JPH08123905A - Method and device for character recognition - Google Patents

Abstract

PURPOSE: To provide the method and device which shorten the processing time to attain the processing time of practical font recognition ana improve the precision of font recognition. CONSTITUTION: The picture read by an image scanner 3 is segmented into each character, and its size is normalized. Thereafter, the feature vector is extracted based on the normalized character image. The vector component of the extracted feature vector is divided into plural groups and is decomposed to partial vectors consisting of components of respective groups. Distances from these partial vectors to standard patterns stored in a first dictionary preliminarily stored in a recognition dictionary 9 are calculated, ana the character category of the standard pattern having the highest degree of similarity is obtained. Thereafter, distances between this character category and standard patterns in a second dictionary are calculated to determine a character font.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition apparatus and method for recognizing characters in input image data.

[0002]

2. Description of the Related Art Generally, most character recognition devices only output a character code obtained by recognition. In order to recognize even character fonts, it is necessary to increase the number of dimensions of the feature vector in order to identify subtle differences in the character shape, or to make the identification calculation more complicated.

In the case of character recognition, if the recognition target is limited to English, the target character category is 52 characters and some symbols are added thereto, and the number of categories is about 100 at most.
On the other hand, the types of character fonts that are often used are generally four, such as Courie, Times Roman, Helvetica, and italics. When trying to recognize fonts, the first method that can be considered is to have recognition dictionaries for each font for each font (for example, 4), and each recognition dictionary is created from standard patterns learned from the learning characters of one font. To be Then, a feature vector is extracted from one character image cut out by the character cutout, and the similarity (or distance) with the standard pattern of each dictionary is obtained by a predetermined identification calculation method. The similarity is determined by the number of fonts, that is, four, and the font of the recognition dictionary that gives the maximum similarity is used as the solution. The four similarities obtained here are generally very close values. This is because the difference in fonts makes it difficult to capture the features. Therefore, if the number of dimensions of the feature vector is increased or the identification calculation is made sophisticated (complex), it may be possible to catch the font, but this time, the identification calculation will increase. Further, the amount of calculation reaches the number of times of the number of recognition dictionaries (the number of font types). In the above example, if the similarity calculation load with one standard pattern is ρ, the total calculation amount ρ × 100 × 4 = 400ρ. Since the number of characters is about 3,500, the amount of calculation is ρ × 3,500 × 4 = 14,000ρ, assuming that Japan is the target. This means that the similarity calculation load ρ for catching the case of a font is originally very large, and therefore font recognition cannot be substantially performed in a realizable processing time.

[0004]

SUMMARY OF THE INVENTION The present invention provides an apparatus and method for reducing the processing time and improving the accuracy of font recognition in order to achieve a realistic processing time for font recognition. Also, it is meaningful that the document is output in the same way as it is input. However, due to the widespread use of color printers in recent years, there is a natural need to reprint monochrome documents in color. In particular, there is an increasing desire to colorize a document that has been printed by a conventional monochrome printer and accumulated.
Similarly, the desire to display on a color display also increases. The present invention also addresses these needs as an application of font recognition.

[0005]

[Means for Solving the Problems] and

In order to solve the above problems, for example, the character recognition device of the present invention has the following configuration. That is, a character recognition device that cuts out a character portion in an input document image and recognizes a character, and a feature extraction unit that extracts a feature vector from the cut out character image, and statistical information for determining a character category. A first recognition dictionary that stores a standard pattern that includes the second recognition dictionary that has a standard font pattern for determining a feature conversion matrix and a character font for each character category for each character category; Feature decomposition means for dividing the vector component into a plurality of groups and decomposing into a plurality of partial vectors composed of the components of each group,
First identifying means for calculating a distance between one of the partial vectors and a standard pattern of the first recognition dictionary and outputting a character category of the standard pattern having the highest similarity; and the feature conversion of the determined character category. A second identifying means for converting the feature vector into a new feature vector by a matrix, calculating the distance between the new feature vector and the standard font pattern of the second recognition dictionary, and determining the character font.

