIG.
The coordinates of the second feature point BR are stored in the maximum value coordinate storage means 343, and the coordinates of the feature point TL are stored in the minimum coordinate storage means 345.

[Explanation focusing on the coordinate detection means 36] (Explanation of X-Y maximum value/minimum value detection method) The coordinate detection means 36 performs the same operation as the coordinate detection means 34 in parallel with the operation of the coordinate detection means 34 described above. Since the detection means 36 operates in the following manner, a description of the operation of the detection means 36 will be omitted.

In the coordinate detection means 36, since α=1 and β=1l, when processing is completed for all pixels, for example, the coordinates of the feature point TR of the input character pattern M2 shown in FIG. The coordinates of the feature point BL are stored in the value coordinate storage means 363, and the coordinates of the feature point BL are stored in the minimum value coordinate storage means 365.

Using the coordinates of TL, BL, TR, and BR for the feature points of the input character pattern, the feature value F is calculated using the following equation (1).

Note that in equation (1), the feature points TL, BL, TR,
The X coordinate of BR is TLX, BLX, TRX, B
Expressed as RX.

F=k (I2(TLX-BLX)+m(TRX-BR
X)}...Equation (1) In Equation (1), k, β, and m are arbitrary constants.

The font is determined according to the condition of the following formula (2) of F calculated by the formula (1), and the font determination signal N is sent to the dictionary section l6.
Output.

In equation (2), CIl C2 is a fixed threshold value and can be changed arbitrarily.

In addition, in this embodiment, the font determination signal N to be output is rl
There are three types, l, r2J, and r3J, but there is no problem even if the font determination signal N is other than the three types.

Next, identifying an input character pattern and outputting the result will be described.

The font determination signal N output by the font determination section 15 in FIG. 1 is output to the dictionary section l6, and a dictionary corresponding to the font determination signal N is selected.

The dictionary section l6 in FIG. 6 includes a dictionary selection section 60, a first dictionary matrix 61, and a second dictionary matrix 62.

In this embodiment, there are two dictionary matrices, but there may be no problem with three or more dictionary matrices.

The dictionary selection unit 60 selects a first dictionary matrix 61 and a second dictionary matrix 6 in response to the font determination signal N=1, N=2, or N=3 output from the font determination unit 15, respectively.
2, or the first and second dictionary matrices 61. 62 and outputs the selected dictionary matrix to the identification unit l7.

Next, the identification section l7 will be explained.

The identification unit 17 performs feature extraction processing and recognition of the input character pattern for this input character pattern. Although various conventionally known methods can be used for this feature extraction method, in the case of this embodiment, the method described below is used.

First, a rectangular frame circumscribing the input character pattern is detected and used as a character frame. Furthermore, the line width W is calculated for the input character pattern. This line width calculation can be performed using, for example, the well-known approximation formula (3) shown below.

Vl=1/{1-(Q/A))...Formula (3
) However, in equation (3), when Q is divided into a window in which each point constituting the input character pattern can be seen in a range of (2×2) points, all points within this window are black bits. is the number of windows, and A is the number of black bits in the character frame.

Furthermore, this input character pattern is scanned in multiple directions to detect the number of consecutive black bits in each scanning line, and based on this number of consecutive black bits and the above-mentioned line width, the input character pattern is scanned in multiple directions. Extract each corresponding subpattern. Then, the region within the character frame of this input character pattern is divided into (N The feature values are calculated for each area, and the feature values are normalized using the size of the character frame to obtain a feature matrix.

In this embodiment, the feature amount is divided by a value of (Δχ + ΔY)/2 for normal viewing. Here, ΔX is the horizontal length of the character frame, and ΔY is the vertical length.

The identification unit 17 compares the feature matrix 2 3 2 4 extracted in this way with the dictionary matrix after dictionary selection output from the dictionary unit 16, and selects a dictionary that shows the largest value of similarity. The character name (JIS code, etc.) corresponding to the matrix is output to an external device via the output terminal l8.

In the case of this embodiment, the above-mentioned similarity is calculated based on equation (4) shown below.

However, in equation (4), R is the degree of similarity, f is the input character/pattern, g1 is the dictionary matrix stored in the dictionary, and i = 1. 2, 3,・
..., NXM.

Next, the present embodiment will be specifically explained with reference to FIG. 1.

First, a form with characters written on it, for example as shown in FIG. 9, is input as image data S to the photoelectric conversion section 11. The photoelectric conversion unit 1l detects each character area of the form, photoelectrically converts it line by line, converts it into binary digital image data, and stores it in the line buffer l2. In this embodiment, the detection of the row area is performed using the 10th line area set in advance in the photoelectric conversion unit 11.
It is assumed that the steps are performed sequentially with reference to the input format table shown in FIG.

As shown in FIG. 10(b), the input format table records the distance of the first line area from the top and left edges of the form, the size of the line area, the line pitch, and the number of lines.

As shown in FIGS. 2(a) and 2(b), the line image data, which is a binary digital signal read from the line buffer l2, has a distribution of black points in the character cutout portion 13.
Detecting from the line image data from the point where the character changes from 1 or more to the point where the character changes from 1 or more to O as the expected character area,
128 Xl Extracts an input character pattern of 28 pixels.

This input character pattern is stored character by character in the pattern register l4.

The font determining unit l5 that reads the input character pattern from the pattern register l4 determines the 128 X1 of the input character pattern.
Assign an X coordinate and a Y coordinate to each 2g pixel, and calculate the maximum and minimum calculated values αX for the pixels of the input character pattern.
+βY is detected, and feature point coordinates that give these maximum and minimum calculated values are extracted.

