JPH03126188A - Character recognizing device - Google Patents

Abstract

PURPOSE:To accurately identify characters with the same character type and different size by calculating the reference line of character row data by using the recognition result of first to n-th characters to be recognized of the character row, and hereafter performing recognition based on the character type of the character and the relative position relation of the coordinate of a character pattern in character row data and that of the reference line. CONSTITUTION:A photoelectric conversion part 112 which obtains the input character row data of the character row on a medium by performing the photoelectric conversion and quantization of an optical signal 111 from the medium, a line buffer 113, a character segmenting part 114, and a pattern register 115 which stores a segmented character pattern are provided. And the reference line of the character row is decided based on the recognition result of the first to n-th characters of the character row when each character on the character row is recognized, and when the characters behind an (n+1)th character is recognized, the recognition is performed based on the character type of the character and the relative relation of the quantizing pattern of the character and the reference line. Thereby, it is possible to accurately identify the characters even when the characters with the same character type and different size are contained.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a character recognition device that can obtain high recognition accuracy.

(Prior Art) If a machine can automatically identify characters and graphics, various advantages can be obtained, such as being able to enter data into a computer more efficiently and accurately than humans can.

For this reason, research into character recognition devices has been actively conducted.

Conventional character recognition devices generally include the following components.

[1]...The optical signal obtained by scanning a medium, such as a form, on which characters, figures, etc. are written is photoelectrically converted, and the text lines are roughly represented by black bits, and the background part is represented by white hits. A photoelectric conversion unit that obtains input character line data.

[2]...Character cutting unit that cuts out a character pattern from this input character line data.

■... Extract the feature amount from this character pattern, compare this feature Jl with the feature amount of the standard character prepared in advance, and select the character name of the standard character pattern with the highest degree of similarity as the recognition result character of the character to be recognized. Recognizer that outputs as a name.

However, when such a character recognition device recognizes a sentence in a foreign language, such as English, or a line of characters written in English, such as a name or address, the first line in the line is a comma [, J, and a post O letter. Also, since there are characters with exactly the same shape, such as the capital letter "P" and the lowercase letter "PJ," there is a problem that character recognition cannot be performed accurately based only on the shape of the character pattern. there were.

Therefore, in order to solve this problem, a method has been used in which literary journal m is performed using the size of the character pattern and the relative position Mlfr of the character pattern in the character line in addition to the shape of the character pattern.

Examples of this type of method include the literature (1988 Institute of Electronics, Information and Communication Engineers Spring National Conference (March 15, 1988) D-4
48) was disclosed.

According to the method disclosed in this document, first, a rectangular frame circumscribing a character is extracted from a character line. Next, the circumscribing rectangular frames of each character in the character line are compared, and the rectangular frame circumscribing each character in the character line is compared. Extremely small characters are removed Next, a histogram is created based on the height positions of the top and bottom edges of the circumscribed rectangle of the remaining characters. From this histogram, the lowest position at the top of the rectangle is The beak and the peak located at the highest position at the bottom of the rectangle are detected, and a straight line giving the slope of the character line is found by the least squares method using the coordinates of the upper and lower ends of the character whose size is approximately the same as the distance between these beaks. Next, the deviation in character height due to skew is corrected from the slope of the obtained M line, and then a histogram is created again in the same manner as described above. Next, from this histogram, two peaks are detected in the same way as explained earlier, and these and a portion are taken as the upper reference line and the lower reference line. Next, the distance between these upper and lower reference lines is is set as a standard size character, and the size of each character in the character line is compared with the size of this standard size character, and the position of each character in the line is compared with the upper and lower reference lines. Based on the results of this comparison, each character in the character line is classified into multiple categories, thereby improving recognition accuracy.

(Problem to be Solved by the Invention) However, the conventional character recognition method disclosed in the above-mentioned literature targets all characters in a character line, inputs rectangular information, removes lowercase characters, and corrects line inclination. Since it is necessary to perform a series of processes such as rough reference line calculation, there is a problem in that the recognition speed is significantly reduced.