According to a preferred embodiment of the present invention,
The statistical information of the standard pattern included in the first recognition dictionary is an average vector, which is necessary for the calculation of the pseudo pay discriminant equation,
It is desirable to include eigenvalues, eigenvectors, and higher-order eigenvalue replacement parameters. Further, it is preferable that the standard font pattern of the second recognition dictionary includes an average vector of new feature vectors, an eigenvalue, an eigenvector, and a high-order eigenvalue replacement parameter, which are necessary for the calculation of the pseudo-pause identification formula.

Furthermore, after the character category and the character font are determined, there is a character information output stage for outputting the character code in association with the font type. Different character information has different colors for different fonts. It is desirable to display and / or print at.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An example of a real sister according to the present invention will be described in detail below with reference to the accompanying drawings. FIG. 1 is a block diagram of the character recognition device in the embodiment. In the figure, reference numeral 1 is a CPU that controls the entire apparatus, and also functions as a processing unit that performs processing such as various calculations. 2 is a bus, 3 is an image scanner,
4 is a RAM, 5 is a feature extraction unit, 6 is a display unit, 7 is a pointing device, 8 is a ROM for storing processing procedures (programs, etc.), 9 is a recognition dictionary, 10 is an external storage unit, 11
Is a keyboard and 12 is a color printer.

FIG. 2 is a flow chart for explaining the operation. The processing contents of the above configuration will be described below with reference to FIG. In step S210, the scanner 3 reads a document, which is an original, and stores it in the RAM 4 as image data. In step S220, the image data is displayed on the display unit 8. In step S230, an area is designated by enclosing a character portion in a frame with a pointing device (hereinafter referred to as PD). In step S240, a black pixel is projected in the horizontal and vertical directions to find a character cutout position, and the character is cut out for each character.

In step S250, the feature vector is obtained as follows. FIG. 3 shows one character image extracted in step S240. In step S250,
First, in order to make the size constant (normalize the size), FIG.
The character image 30 (L × L pixels) is converted into a character image 31 having a size of 62 pixels × 62 pixels. In the conversion, the pixel value of the coordinate (x, y) of the normalized image is calculated by the following formula (coordinate (x
_This is performed by setting the pixel value corresponding to the character image 30 of ( ₀ , y ₀ ).

[0011]

[Equation 1]

[0012]

[Equation 2]

However, N = 62. 6 made with this
An outer frame of white pixels (pixel value: 0) having a 1-dot width is added to the outside of the image of 2 pixels × 62 pixels, and an image of 64 pixels × 64 pixels is obtained as a final normalized character image. Then 0
A 63 × 63 pixel area of ≦ x ≦ N and 0 ≦ y ≦ N is set to 9 × 9.
Divide into small areas of pixel size. Therefore, there are a total of 7 × 7 = 49 small areas. The squares of the image 31 in FIG. 4 indicate this small area.

In the following, i-th (i = 0 to 0) in the x-direction
6), to represent the j-th (j = 0 to 6) small area in the y direction, it is designated by (i, j). It should be noted that the small area in the row of i = 6 and the small area in the column of j = 6 include a 1-dot width outer frame of white pixels. As a preliminary preparation, there are 16 possible states of the 2 × 2 pixel area. Of these, except for the case of all white pixels or all black pixels, the remaining 14 images are classified as shown in FIG.
The direction indices k (k = 0, 1, 2, 3) are associated with each other as illustrated. FIG. 6 shows the four directions represented by each direction index.

In the embodiment, the 2 × 2 pixel area in the small area (i, j) is set and the frequency H _ij (k) of each direction index is calculated as follows. FIG. 4 shows a small area (0, 0) of the normalized image 31 shown in FIG. In this small region, the 2 × 2 mask region 40 is scanned in the right direction by shifting it by one pixel in contact with the upper left corner. Then, scanning is sequentially performed with the position shifted downward by one pixel as the scanning start position. On the way, mask areas 41, 42,
As in 43, a portion that spans adjacent small areas also occurs.

