JP2918363B2 - Character classification method and character recognition 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 classification method and a character recognition device.

[0002]

2. Description of the Related Art Various advantages can be obtained if a machine can automatically recognize a character or graphic, for example, data can be input to a computer more efficiently and more accurately than a human. For this reason, research on a character recognition device has been actively conducted.

[0003] The conventional character recognition apparatus is generally performed in the following procedure.

First, an optical signal obtained by scanning a medium on which characters, figures, and the like are written, for example, a form, is photoelectrically converted, and further, a character line portion is expressed by, for example, a black bit and a background portion is expressed by a binary bit. To get the input character line data. Next, a character pattern is cut out from the input character line data. Further, feature extraction is performed from this character pattern, and this feature is compared with the feature of a standard character prepared in advance, and the character name of the standard character pattern having the highest similarity is output as the recognition result character name of the recognized character.

However, when such a character recognition device recognizes an English document or a character line such as a name or address written in English characters, a comma "," and an apostrophe "'" are included in the character line. In addition, since characters of exactly the same shape, such as uppercase "P" and lowercase "p", are mixed, there is a problem that character recognition cannot be performed accurately with only the character shape.

In order to solve this problem, a method has been used in which character recognition is performed using the size of the character pattern and the relative position of the character pattern in the character line in addition to the character shape of the character pattern. As this kind of method, for example, there is a method disclosed in a literature; 1988, IEICE Spring National Convention D448.

According to this method, first, a rectangular frame circumscribing a character is extracted from a character line. Next, a circumscribed rectangular frame of each character in the character line is compared, and a character extremely small as compared with the largest character is removed. Next, a histogram based on the height positions of the upper and lower ends of the circumscribed rectangular frame of the remaining characters is created. Next, from this histogram, the peak at the lowest position at the upper end of the rectangle and the peak at the highest position at the lower end of the rectangle are detected, and the coordinates of the upper and lower ends of the character having substantially the same size as the distance between these peaks are used. A straight line giving the inclination of the character line is obtained by the least square method. next,
After the deviation of the character height due to the skew is corrected from the obtained inclination of the straight line, a histogram is created again in the same manner as described above. Next, two peaks are detected from this histogram in the same manner as described above, and these peaks are used as an upper reference line and a lower reference line. Next, the distance between the upper and lower reference lines is regarded as a character of the reference size, the size of each character pattern in the character line is compared with the size of this reference size character, and the position of each character pattern is set to the upper and lower positions. Compared to the lower reference line. Each character of the character line is classified into a plurality of categories based on the comparison result, thereby improving recognition accuracy.

[0008]

However, in the above-described method, it is necessary to perform a series of processes such as inputting rectangular information, removing small characters, correcting line inclination, and calculating a reference line. Was. In addition, since the calculation is performed by the least square method for the correction of the inclination, etc., even for a character line that is slightly inclined, it takes a considerable time to classify each character, and at the same time, a simple inclination such as frequently appearing in a FAX image or the like. When the reference line is uneven, the reference line cannot be calculated correctly, and the classification method does not function effectively. Therefore, with high accuracy,
Moreover, it has been difficult to realize a character recognition device capable of high-speed processing.

According to the present invention, since the classification speed described above is slow, and the reference line cannot be detected correctly from a document image in which the arrangement of characters is uneven, characters cannot be correctly classified. A character classification method that eliminates the problem that accurate recognition cannot be realized, and that can perform character classification accurately and easily from character lines with irregularities and inclinations, with simple processing, and character processing. It is an object of the present invention to provide a character recognition device that provides high recognition accuracy using the character classification method and that can perform high-speed processing.