In addition, in the case of short character lines consisting of one or two characters, there is very little data to create a histogram, so it is not possible to accurately detect the peak that will serve as the reference line. cannot be applied to short character lines, and even if it were applied, there was a problem in that the recognition accuracy would actually decrease.

In addition, in the case of a character line consisting of characters that are mostly the same size, the unevenness in the histogram is small, so the reference line and reference size cannot be detected, and therefore accurate character recognition cannot be performed. The problem was that there was no.

The present invention has been made in view of the above points, and therefore, an object of the present invention is to provide a character recognition device M that can solve the above-mentioned problems and can recognize characters of the same shape accurately and at high speed.
Our goal is to provide the following.

(Means for solving the problem) In order to achieve this objective, the inventor of this application
We have repeatedly considered various issues. As a result, when recognizing each character in a character line, the standard II for this character line is determined based on the recognition results from the beginning of the character line to the nth character, and when recognizing characters from the n+1st character onward, the It was concluded that the object of the present invention can be achieved if recognition is performed based on the MIWi relationship of the character shape and the quantization pattern (character pattern) of the character relative to the reference line.

Therefore, according to the present invention, there is provided a photoelectric conversion unit that photoelectrically converts and quantizes light from a medium to obtain input character line data of a character line on the medium, a character cutting unit that cuts out a character pattern from the input character line data, and a character cutting unit that cuts out a character pattern from the input character line data, and In a character recognition device comprising a recognition unit that extracts features of a pattern and outputs a recognition result character name of a recognized character, the recognition unit described above is configured to determine a recognition result character name according to (A) and (B) below. A character recognition device characterized by:

(A) For the characters to be recognized from the beginning of the character line to the n is any positive integer).

(B) For the characters to be recognized after the (n+1)th character from the beginning of the character line mentioned above, each character is: [1]...Features including features resulting from the shape of the character pattern of the character to be recognized, and [2]... The above-mentioned recognition result of the characters up to the n-th character is calculated using some or all of the predetermined coefficients corresponding to the characters and the coordinates in the input character line data of the character pattern corresponding to the coefficients to be used. A recognition result character name is determined based on the relative position of the reference line coordinates of the input character line data and the coordinates of the character pattern of the character to be recognized in the input character line data.

Here, the features that include features resulting from the shape of the character pattern of the recognized character include, for example, features of only the shape of the recognized character, features due to the size of the recognized character, and features of the character shape and size of the recognized character. It refers to the characteristics of each.

In implementing this invention, the above-mentioned recognition section is replaced with a candidate character name extraction section that extracts a candidate character name based on the characteristics resulting from the shape of the character pattern of the character to be recognized, and the above-mentioned extracted candidate character name. It is preferable that the character name determination unit is configured to include a character name determination unit that determines a recognition result character name according to (a) and (b) below.

(a) For the above-mentioned characters up to the n-th character, the candidate character name with the greatest degree of similarity to the character to be recognized among the candidate character names is set as the recognition result character name.

(b) For the n+1st and subsequent characters to be recognized, calculate the positional characteristics of the character pattern in the input character line data based on the reference line coordinates and the coordinates of the character pattern of the character to be recognized; The feature of the position is compared with a predetermined value related to the candidate character name of the character to be recognized, and if the predetermined condition is satisfied, the candidate character name is set as the recognition result character name; A comparison is made between the predetermined value related to the candidate character name and the above-mentioned positional features for the candidate character names ranked 3rd or lower, and the aforementioned predetermined conditions are satisfied among the candidate character names ranked 3rd or lower. The candidate character name obtained is set as the recognition result character name.

(Operation) According to the character recognition device of the present invention, n characters from the beginning of a character line are
Using the recognition results of up to the recognition character, calculate a reference line of character positions used to determine whether candidate character names of the n+1st and subsequent recognition characters are appropriate as recognition result character names,
The recognition result character name can be determined based on this reference line and the obtained character pattern data of the character to be recognized. For this reason, for example, the capital letter "P" is included in the characters to be recognized after the n+1st
Even if characters with the same shape but different sizes are included, such as ``'', lowercase rP, J, etc., the two can be accurately identified.