In this example, α and β for detecting feature points are (α
= 1, β = 1) and (α = 1, β = -1), and the feature point coordinates based on the calculated values are TL =
(0, O), BL= (0, 4?), TR
= (42, O) '+ BR= (42. 47)
, In Fig. 7(b), TL= (8, O) , B
L= (0.47), TR=(42, O),
BR=(35.47). A feature value F is calculated using equation (1) based on the feature point coordinates.

At this time, in this embodiment, the constants k, j2, and m are respectively (k=station, flood=1, m=1), and the fixed threshold values C,=5.0, C2=3, 0. . The characteristic value calculated under the above conditions is F=0 in FIG. 7(a), and F=7.5 in FIG. 7(b).

Therefore, according to equation (2), when the form shown in Figure 9 is input, the value F shown in Figure 2 (C) is calculated, and for the input character pattern where F>C, the font judgment signal N = l F<C
For input character pattern No. 2, a font determination signal N=2 is output to the dictionary selection section 60 of the dictionary section l6. In this example, M
, Y, N, A, M, E, I, and S, N=2 is output and the first dictionary matrix 6l is selected.

The identification unit 17 extracts sub-patterns into line widths in four directions (horizontal, vertical, diagonal right, and diagonal left) of the input character pattern read from the pattern register l4, and converts each sub-pattern into NXM for the area within the character frame. To divide. In this embodiment, it is 5×5. In each region, a feature amount representing the character line length is calculated to obtain a feature matrix.

This feature matrix and FIGS. 8(a) and 8(b)
The character name (jIs code, etc.) corresponding to the dictionary matrix that shows the largest similarity value is shown through the output terminal l8 by comparing it with the dictionary matrix that has been selected after font determination, as shown in FIG. Output to an external device that is not available.

(Effects of the Invention) As explained above, according to the present invention, the maximum and minimum calculated values of αX+2 7 2 8 βY and αX-βY are calculated for each pixel having a character line of each character on a form. death,
By extracting feature point coordinates, italicized characters such as italics are detected, the font of the input character pattern is determined, and a dictionary mask for comparison is selected based on the determination result. Therefore, since the dictionary mask is compared only with the selected dictionary mask, the time required for the comparison is shortened, and it is possible to realize a character recognition device that can read documents containing a mixture of characters with different fonts at high speed and with high accuracy.

[Brief explanation of drawings]

Figure 1 is a block diagram showing one embodiment of the present invention, Figure 2 (
a) is a diagram showing binary image data in a character line area, FIG. 2(b) is a diagram showing peripheral distribution by binary image data,
FIG. 2(c) is a diagram showing the feature value F of the input character pattern,
FIG. 2(d) is a diagram showing the recognition result for the input character pattern, FIG. 3 is a block diagram showing the font determining unit l5 of FIG. 1, and FIG. 4 is a diagram showing the operation of the maximum value coordinate detection means in this embodiment. 5 is a flowchart showing the operation of the minimum value coordinate detection means in this embodiment, FIG. 6 is a block diagram showing the configuration of the dictionary l6 in FIG. 1, and FIG.
a) is a diagram showing the standard type input character pattern and minutiae coordinates of Roman itself, Figure 7 (b) is a diagram showing the italic type input character pattern and minutiae coordinates of Roman itself, and Figure 8 is Roman itself. An explanatory diagram of the standard character pattern of the standard type and the italic type of Roman itself and the dictionary matrix of the characters, Figure 9 is a diagram showing the form in which the characters used in this example are written, and Figure 1O (a) is the input FIG. 10(b), which is a diagram showing an example of a format table, is a diagram illustrating an input format table. 10, 1l, 12, l3, l4, l5, l6, character recognition device, photoelectric conversion unit, line buffer, character extraction unit, pattern register, font determination unit, dictionary unit, 17... identification unit, l8... Output terminal.

Claims (3)

[Claims] (1) Obtain quantized image data by photoelectrically converting the feature extraction target on the medium, cut out a character pattern one character at a time from the image data, judge the font of the cut out input character pattern, and make a judgment. A recognition dictionary mask is selected based on the result, the input character pattern is collated using the selected dictionary mask, and the image data of the input character pattern is extracted one character at a time in a character recognition device that recognizes characters. X coordinate generation means for assigning an X coordinate to a pixel of the input character pattern; Y coordinate generation means for assigning a Y coordinate to a pixel of the image data of the input character pattern cut out character by character; and at least two sets of the X and Y coordinates. Detect maximum and minimum calculated values αX+βY for pixels of the input character pattern having a predetermined pixel value using specific α and β values of the pixels of the input character pattern that give these maximum and minimum calculated values. a coordinate detection means for detecting the X, Y coordinates of the object; and a feature amount calculation means for calculating geometric feature amounts based on the detected X, Y coordinates, and the features calculated by the feature amount calculation means. A font determination unit that determines the font based on the amount, a dictionary unit that selects a recognition dictionary mask corresponding to the determined font, and a character identification unit that uses the selected dictionary mask to match input character patterns and identify characters. 1. A character recognition device comprising: an identification unit. (2) In the coordinate detection means, α=β=1 and α=
1. Detect the maximum and minimum calculated values X+Y and X−Y regarding the pixels of the input character pattern having a predetermined pixel value using the two sets of α and β values with β=−1, and calculate the maximum and 2. The character recognition device according to claim 1, further comprising detecting the X and Y coordinates of pixels of the input character pattern that give the minimum calculated value. (3) In the feature quantity calculation means, the X coordinate of the maximum value of X+Y in the detected coordinates is TRX, the X coordinate of the minimum value is BLX, the X coordinate of the maximum value of X-Y is BRX, and the minimum value x The coordinates are TLX, and the feature amount F is F=k{l(T
LX-BLX)+m(TRX-BRX)} (however, k,
3. The character recognition device according to claim 2, wherein l and m are arbitrary constants.