[0010]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention provides a character classification method for classifying characters cut out from a character line. , And for the second and subsequent characters, calculate the difference between the upper side position and the lower side position of the circumscribed frame of the character and the immediately preceding character, respectively, and calculate the displacement amount and the displacement direction based on the absolute value of the position difference and the sign. Is calculated, based on the displacement amount, the displacement direction, and the provisional classification value assigned to the character immediately before,
A provisional classification step of assigning two types of provisional classification values to the character, a provisional classification value based on the upper side position of the circumscribed frame and a provisional classification value based on the lower side position of the circumscribed frame; A totaling step of totaling the classification values for each of the two types of temporary classification values given to the character to obtain a frequency for each classification value; and setting each character in the character line in a character height direction in advance. In order to classify by a plurality of reference lines, each reference line is identified by a number, and for the character group A in which the frequency of the provisional classification value based on the position of the upper side of the circumscribed frame is the largest, the frequency of appearance in the document is high. The identification number of the reference line corresponding to the mean line where the frequency of the upper side position of the circumscribed frame of the character group is higher is assigned as the upper side classification value, and the upper side classification value of the other character group is set at the upper side position of the circumscribed frame of the character group Provisional classification value based on the characters A character group B in which the frequency of the provisional classification value based on the position of the lower side of the circumscribed frame is maximum in the character line given in accordance with the magnitude relation with the provisional classification value of A, The identification number of the reference line according to the baseline in which the existence frequency of the circumscribed frame lower side position of the group is higher is given as the lower side classification value, and the lower side classification value of the other character groups is based on the circumscribed frame lower side position of the character group. A new classification result value is provided for the combination of the upper side classification value and the lower side classification value given for each character by giving a magnitude relationship between the provisional classification value and the provisional classification value of the character group B. And a classification step.

[0011]

[0012]

According to another aspect of the present invention, there is provided a character recognition device for inputting image data, a character line cutout unit for cutting out a character line from the input image data, and a character line from within the character line. In a character recognition device including a character cutout unit that cuts out a pattern, a character recognition unit that recognizes a cutout character pattern, and a dictionary unit that stores a dictionary mask for character recognition, a character that classifies characters by the character classification method. A characterizing unit is provided, and character recognition is performed using a dictionary selected based on the character classification result.

[0014]

According to the character classification method of the present invention, in the provisional classification step, for example, the displacement of the position is determined based on the difference between the circumscribed frame position of the character and the immediately preceding character and the temporary classification value of the immediately preceding character. A process of assigning the same temporary classification value to the characters in the range is performed for each of the upper side position and the lower side position of the circumscribed frame, and two types of temporary classification values are assigned to each character. The frequency of each provisional classification value is determined for each provisional classification value. In the classification step, for example, the upper-side classification value of the character group A in which the frequency of the provisional classification value based on the upper-side position of the circumscribed frame is the maximum is set to a reference line ( For example, the character group A is classified into one group by assigning it to a number of a reference line according to the mean line, and the upper side classification value of the remaining character group is provisionally classified based on the position of the upper side of the circumscribed frame of the character group A. And the provisional classification value of the remaining character group. In the classification step, the lower side position of the circumscribed frame is further classified in the same manner to obtain a lower side classification value. Thereafter, each character is re-classified by adding a new classification value to the combination of the two types of classification values, the upper classification value and the lower classification value. Therefore, characters can be accurately classified by simple processing even from character lines having irregularities or inclinations.

Further, according to the character recognition device of the present invention, a character classification unit using this character classification method is provided, and a dictionary is selected based on the classification result. Therefore, a character recognition device with high recognition accuracy and capable of high-speed processing is provided.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A character classification method and a character recognition device according to the present invention will be described below with reference to FIGS.

FIG. 1 is a block diagram showing one embodiment of a character recognition device using the character classification method of the present invention. The character recognition device 100 includes an image input unit 110, a character line cutout unit 12
0, character segmentation unit 130, pattern memory 140, character classification unit 150 using the character classification method of the present invention, dictionary unit 1
60, character recognition unit 170, output terminal 180, control unit 19
With zero. The character classifying unit 150 includes a storage unit 151, a temporary classifying unit 152, a counting unit 153, and a classifying unit 154.
It is the structure which consists of.

The image input unit 110 photoelectrically converts an optical signal S from a form as shown in FIG. 2 in which characters, figures, symbols, and the like (hereinafter, referred to as characters) are stored. A black and white binary quantized electric signal (hereinafter, referred to as form image data) in which pixels and a background portion are represented by white pixels is generated and output to the character line cutout unit 120.

The character line cutout unit 120 is provided in the image input unit 1.
The form image data input from step 10 is sequentially scanned with the character line direction as the main scanning direction and the character string direction as the sub-scanning direction, and a distribution of black pixels is created. From the position changing to “1” or more to the position immediately before the position changing from “1” or more to “0” as a character line, the character line data is sequentially extracted from the character line data.
Output to 30.