Therefore, recognition accuracy for the entire character line can be improved.

However, since the reference line is calculated only by using one or more characters at the beginning of a character line, the time required to calculate the reference line can be extremely shortened compared to the conventional method. Therefore, the recognition time can be shortened.

(Embodiments) Hereinafter, embodiments of the character recognition device of the present invention will be described with reference to the drawings.

Figure 1 is a diagram schematically showing the configuration of a character recognition device according to an embodiment.

In FIG. 1, 100 is a character recognition device, 111 is an optical signal from a medium (for example, a form), and 112 is a photoelectric conversion and quantization of the optical signal 111 from the medium to obtain input character line data for character lines on the medium. 113 is a photoelectric conversion unit; 113 is a line buffer for storing this input character line data; 114 is a character cutting unit that cuts out a character pattern from the input character line data in the line buffer; and 115 is a pattern register that stores the cut out character pattern. These photoelectric conversion unit 112, line buffer 113, and character cutout g are shown respectively.
E+14 and pattern register 115 are each constructed from conventionally known circuits. In this embodiment, the line buffer 113 is composed of a memory having a capacity of 128 x 4096 pixels, and the pattern register 115 is composed of a memory having a capacity of 128 x 4096 pixels.
It consists of a memory having a capacity of 28 x 512 pixels.

Briefly, in FIG. 1, reference numeral 116 indicates a recognition unit according to the present invention. This recognition unit 116 is configured to determine a recognition result character name according to (A) and (B) below.

(A) For each character to be recognized from the beginning to the nth character in a character line, a recognition result character name is determined for each character based on features including features resulting from the shape of the character pattern of the character to be recognized.

(B) For the characters to be recognized after the (n+1)th character from the beginning of the character line, [1]...Characteristics due to the shape of the character pattern of the character to be recognized, character by character, and [2]...Up to the nth character above. The standard of the input character line data calculated using some or all of the predetermined coefficients corresponding to each character name and the coordinates in the input character line data of the character pattern corresponding to the coefficient used as the core. A recognition result character name is determined based on the relative position of the line coordinates and the coordinates of the character pattern of the character to be recognized in the input character line data.

In order to facilitate the processing of (A) and (B) above, the recognition unit 116 of this embodiment extracts candidate character names based on the features resulting from the shape of the character pattern of the character to be recognized. A character name extraction unit 117, a character position determination unit 118a, a character position feature table 118b, and a reference line calculation unit for determining recognition result character names from the extracted candidate character names according to (a) and (b) below. 118C and a character name determining section 118 comprising a reference line storage section ++8 dM.

(a) For the characters to be recognized from the beginning of the character line to the n-th character, the candidate character name with the greatest degree of similarity to the character to be recognized among the candidate character names is set as the recognition result character name.

(b) For the characters to be recognized after the n+1th character from the beginning of the character line, the characteristics of the position of the character pattern in the input character line data are based on the reference line coordinates and the coordinates of the character pattern of the character to be recognized. is calculated, and the characteristics of the position are compared with a predetermined value related to the candidate character name of the character to be recognized, and if the predetermined condition is satisfied, the candidate character name is set as the recognition result character name; In this case, the predetermined value related to the candidate character name and the characteristics of the position are compared for the candidate character name in the third place or below, and the predetermined condition is determined for the candidate character name in the third place or below. The satisfied candidate character name is set as the recognition result character name.

Here, the reference line calculation table 118c stores the above-mentioned predetermined coefficients corresponding to the recognition result character names. Further, the character position feature table 118b stores the above-mentioned predetermined value that is incorrect as a candidate character name of a character to be recognized.

Briefly, in FIG. 1, reference numeral 119 indicates a character name output terminal for outputting the character name determined by the character name determining section 118 to an external console, display screen, etc.