During this scanning, it is checked which direction index in FIG. 5 the masked 2 × 2 image area is. Then, every time there is a corresponding direction index k, H _ij (k) (in this case H
₀₀ (k) will be counted up. At this time, the 2 × 2 image of all white pixels or black pixels is ignored. This is performed for 9 × 9 = 81 pieces in the small area. The above is performed for each small area to obtain the direction index histogram H
_ij (k) (i, j = 0,1, ..., 6; k = 0,1,2,
3) is obtained.

Note that a small area having an x-coordinate of 6 or a y-coordinate of 6 has a blank portion of 1 dot as its outer frame, and therefore 81 pieces of 2 × are provided for each of them. The orientation index of the two image areas is determined. Next, of the 49 small areas, i and j with even numbers are selected as the representative small areas. Since both i and j can take values of 0 to 6, 4 × 4 = 16 representative small areas can be specified in total. Here, in order to make the position of the representative small area easy to understand, (i ', j') (i ', j' = 0, 1, 2, 3)
Express with. As described below, the histograms of the representative small area and the small areas in the vicinity thereof are weighted and added, and new variables h _{i'j '} (k) (i', j '= 0, 1, 2, 3; k = 0 ，
1, 2, 3) is calculated.

[0018]

(Equation 3)

Here, the set G (i ', j') includes a representative small area and small areas in the vicinity thereof, and the small area in the vicinity is eight small areas diagonally above, below, to the left, to the right of the representative small area. is there. In the weighting factors g _ij , the focused small area is represented by the representative small area (i,
In the case of j), it is 4, the upper, lower, left and right small areas (i, j) are 2, and the oblique small areas (i, j) are 1, which is close to a two-dimensional Gaussian distribution function. However, when (i, j) becomes an undefined small area, g _ij = 0. The three-dimensional array h _{i'j '} (k) is appropriately rearranged in one dimension to obtain the feature vector x _i (i = 1, 2, ..., N). In the case of the embodiment, n is n = 4 × 4 × 4 × = 64 (there are 16 per k, and k can be a value of 0 to 3, and thus 64).

Now, consider further extension of this feature vector. Raster scanning is performed for each small area to find the number of black pixels, and the number of black pixels in the small area (i, j) is set to H.
_ij (4) (that is, the range that k can take is 0 to 4)
Then, by applying the above formula (A-3), similarly, h _{i'j '} (k
+1) is obtained. In this way, again the three-dimensional array h
_{i′j ′} (k) is appropriately rearranged in one dimension to obtain the feature vector x _i (i = 1, 2, ..., N ′). In this case,
n ′ = 4 × 4 × 5 = 80.

Here, the feature vector x _i (i =
1, 2, ..., N ′) is decomposed into a direction index part and a black pixel number part, and the first part vector of the direction index is x.
_i (i = 1, 2, ..., N), and the second partial vector of the number of black pixels is represented by _x′i (i = 1, 2, ..., N ″), where
It is obvious that n = 4 × 4 × 4 = 64 and n ″ = 4 × 4 × 1 = 14.

A method for creating a general recognition dictionary in the pseudo Bayes function method will be described below. The recognition dictionary is a correspondence table of character codes, character attributes, and standard patterns, and includes information about all character categories to be recognized. Here, the character attribute is a character type (alphabet, number, kanji, hiragana, katakana, symbol, code indicating other classification), font type, character size (upper or lower case, for example, o [o ], Etc., which distinguish between characters that have the same letter shape in two letter sizes), and other characteristics of the character. The character code may be a 1-byte ASCII code if the target character type is only alphanumeric characters or symbols, or a 2-byte JIS code if the target character type is Japanese. The Japanese character code is not limited to this because there are shift JIS and other code systems besides this.