In character extracting section 130, the character line data input from character line extracting section 120
A character pattern for each character is cut out, the character pattern data is sequentially stored in the pattern memory 140, and the position information of the character pattern is stored in the storage unit 151 in the character classification unit 150.
And it outputs to provisional classification part 152. This character pattern extraction for each character is performed by scanning the character line data input from the character line extraction unit 120 with the character string direction as the main scanning direction and the character line direction as the sub-scanning direction. This is performed by cutting out from a position where the distribution of the black pixels changes from “0” to “1” or more to a position immediately before changing from “1” or more to “0”. In the present embodiment, the position information of the cut-out character pattern is the upper side coordinate and the lower side coordinate of the circumscribed frame of the character pattern.
Are the absolute coordinate values on the form expressed in the xy coordinate system where the character line direction is the x-axis direction and the character string direction is the y-axis direction.

The pattern memory 140 includes a character extracting unit 1
This is a storage unit for sequentially storing character pattern data cut out by 30 and can be easily configured with an IC memory or the like. In this embodiment, the pattern memory stores 12 characters per character.
It has a capacity of 8 × 128 pixels and can store the character pattern data so that it can be reproduced in a two-dimensional format.

In the storage unit 151 in the character classifying unit 150, position information including the upper y-coordinate of the circumscribed frame and the lower y-coordinate of the circumscribed frame of the extracted character pattern input from the character extraction unit 130, and a temporary information described later. This is a storage unit for storing, for each character line, a provisional classification value assigned by the classification unit 152 for each character pattern for each character pattern constituting the character line. For example, for the first line of the form shown in FIG. 2, a character pattern for each character is cut out by the character cutout unit 130. In this case,
As shown in FIG. 3, a temporary classification value for the upper side of the circumscribing frame and a temporary classification value for the lower side of the circumscribing frame and a temporary classification value for the upper side of the circumscribing frame corresponding to each character of the character line are stored in the storage unit 1.
51 is stored in a table format.

The provisional classification unit 152 includes a character extracting unit 130.
Based on the input character pattern position information of the character line unit, each character pattern is classified and a provisional classification value is assigned,
While being stored in the storage unit 151 in association with each character pattern, the assigned temporary classification value is output to the totaling unit 153.

The provision method of the provisional classification value is as follows.

First, 5 is assigned to the character line head character pattern as a provisional classification value.

Next, for the second and subsequent characters of the character line, the magnitude of the displacement in the y-axis direction is determined from the difference between the immediately preceding character pattern and the upper-side position of the circumscribed frame or the lower-side position of the character pattern. The direction is obtained, and a provisional classification value is determined by equation (1).

[0027]

(Equation 1)

Where y _s = y _n −y _n−1 , where y _n is the y coordinate value of the upper or lower side of the circumscribed frame of the character pattern, y
_n-1 is the y coordinate value of the upper side or lower side of the circumscribed frame of the character pattern immediately before the character pattern. C _n is a provisional classification value of the character pattern, and C _n-1 is a provisional classification value of the character pattern immediately before the character pattern.

The value of T is a predetermined threshold value set in accordance with the size of the character to be recognized on the image memory and the design of the character font to be recognized.
Any value larger than / 2 and smaller than the descender height may be set. In this embodiment, T = 15.

From the first character line of the form as shown in FIG. 2, a provisional classification value as shown in FIG. 3 is obtained for each of the upper side position and the lower side position of the circumscribed frame.

In the counting section 153, the provisional classification section 152
Of the provisional classification values of the respective character patterns of the character line input from, the number of character patterns of the provisional classification values existing in the character line is counted, and the counted result is classified into the classification unit 15.
4 is output. From the provisional classification values as shown in FIG. 3, the counting results of the provisional classification values for each of the upper and lower sides of the circumscribed frame as shown in FIG. 4 are obtained. here,
The upper-side frequency in FIG. 4 is the total result value of the provisional classification value based on the upper-side position in FIG. 3, and the lower-side frequency is the total result value of the provisional classification value based on the lower-side position in FIG. .

The classifying unit 154 re-classifies the character based on the provisional classification value of each character pattern of the character line stored in the storage unit 151 and the tabulation result input from the tabulation unit 153. , And outputs the classification result to the dictionary unit 160.

The reclassification is performed by first obtaining a classification value representing the position of each of the upper side position and the lower side position of the circumscribed frame, and then classifying the character pattern based on the classification values of the upper side position and the lower side position. Do.

A method for obtaining the classification value for each of the upper side position and the lower side position of the circumscribed frame will be described below.