In order to deepen the understanding of the character recognition device of the example, we will refer to Figure 1, Attached Table 1, Attachment 2, Figure 2 (A) and CB), and Attachment 3 to understand the character recognition device of the example. The operation of the recognition device will be explained. Here, Appendix 1 shows the reference line calculation table 118c.
Attached Table 2, which provides an explanation of the character position feature table +
18b is a table for explaining the input character line data 21 stored in the line buffer. FIG. 2(B) is a table for explaining the input character line data 21 stored in the line buffer. Table 3 is a table for explaining candidate character names and recognition result character names when the character to be recognized is the lowercase letter "P".

First, write down the characters, figures, etc. (hereinafter simply referred to as characters)!
An optical signal 111 from the generated form is input to a photoelectric conversion unit 112. The photoelectric conversion unit 112 photoelectrically converts this optical signal 111 into a 247 m digital signal in which the character line portion is expressed by black bits and the background portion is expressed by white bits (this signal corresponds to input character line data). This input character line data is stored in the line buffer 113.

The line buffer 113 stores input character line data input from the photoelectric conversion unit 112 in a format that allows reproduction of two-dimensional coordinates. FIG. 2A visually shows the input character line data 21 stored in the line buffer 113.

Next, the character cutting unit 114 reads the input character line data from the line buffer 113 and stores it in a direction perpendicular to the character line (direction indicated by Y in FIG. (referred to as "direction") is the main scanning direction, and scanning is performed sequentially from the left end to the right end, and the number of black hits in each column is counted to create a histogram of black bits. The rough character segmentation unit 114 examines the created histogram, and extracts character pattern data from end areas in which columns in which the number of black bits is equal to or greater than a predetermined first threshold of 8 are continuous for a predetermined second threshold or more. and extract this as pattern register 115.
Store in. Furthermore, when storing the character pattern data in the pattern register 115, the character cutting unit 114 also generates a character pattern number m (see FIG. See A))
, the coordinates Yt of the highest point and the coordinates Y of the lowest point on the line buffer 113 of the character pattern are output to the character position determining section 118a. In addition, in the case of this example, the first
The character pattern data was extracted by setting the threshold value B to 1 and setting the second threshold value B to 5. Further, the coordinates Yt and Yb are indicated by absolute coordinates (Y coordinates in FIG. 2(A)) assigned in the line buffer 113.

The pattern register 115 that receives the character pattern data output from the character cutting section 114 stores the character pattern data in a format that can reproduce its two-dimensional coordinates.

Next, the candidate character name extraction unit 117 of the recognition unit 116 reads the character pattern data stored in the pattern register 115, extracts the characteristics of this data using a predetermined method, and creates a feature matrix. Furthermore, this feature matrix and
The character name determination unit 11 calculates the degree of similarity between standard character patterns prepared in advance and the dictionary matrix, and selects character names from the dictionary matrix in descending order of degree of similarity as candidate character names.
8 to the character position determination unit 118a. Note that in this embodiment, K=5. Here, extraction of features from character pattern data can be performed using various conventionally known methods, but in this embodiment, the extraction was performed using the method described below.

First, a rectangular frame, for example, circumscribing a character line portion is detected using character pattern data.

Next, the line width W% of this character pattern is calculated using a well-known approximation formula shown by the following formula (1).

W= 1/ (1-Q/A)...(1) where (1)
In the formula, Q is the number of windows in which all four pixels within this window are black bits when a character pattern is viewed through a 2 x 2 window, and A is the number of all black bits in the character pattern. be.

Next, this 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 line width W described above, the number of consecutive black bits is detected. A subpattern corresponding to each direction is extracted. Then, the above-mentioned circumscribed frame of this character pattern is divided into (NXM) regions tft (N, M are constants) for each subpattern, and the feature amount representing the character line in each divided region is calculated for each divided region. This feature quantity is normalized by the size of the character frame to obtain a feature matrix.

In this embodiment, the feature amount is normalized by dividing it by a value of (Δχ+ΔY)/2. Here, ΔX is the length of the circumscribing frame in the horizontal direction, and ΔY is the length of the circumscribing frame in the vertical direction.