The standard pattern is created for each character category as follows. Now, the character category is represented by ν [new] (ν = 1, 2, ..., L). The character category ν is observed n times (referred to as a process of reading out with a image scanner as an image of one character), and the feature vector is obtained by the above method. The feature vector obtained by the α-th observation is represented by _v x α (left side of the following equation). The mean vector _v x _ave of n _s observations is

[0024]

[Equation 4]

It is obtained by Any vector a

[0026]

(Equation 5)

The transposed vector a ^{t of a} can be expressed as a ^t =
(A ₁ a ₂ ... A _n ). here,

[0028]

(Equation 6)

By defining _{v v} is a so-called n × n covariance matrix. For each matrix _v V, eigenvalues and eigenvectors are obtained, and the eigenvalues are _v λ _i (i = 1, 2, ...,
n), the eigenvectors belonging to _v λ _i are _v ψ _i (i = 1, 2,
, N). However, the eigenvalues _v λ _i are arranged in the descending order of values in the order of i. When the feature vector x of the unknown input character β is obtained, the probability P (x | ν) that the unknown input character β is the character category ν is

[0030]

(Equation 7)

Is given by However, there is a rational process in which the unknown input vector follows an n-variable normal distribution. Now, if v _d (x) =-2logP (x | v),

[0032]

(Equation 8)

It becomes Here, (a, b) means a vector dot product. The smaller the value _v d (x) is, the higher the probability that the unknown input character belongs to the category character ν is. Therefore, _v d (x) is a dissimilarity function. This is called the pseudo Bayes discriminant function. By the way, the eigenvalue _v λ _i (i =
1, 2, ..., N), n _s learning characters are used to obtain the eigenvector _v ψ _i (the arrow in the equation 8 has an arrow above) belonging to _v λ _i (i = 1, 2, ..., N). Although necessary, since n _s is a finite number, an error is included in the eigenvalue _v λ _i and the eigenvector _v ψ _i . In particular, the higher-order terms of the eigenvalues are inaccurate because their absolute values are small. Therefore, all eigenvalues after i = k + 1 are replaced with a constant value _v Λ. Parameter _v
Λ is, for example, a method of making it equal to the eigenvalue of _v λ _k , or a method of making the average value of all eigenvalues of i = k + 1 or higher,
There is a method to set it to any other value. The order k (1 ≦ k ≦ 64) of the highest eigenvalue that is not replaced with a constant value is set to 10, for example. If this replacement is performed, the above dissimilarity function becomes as follows.

[0034]

[Equation 9]

Here, the standard pattern of the character category ν is the average vector _v x _ave and the eigenvalue _v λ _i (i = 1, 2,
, K), an eigenvector _v ψ _i (indicated by an arrow at the top in Equation 9) (i = 1, 2, ..., K), and a parameter _v Λ.
It is defined as a set of data. Now, for the character category ν (ν = 1, 2, ..., L), a standard pattern is previously obtained from a sufficiently large number of learning characters (for example, n _s = 500) by the above method, and the standard pattern of the character category ν A table in which a set of character codes and character attributes (character code, character attribute, standard pattern) is arranged in the order of character code is created as a recognition dictionary. In the recognition dictionary, L
Contains standard patterns for this character category.

In this embodiment, the recognition dictionary 9 stores two recognition dictionaries, a first recognition dictionary and a second recognition dictionary. In the present case, these are stored in one memory, but they may be separately stored in different memories. The first recognition dictionary is for determining character categories,
It is a table in which a set of character codes and character attributes of the character category ν for each category (character code, character attribute, standard pattern) is arranged in the order of character code. Include a sufficient number of sample characters of the recognition target font. In addition, in order to reduce the amount of calculation, the standard pattern uses the first partial vector (64 dimensions) as a feature vector.