Each line numbered from 1 to 4 in FIG. 5B is provided at a position similar to the ascender line, mean line, base line, descender line used in English and the like. This is a reference line for classifying the upper end position and the lower end position of the circumscribed frame of the character pattern by these 1 to 4 lines. Here, the above-mentioned body height means the interval from line 2 to line 3 in FIG. 5B, that is, the standard height of lowercase alphabetic characters, and the descender heights are lines 3 and 4. Is the height of the descender part between

Regarding the position of the upper side of the circumscribed frame, the classification result of the character pattern having the provisional classification value having the highest frequency is set to 2 as a result of the totaling input from the totaling unit, and the provisional classification value of a value smaller than the provisional classification value is set to 2. A classification value of 1 is assigned to a character pattern having a classification value.
A classification value 3 is assigned to a character pattern having a temporary classification value larger than the temporary classification value.

Here, the reason why the classification value of the character pattern having the most frequent provisional classification value is 2 is that in a normal English document, the characters with high appearance frequency are e, s, r, and
n, a, o, i, u, p, y, etc., due to the extremely high frequency at which the upper side of the circumscribed frame of these characters is located near line 2 (the number of the reference line that is similar to the mean line). .

In the case of FIG. 4, the provisional classification value having the largest upper side frequency is 6, so in FIG. 3, the character pattern having the provisional classification value 6 at the upper side position is assigned the classification value 2 and the upper side is assigned the classification value 2. A character pattern with a provisional classification value of 5 or less is assigned a classification value of 1, and a character pattern with an upper side provisional classification value of 7 or more is assigned a classification value of 3.

Regarding the position of the lower side of the circumscribed frame, the classification value of the character pattern having the provisional classification value with the highest lower side frequency is set to 3 in the totaling result input from the totaling unit (this is the same as in the case described above). Because the frequency of the lower side of the circumscribed frame of the character having a high frequency of appearance is the highest in the vicinity of the reference line, that is, the line 3, which is similar to the base line), and has a temporary classification value smaller than the temporary classification value. A classification value of 2 is assigned to a character pattern, and a classification value of 4 is assigned to a character pattern having a temporary classification value larger than the temporary classification value.

In FIG. 4, the provisional classification value at which the lower side frequency is the maximum is 5, so in FIG. 3, the classification value 3 is assigned to the character pattern having the provisional classification value 5 on the lower side, and A classification value of 2 is assigned to a character pattern with a provisional classification value of 4 or less, and a classification value of 4 is assigned to a character pattern with a provisional classification value of 6 or more on the lower side.

When the classification values for the upper side and the lower side of the circumscribed frame are obtained, this embodiment further obtains the classification result value of the character pattern according to the table shown in FIG.
For example, a character [M] having an upper side classification value of 1 and a lower side classification value of 3 is assigned a classification result value of 2, and an upper side classification value of 2 and a lower side classification value of 4 A classification value 5 is assigned to a certain character [y].

By such processing, the classification unit 154 obtains a classification result for each character pattern as shown in FIG.

The dictionary unit 160 includes a character recognition unit 17 described later.
This stores the characteristic matrix of the standard character pattern used in the case of 0, as shown in FIG.
(F) is composed of six dictionaries classified into each of the six categories, and the classification unit 154 in the character classification unit 150
Classification result value (1
7) selecting a dictionary to be collated based on the above (7), and outputting a reference character mask of the selected dictionary to the character recognition unit 170.

The selection of the dictionary is determined based on a table as shown in FIG. For example, for the character pattern [M] of the first character in FIG. 3, since the classification value 2 is input, the dictionary matrix of the dictionaries (A) and (F) is output to the character recognition unit 170.

The character recognizing section 170 sequentially reads the character pattern data from the pattern memory 140, extracts a characteristic matrix from the character pattern data,
The feature matrix is compared with the dictionary matrix data to be matched corresponding to the character pattern data input from 60, and one or more candidate characters are obtained, and the candidate characters are output to the output terminal 180 as a recognition result. .

The extraction of the feature matrix from the character pattern data is performed as follows.

First, a sub-pattern is extracted from the target character pattern data. The character pattern data is scanned in a plurality of directions, and a black pixel row continuous over a predetermined value h (h = 5 in this embodiment) is detected on each scanning line, and the continuous black pixel row is detected. By extracting the sub-pattern as a black pixel component, a sub-pattern for each scanning direction is extracted from the character pattern.