Further, in this embodiment, the degree of similarity between the feature matrix obtained in this manner and a dictionary matrix of standard character patterns prepared in advance is calculated according to the following equation (2).

However, in formula (2), day is the degree of similarity, fl is the element value of the feature matrix of the character pattern data of the character to be recognized, and 9. is the element value of the dictionary matrix, and NXM is the number of dimensions of the feature matrix of the character to be recognized and the dictionary matrix, respectively.

Next, the operation of the character name determining section 118 will be explained. In addition, in order to facilitate understanding of this explanation, Fig. 2 (A)
The operation will be explained using an example of processing the input character line data 21 shown in FIG.

The character position determining section 118a of the character name determining section 118 compares the character pattern number m input from the character cutting section 114 with a predetermined specific value n, and operates as described below according to the comparison result. Here, n is a predetermined positive integer, and in this embodiment, n=1.

(a) If it is n5m, that is, if the character to be recognized is the nth character from the beginning of the character line, in the example of Figure 4 (A), if the first character is The position determination unit 118a outputs the character name with the greatest degree of similarity to the recognized character among the candidate character names output from the candidate character name extraction unit 117 to the character name output terminal 119 as a recognition result character name. . Additionally, the character position determination unit 118
a is a predetermined coefficient α (hereinafter sometimes referred to as a reference line calculation coefficient) corresponding to the above-mentioned recognition result character name (character name "■") from the reference line calculation table ll8c (Appendix 1).

) is read out from the reference line calculation table 118c, and this coefficient α and the character pattern “K” corresponding to the recognition result character name are read out.
In this example, the reference line coordinate Y of the input character line data is calculated according to the following formula (3) using the highest point coordinate Y and the lowest point coordinate Yb input from the character cutting section 114. do.

This configuration is such that the character name of each alphabetic character and the reference line calculation coefficient α are registered in advance in association with each other.

The reference line coordinate Y of the input character line data 21 in FIG. 2(A),
Considering, the reference line calculation coefficient α, the highest point coordinate Yt, and the lowest point coordinate Y of the first character pattern rIJ are α=0.0 Yt=98 Yb=30, so the reference line coordinate Y, is: Y,=30+Ox (98-30)=30.

Y, = Yb + α (Yt-'v'b) -...
(3) Next, the character position determination unit 118a stores the calculated reference line coordinate Ys in the reference line storage unit 118d.

Note that the reference line calculation table 118C is for uppercase English letters and lowercase English letters, for example, as shown in Attached Table 1 (b).On the other hand, if m>N, that is, if the recognized character is from the (n+1)th or later position from the beginning of the character line, If it is a character, the character position determination unit 11
8a determines the recognition result character name as described below. This operation will be explained using the example of the lowercase letter "P" which is the third character in FIG. 2(A).

The character position determining unit 118a uses the highest point coordinate Yt and the lowest point coordinate Yb of the character pattern rpJ input from the character cutting unit +14 and the reference line coordinate Y3 stored in the reference line storage unit 118d, A feature e representing the 11th digit of the character pattern "P" is calculated according to the following equation (4).

However, 4) In the formula, Z is a constant, and in this example, Z=
It is set at 10.

Next, the character position determination unit 118a selects the candidate character name extraction unit 1.
Predetermined values (referred to as character position features 9L and 9H) corresponding to the candidate character names inputted from 17 are sequentially stored in the character position feature table 118b in descending order of similarity to the character to be recognized. Read from.

Note that the character position feature table 118b has a structure as shown in Attached Table 2, for example, when considering uppercase English letters and lowercase English letters. 9L and the upper limit coordinate 9N) are registered in advance in association with each other.

Next, the character position determining unit 118a determines the character position characteristics 9. read from the character position characteristic table 118b. and 9H,
The character pattern is compared with the positional feature e calculated according to equation (4). If the comparison result satisfies 9L≦e≦9H, the candidate character name is outputted to the character name output terminal 119 as a recognition result character name.