In this way, the degree of difference between the feature vector (first partial vector) of the unknown character obtained in step S250 and the standard pattern of each character category in the first recognition dictionary is calculated by the formula (B-6). Yes (step S26
0). When the dissimilarity of the number L of character categories is obtained, the character categories are sorted in ascending order of dissimilarity. The character category that gives the smallest difference is the recognition result.

Next, in step S270, the font is recognized. Multiple classes for font recognition (class = font)
Discriminant analysis method is used. Assuming that the character category (that is, the character code) of the unknown character is fixed, it is necessary to recognize the font next. Here, the number of fonts to be recognized is F. In order to discriminate which font (hereinafter referred to as class) is an unknown font character (character whose character category is fixed but whose font is unknown), the method of the F class discriminant analysis is used here.
The explanation will be given below. In the following, the feature vector x represents an 80-dimensional feature vector obtained by combining the first partial vector and the second partial vector.

Letting A (m × n) be the matrix for changing from the feature vector x of an unknown font character to the new feature vector y (m dimension: m ≦ n) effective for discrimination,

[0040]

[Equation 10]

Average vector x _iave of feature vector x of each class c _i (i = 1 ... F), covariance matrix Σ of class c _i
i is given by the following equation.

[0042]

[Equation 11]

[0043]

(Equation 12)

Ec _i [...] represents the arithmetic mean in class c _i . Mean vector x _{i ave} , covariance matrix Σ of class c _i
i can be obtained from a sufficient number of classes c _i (for example, 50). The pre-occurrence probability of each class (probability that represents how often a font occurs) ω _i
The in-class covariance matrix Σ _W can be defined as

[0045]

(Equation 13)

Here, the pre-occurrence probability ω _i of each class can be obtained in advance by statistically investigating the frequency with which each class (font) is used. Then, the interclass covariance matrix Σ _B is defined as follows.

[0047]

[Equation 14]

Here, x _Tave is the average vector of the feature vectors over the entire class C. Also, (C-
4), (C-3), and (C-5), x can be replaced by y to define the intra-class covariance matrix Θ _W and the inter-class covariance matrix Θ _B for the new feature vector y. Then, the following relationship can be easily understood.

[0049]

(Equation 15)

Therefore,

[0051]

[Equation 16]

In other words, discriminant analysis in many classes shows that if the transformation matrix A that maximizes J (A) is obtained, accurate discrimination can be performed using the new feature vector y. With (C-6) and (C-7), this may solve the following eigenvalue problem.

[0053]

[Equation 17]

Where Λ is an m × m matrix in which only diagonal elements have non-zero eigenvalues (λ ₁ ≧ λ ₂ ... ≧ λ _m ). If the normalized eigenvector belonging to λ _i is φ _i , then A = (φ ₁ φ ₂ ... φ _m ). The eigenvector normalization condition is

[0055]

(Equation 18)

It is The conversion matrix A of x → y is given by the formula (C−
4) and (C-5), the eigenvalue problem (C-8) is solved from the learning data of each class. Next, as a second preparation, from the learning error of each class, the average vector y _iave for the new feature vector y _i and the covariance matrix (B → x →
y)) eigenvalues / eigenvectors and parameters
Find _i Λ.

In this way, if the mean vector y _iave about the transformation matrix A and the new feature vector y _i , the eigenvalues / eigenvectors of the covariance matrix and the parameter _i Λ are obtained in advance, the font recognition after the character category is determined is determined. Similarly to the above, the font can be determined by the input unknown font character and the pseudo Bayesian discriminant. Here, if you write a pseudo Bayes identification formula for font determination,

[0058]

[Formula 19]

However, ν is an index designating a class. In the second recognition dictionary, the conversion matrix A for all character categories and the standard font patterns (F) for each character code font are stored correspondingly. Here, the standard font pattern is the mean vector y _iave for the new feature vector y _i , the eigenvalue / eigenvector of the covariance matrix, and the parameter _i Λ. It is natural that these statistics are obtained by learning with the character samples of the corresponding character category / character font.