In the present embodiment, the scanning direction is a direction (vertical direction) perpendicular to the character line direction (hereinafter, X-axis direction), a direction parallel to the character line direction (horizontal direction), and 45 ° counterclockwise from the X-axis. The direction is set to the direction (diagonally left) and the direction of 45 ° clockwise (diagonally left), and a character pattern is scanned in each direction to extract four sub-patterns for each direction.

For example, in the extraction of the vertical sub-pattern, the vertical direction is set as the main scanning direction, and continuous black pixels (black runs) are detected on the vertical scanning line of the pattern register.
The vertical sub-pattern is extracted by extracting a black run of length 1 as a black pixel portion of the vertical sub-pattern.

Similarly to the vertical sub-pattern extraction,
Sub-pattern extraction when the remaining other directions are set as the main scanning direction is also performed.

Next, a rectangular area corresponding to the character circumscribed frame of the character pattern is set on the extracted sub-patterns in each direction, and N × M square areas (N and M are arbitrary suitable natural numbers) are set. of divided into small regions, extracting a feature value representing the length of the character lines in each sub-patterns included in each small area, normalizing the feature amount of the character bounding box with the size, characteristics consisting of feature amounts f _i Create a matrix F.

In this embodiment, the number of divisions is N, M =
8, and the feature quantity is normalized by a value of (dX + dY) / 2. Here, dX is the horizontal length of the character circumscribed frame, and dY is the vertical length of the character circumscribed frame. The feature value f _i is the number i (i = 1,2, ..., N × M) up to 1 to N × M in each small area when representing the subregions sequentially subjected, the number i It represents the feature amount of the small area and is an element value of the feature matrix F.

The matching between the feature matrix and the dictionary matrix is performed by calculating the distance R between the feature matrix F and the dictionary matrix G by the following equation (2).
This is performed by setting the character names of the following dictionary matrix as candidate character names, and ranking the first and second candidate characters in ascending order of distance to obtain a candidate character string.

[0054]

(Equation 2)

Here, gi represents an element of the dictionary matrix.

The control unit 190 controls each unit constituting the character recognition apparatus 100, controls the operation of each unit and data synchronization, and controls the interface with the outside through a control signal line (not shown).

The output terminal 109 is a data output terminal for outputting a recognition result to the outside, and connects other systems, a medium for recording the recognition result, a communication network, other information processing systems, and the like. .

It should be noted that the present invention is not limited to only the above-described embodiments, and the operation of each component, the manner of processing, the flow of input / output signals, the number of arrangements, the position, the shape, the number, and other conditions are not limited. Any suitable changes can be made.

For example, the character classifying section of this embodiment is configured to first classify the upper and lower positions of the circumscribed frame of a character pattern, and to obtain a classification result value of the character pattern by combining the two. However, the present invention is not limited to this, and the classification based on only the upper side position of the circumscribed frame or the classification based only on the lower side position may be used as the classification result of the character pattern.

In the character recognition apparatus of this embodiment, the dictionary is switched based on the classification result by the character classification method of the present invention. However, the present invention is not limited to this. The configuration using the classification result for switching, switching of the character pattern cutting method, or the like may be used.

The image input unit is configured to photoelectrically convert an optical signal from a form to obtain form image data. However, the present invention is not limited to this. For example, compressed image data transmitted by facsimile or the like may be used. An image input unit having a function as a decoding unit that develops the form image data to obtain the form image data may be used.

In addition, it is apparent that a method of extracting a character line, a method of extracting a character, a method of configuring a pattern memory, a method of recognizing a character, a configuration of a dictionary unit, and the like may be freely configured within the scope of the present invention. .

[0063]