On the other hand, if the comparison result is 9,〉e or eggH, it is determined that the candidate character name is not a recognition result character name, and the candidate character name with the next highest degree of similarity is selected. The same processing as described above is performed.

Considering the example of determining the recognition result character name of the lowercase letter "P", the highest point coordinate Yt and the lowest point coordinate Y of the character pattern of this character are vt = 50 Y, = 15, and the input character line data Since the reference line coordinate Y, is Y, = 30 as calculated earlier, the character pattern position characteristic e of the lowercase letter "P" is
From equation (4), it becomes.

Further, candidate character names for the lowercase letter "P" are "P", rpJ, "0", rl)J, and rcJ as shown in Appendix 3 in the similarity ranking layer.

Therefore, the character position determining unit 118a first reads the character position features 9L and 9H of the first candidate character name "P" from the character position feature table 118b, and reads these 9L and 9□
and the calculated positional feature e are compared. However, the first
As is clear from Attached Table 2, the candidate character name rpJ for the position is 9L = O and 9H = 1, so the character pattern ``
In relation to the feature e=4.2 of the position of "P", g, < e
Therefore, the character position determination unit 118a
It is determined that the first candidate character name rpJ is not a recognition result character name.

Next, the character position determination unit 118a determines the second candidate character name “
The same process as for the first candidate character name is performed for "P". In this case, the second candidate character name "P" is
As shown in the figure, since 9L=4 and 9H=6, 9L≦e≦9H is satisfied in relation to the character pattern rPJ positional feature e=4.2. Therefore, the character position determination unit 118a determines the second candidate character name [PJ as the recognition result character name, and outputs it to the character name output terminal 119.

The above is the description of the embodiment of the character recognition device of the present invention.

However, the present invention is not limited to the above-described embodiments, and various modifications as described below can be made.

In the above embodiment, the character to be recognized at the beginning of the character line, and the two
This is an example in which the recognition processing is different depending on the characters to be recognized after the first character. That is, the number of characters from the beginning of the character line n8n = 1
This is an example of setting. However, it is clear that the number n can be changed depending on the design. However, when n is set to a value of 2 or more, the reference line coordinate Y in the input character line data
, is preferably calculated as follows, for example.

<First calculation method> As a first method, there is a method in which the average value of the reference line coordinates calculated for each character from the beginning to the nth character is set as the reference line coordinate Y.

For example, for the input character line data 21 of Figure 2 (A), n
To explain using an example where = 2 is set, first calculate Ys+ for the first character "ENG" according to formula (3), then calculate Y, 2 for the second character rnJ according to formula (3).
@Calculate and average these values (Y * I + Y 12)
/2 is the reference line coordinate Y in the case of n=2.

Second calculation method> The second method uses a predetermined coefficient corresponding to the recognition result character name with the highest degree of similarity among the recognition result character names of the respective characters from the beginning to the nth character, and the recognition result character name. There is a method in which the reference line coordinates calculated using the coordinates of the character pattern from which the resulting character name is obtained are set as the reference line coordinates Y.

To explain a specific example of the second calculation method using the same example as the first calculation method, the similarity of the first candidate character name at the time of recognition of the first character "ku" and the second Compare the similarity of the first candidate character name when recognizing the character "n" and consider the first candidate character name of the character with the greater similarity (3)
The reference line coordinate is calculated according to the formula, and this is set as the reference line coordinate Y3 in the case of n=2.

Note that the first and second calculation methods when n is 3 or more can be performed using the same procedure as when n=2.

Further, in the above embodiment, there is only one reference line, and the reference line is used as the baseline. However, in carrying out the present invention, the reference line is not limited to the baseline, and may be other lines such as a descender line, mean line, cap line, or ascender line, and moreover, two or more reference lines may be set. Also good.

(Effects of the Invention) As is clear from the above description, the character recognition device of the present invention can determine the reference line of character line data using the recognition results of the n-th characters from the beginning of the character line. The n+1st and subsequent characters to be recognized are recognized based on the shape of the character, the coordinates of the character pattern of the character in the character line data, and the relative positional relationship between the reference line coordinates of the character line data. , Therefore, among the characters to be recognized after the n+1st character, for example, the uppercase letter "P", the lowercase letter "PJ
Even if characters with the same shape but different sizes are included, such as , the two can be accurately identified.