A new feature vector is obtained from the feature vector of the unknown character by the feature conversion matrix A, the degree of difference between each font and the standard font pattern is calculated by the formula (C-10), and the font giving the smallest degree of difference is recognized. The result. Note that the feature conversion matrix A is different for each character category. In step S280, the character code and the font code are output to the RAM 4, the color control code (using the ESC code of ASCII control characters, etc.) and the color code corresponding to the font code are output to the RAM 4,
It returns to step S250. When there are no more characters to be recognized in step S250, the process proceeds to step S290. In step S290, the character code of the RAM 4 and the color control code are input, and the character is printed in the color corresponding to the color code.
Printing is performed by the color printer 12.

[0061]

Second Embodiment In the above embodiment, the pseudo Bayes discrimination function is used to determine the character category, and the pseudo Bayes discrimination function is used to determine the character font. However, in the second identification (determination of character font), since the features have already been converted to effective new feature vectors by discriminant analysis,
It is not always necessary to use the pseudo Bayes discriminant function, and even if discriminated by a simpler Euclidean distance function, a simple similarity, a city block distance function, or the like, a large decrease in precision is not seen, and a slight speedup in processing can be expected.

[0062]

[Third Embodiment] By the way, the first recognition dictionary is created by learning all the fonts to be recognized, but the shape of the font may be very different from other fonts. For example, italics are the same in English. In such a case, it is difficult to learn all fonts using only the first recognition dictionary. Therefore,
Learn only special fonts (italics) separately and create a third recognition dictionary consisting only of those fonts. Then, the first discriminating means discriminates between the unknown input character and each of the first recognition dictionary and the third recognition dictionary, and obtains the degree of difference 1 with the first recognition dictionary and the degree of difference 2 with the first recognition dictionary. . Then, there is a method in which the character category corresponding to the smaller one of the difference 1 and the difference 2 is used as the recognition result and the process proceeds to the second identifying means to determine the character font. It is clear here that the second recognition dictionary does not need to have a special font separately.

As described above, according to this embodiment, the character font can be recognized at high speed and with high accuracy. Therefore,
An existing monochrome document can be colorized by associating a preset color with each font. As color copiers and color printers become widespread in the future, it will be required to create easy-to-understand color documents, but existing documents are monochrome.
Therefore, according to the present invention, since a color document can be easily created, there is a great effect in colorizing the representation of information.

When the present invention is applied to a copying machine, the output color for each font is set in advance on the operation panel or the like. Then, it is provided with means for generating a character pattern based on the character code. Then, based on the character code and font type information obtained in the above processing,
A corresponding character pattern is generated and printed in the color set on the operation panel or the like.

Further, not only the copying machine but also a host computer which outputs print data to a printer can be applied. That is, the print data is formed (the output color for each font is set in advance) based on the character code and the font type obtained in the above process, and the print data is output to the printer. Further, in the above embodiment, the image data from the device for optically reading the original image was recognized, but the present invention is not limited to this. For example, a facsimile receiver is provided, and the received image is recognized and output. It may be adapted to the device. In this case, the character size information for printing may be incorporated in a part of the print data and output based on the size of the character image to be recognized.

Regarding the processing time in the example of English recognition, the amount of calculation required to identify the standard pattern of the first recognition dictionary is
Since ρ × 100 = 100ρ and the amount of calculation for identification with the second recognition dictionary is 4ρ (when the number of fonts = 4), the total is 104ρ, and the processing time is greatly reduced. Regarding the accuracy of font recognition, the second identifying means extracts features suitable for font identification by the feature conversion matrix, and therefore the accuracy is naturally improved.

The present invention may be applied to a system composed of a plurality of devices or an apparatus composed of a single device, and is achieved by supplying a program to the system or the device. It can be easily conceived from the description of the above embodiment that it can be applied to the case.

[0068]

As described above, according to the present invention, a character font can be recognized at high speed and with high accuracy.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of a character recognition device in an embodiment.