As described above in detail, according to the present invention, in a character classification method for classifying characters cut out from a character line, a predetermined temporary classification value is assigned to the first character of the character line. For the second and subsequent characters, the difference between the upper side position and the lower side position of the circumscribed frame between the character and the immediately preceding character is calculated, and the displacement amount and the displacement direction are calculated based on the absolute value of the position difference and the sign. Based on the displacement amount, the displacement direction, and the provisional classification value assigned to the character immediately before, the provisional classification value based on the upper side position of the circumscribed frame and the provisional classification value based on the lower side position of the circumscribed frame for the character are obtained. A provisional classification step of providing two types of provisional classification values of the classification values, and a classification value for each of the two types of provisional classification values assigned to each character in the character line. Tabulation step to find the frequency of each sentence and each sentence in the character line In order to classify by a plurality of reference lines set in advance in the character height direction, each reference line is identified by a number, and a character group A in which the frequency of the provisional classification value based on the position of the upper side of the circumscribed frame is the largest. On the other hand, the identification number of the reference line corresponding to the mean line where the frequency of the upper side of the circumscribed frame of the character group having a high appearance frequency in the document is high is given as the upper side classification value, and the upper side classification value of the other character groups is given. Is given by the magnitude relationship between the provisional classification value based on the position of the upper side of the circumscribed frame of the character group and the provisional classification value of the character group A,
For the character group B in which the frequency of the provisional classification value based on the lower side position of the circumscribing frame in the character line is the largest, the base line in which the frequency of the lower side position of the circumscribing frame of the character group that appears frequently in the document increases The identification number of the reference line according to is assigned as the lower side classification value, and the lower side classification value of the other character group is a provisional classification value based on the position of the lower side of the circumscribed frame of the character group and the provisional classification value of the character group B. And a new classification result value is provided for the combination of the upper side classification value and the lower side classification value provided for each character, so that characters can be classified by simple processing. It is also possible to accurately classify character patterns cut out from inclined character lines or character lines in which the reference line is uneven.

Therefore, when this character classification method is applied to a character recognition device, character classification is performed accurately.
A high-performance character recognition device with high recognition accuracy and high processing speed can be realized.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing an embodiment of a character recognition device using the character classification device of the present invention.

FIG. 2 is a diagram illustrating an example of a form.

FIG. 3 is a diagram provided for explanation of a character classification unit.

FIG. 4 is a diagram provided for explanation of a tallying unit.

FIG. 5 is a diagram provided for explanation of a classification unit.

FIG. 6 is a diagram provided for explanation of a dictionary unit. .

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 character recognition device 110 image input unit 120 character line cutout unit 130 character cutout unit 140 pattern memory 150 character classification unit 151 storage unit 152 provisional classification unit 153 counting unit 154 classification unit 160 dictionary unit 170 character recognition unit 180 output terminal 190 control unit

Claims (2)

(57) [Claims] 1. A character classification method for classifying characters cut out from a character line, wherein a predetermined temporary classification value is assigned to the first character of the character line, The difference between the upper side position and the lower side position of the circumscribed frame of the character is calculated, the displacement amount and the displacement direction are obtained from the absolute value of the position difference and the sign, and the displacement amount, the displacement direction, and the immediately preceding character are assigned. Two types of provisional classification values are assigned to the character based on the provisional classification value obtained, that is, a provisional classification value based on the upper side position of the circumscribed frame and a provisional classification value based on the lower side position of the circumscribed frame. A provisional classification step; a totalization step of totalizing the classification values for each of the two types of temporary classification values assigned to each character in the character line to obtain a frequency for each classification value; Characters are set in multiple criteria that are preset in the height direction In order to classify by character line, each reference line is identified by a number, and for the character group A in which the frequency of the provisional classification value based on the position of the upper side of the circumscribed frame is the largest, the character group having a high appearance frequency in the document The identification number of the reference line corresponding to the mean line where the frequency of the upper side of the circumscribed frame is higher is assigned as the upper side classification value, and the upper side classification value of the other character group is provisionally determined based on the upper side position of the circumscribed frame of the character group. For the character group B in which the frequency of the provisional classification value based on the lower side position of the circumscribed frame in the character line is given by the magnitude relation between the classification value and the provisional classification value of the character group A, a document In the character group having a high appearance frequency, the identification number of the reference line according to the baseline in which the existence frequency of the lower side of the circumscribed frame of the character group having a high frequency is given as the lower side classification value, and the lower side classification value of the other character group is Outside A new classification is provided for a combination of the upper side classification value and the lower side classification value given for each character by giving a magnitude relationship between a provisional classification value based on the position of the lower side of the frame and the provisional classification value of the character group B. And a classification step of assigning a result value. 2. An image input unit for inputting image data, a character line cutout unit for cutting out a character line from the input image data, a character cutout unit for cutting out a character pattern from the character line, and recognizing the cutout character pattern. A character recognition device including a character recognition unit and a dictionary unit storing a dictionary mask for character recognition, comprising: a character classification unit configured to classify characters by the character classification method according to claim 1, based on a character classification result. A character recognition device for performing character recognition using a selected dictionary.