However, since the reference line is calculated only by using one or more characters at the beginning of the character line, the reference line calculation can be completed in a much shorter time than in the past. In addition, since the reference line is calculated using one or more characters, the reference line can also be calculated from a character line made up of a small number of characters or a character line made up of characters of similar size. These character lines can also be recognized with high accuracy.

Therefore, it is possible to provide a system M that can accurately and quickly recognize characters of the same shape.

Attached table 3

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a character recognition device according to an embodiment, and FIGS. 2(A) and CB) are diagrams for explaining the embodiment. 00... Character recognition device, 111... Optical signal 2...
・Photoelectric conversion section, +13-...Line buffer 4...
・Character cutting section, 115...Pattern register 6...
・Recognition unit, 117...Candidate character name extraction unit 8・
...Character name determination section, 118a...Character position determination section 8
b...Character position feature table 8c...Reference line calculation table 8d...Reference line storage section, 119...Character name output terminal 21...Input character line data 23...Reference line.

Claims (5)

[Claims] (1) A photoelectric conversion unit that photoelectrically converts and quantizes light from a medium to obtain input character line data of character lines on the medium, a character cutting unit that cuts out a character pattern from the input character line data, and characteristics of the character pattern. A character recognition device comprising a recognition unit that extracts and outputs a recognition result character name of a character to be recognized, characterized in that the recognition unit is configured to determine a recognition result character name according to (A) and (B) below. character recognition device. (A) For the characters to be recognized from the beginning of the character line to the n is any positive integer). (B) For each character to be recognized after the n+1th character from the beginning of the character line, [1]...features including features resulting from the shape of the character pattern of the character to be recognized, and [2]...the above. The input character line calculated using some or all of the predetermined coefficients corresponding to each of the character names as a result of recognition of up to the nth character, and the coordinates in the input character line data of the character pattern corresponding to the coefficients used. A recognition result character name is determined based on the relative position between the reference line coordinates of the data and the coordinates of the character pattern of the character to be recognized in the input character line data. (2) The character recognition device according to claim 1, wherein the recognition unit includes a candidate character name extraction unit that extracts a candidate character name based on a feature resulting from a character shape of a character pattern of a character to be recognized;
A character recognition device comprising: a character name determining unit that determines a recognition result character name from the extracted candidate character names according to (a) and (b) below. (a) For the characters up to the n-th character, among the candidate character names, the candidate character name with the greatest degree of similarity to the recognized character is set as the recognition result character name. (b) For the n+1st and subsequent characters to be recognized, calculate the characteristics of the position of the character pattern in the input character line data based on the reference line coordinates and the coordinates of the character pattern of the character to be recognized, and calculate the position of the character pattern in the input character line data; Compare the characteristics of the character with a predetermined value corresponding to the candidate character name of the character to be recognized, and if the predetermined conditions are satisfied, the candidate character name is set as the recognition result character name, and on the other hand, if the characteristics are not satisfied, the second place For the following candidate character names, a comparison is made between a predetermined value related to the candidate character name and the characteristics of the position, and a candidate character name that satisfies the predetermined condition among the second or lower candidate character names is selected. Use the recognition result character name. (3) The character recognition device according to claim 1 or 2, wherein the reference line coordinate is an average value of reference line coordinates calculated for each character from the beginning to the nth character. . (4) In the character recognition device according to claim 1 or 2, the reference line coordinates correspond to the recognition result character name with the highest degree of similarity among the recognition result character names of the respective characters from the beginning to the nth character. A character recognition device characterized in that the reference line coordinates are calculated using a predetermined coefficient and the coordinates of a character pattern from which the recognition result character name is obtained. (5) In the character recognition device according to any one of claims 1 to 4, the reference line is one or more lines selected from a descender line, a baseline, a mean line, a cap line, and an ascender line. A character recognition device characterized by the following.