FIG. 2 is a flowchart showing the contents of character recognition processing in the embodiment.

FIG. 3 is a diagram showing an unknown input character image and a normalized character image in the embodiment.

FIG. 4 is a diagram showing an outline of a small area of a normalized character image and feature extraction.

FIG. 5 is a diagram showing types of direction indexes in the example.

6 is a diagram showing the relationship between the direction of the direction index of FIG. 6 and its value.

[Explanation of symbols]

 1 CPU 2 Bus 3 Image Scanner 4 RAM 5 Feature Extraction Unit, 6 Display Unit 7 Pointing Device 8 ROM for Storage of Processing Procedure 9 Recognition Dictionary 10 External Storage Unit 11 Keyboard 12 Color Printer

Claims (8)

[Claims] 1. A character recognition device for recognizing a character by cutting out a character portion in an input original image, wherein the feature extracting means extracts a feature vector from the cut out character image and a character category. A first recognition dictionary storing a standard pattern including statistical information; a second recognition dictionary having a standard font pattern for determining a feature conversion matrix and a character font for each character category for each character font; The vector component of the feature vector is divided into a plurality of groups, and the feature decomposing means for decomposing into a plurality of partial vectors consisting of the components of each group, and the distance between one of the partial vectors and the standard pattern of the first recognition dictionary is calculated. , A first identifying means for outputting a character category of a standard pattern having the highest similarity, and a feature vector using the feature conversion matrix of the determined character category. Character recognition apparatus for converting a character into a new feature vector, and calculating a distance between the new feature vector and a standard font pattern of the second recognition dictionary to determine a character font. . 2. The statistical information of the standard pattern included in the first recognition dictionary is necessary for calculation of the pseudo Bayesian discriminant,
The character recognition device according to claim 1, comprising an average vector, an eigenvalue, an eigenvector, and a higher-order eigenvalue replacement parameter. 3. The standard font pattern of the second recognition dictionary includes an average vector of new feature vectors, eigenvalues, eigenvectors, and higher-order eigenvalue replacement parameters necessary for the calculation of the pseudo Bayesian discriminant. The character recognition device according to claim 1. 4. After the character category and the character font are determined, the character information output unit has a column for outputting the character code in correspondence with the font type. The character information is displayed in a different color for each font depending on the character information. The character recognition device according to claim 1, wherein the character recognition device prints. 5. A character recognition method for recognizing a character by cutting out a character portion in an input original image, the step of extracting a feature vector from the cut out character image, and a vector component of the feature vector. A feature decomposing step of decomposing into a plurality of groups and decomposing into a plurality of partial vectors composed of components of each group, one of the partial vectors, and a standard pattern containing statistical information for determining a character category are stored. The first identification step of calculating the distance from the standard pattern of the recognition dictionary of No. 1 and outputting the character category of the standard pattern having the highest similarity, and the feature vector with the feature conversion matrix of the determined character category as a new feature Converted to a vector, and the new feature vector,
Second identification for determining the character font by calculating the distance from the standard font pattern of the second recognition dictionary that has the characteristic conversion matrix for each character category and the standard font pattern for determining the character font for each character font A character recognition method, comprising: 6. The statistical information of standard patterns included in the first recognition dictionary is necessary for calculation of a pseudo Bayesian discriminant,
The character recognition method according to claim 5, comprising an average vector, an eigenvalue, an eigenvector, and a high-order eigenvalue replacement parameter. 7. The standard font pattern of the second recognition dictionary includes an average vector of new feature vectors, eigenvalues, eigenvectors, and higher-order eigenvalue replacement parameters required for calculation of the pseudo-Bayes' discriminant equation. The character recognition method according to claim 5. 8. After the character category and the character font are determined, a character information output unit for outputting the character code in correspondence with the font type is provided, and the character information is displayed in a different color for each font. 6. The character recognition method according to claim 5, wherein the character recognition method is printing.