JP2865611B2 - Character feature extraction method and device, and character recognition method and 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 method and an apparatus for extracting a characteristic of a character, and more particularly, to a method and an apparatus for extracting a characteristic of a character which are performed as preprocessing of character recognition for recognizing a read character.

[0002]

2. Description of the Related Art Conventionally, there has been known an apparatus which reads characters printed or handwritten on paper using a CCD (solid-state image sensor) image sensor or the like and recognizes the characters.
The image data of the read character is converted into two by the binarization circuit.
It is converted into value image data, and each character area is cut out from the binary image in units of one character. Furthermore, each cut-out character area is divided into a matrix area (for example, 8 × 8) mesh area (small area) and normalized (so-called size normalization). Then, the number of black pixels existing in each mesh area (64 in this case) of the normalized character area is obtained. Next, the number of black pixels in each mesh area is divided by the maximum number of black pixels in all the mesh areas, and the obtained values are multiplied by 255, respectively, to obtain a mesh density of 256 gradations (so-called, Concentration normalization). By obtaining this mesh density, the characteristics of the cut-out individual characters are extracted. Then, character recognition is performed by comparing and calculating the feature amount of the extracted character and the feature amount of each character pattern in a dictionary prepared in advance, or character recognition using a neural network for learning the character pattern is performed. The character recognition result is displayed on the screen of the display device.

[0003]

However, in the above-described character recognition processing, due to the relative displacement between the sheet and the CCD image sensor during reading, etc.
As shown in FIG. 5, the character region WE1 cut out by the first reading of the same character “1” and the character region WE2 cut out by the second reading show different distributions of black pixels. Therefore, after the size is normalized, the number of black pixels in each mesh area in the character area WE1 is different from the number of black pixels in each mesh area in the character area WE2.
As a result, there is a problem that the mesh density of the character area fluctuates greatly when the density is normalized. That is, even though the characters are the same, a change in the mesh density causes erroneous recognition in character recognition using a dictionary or a neural network.

[0004] In addition, the above-described problem occurs even with characters having exactly the same shape and size. Furthermore, when character size normalization and density normalization are actually performed in a character recognition device, the mesh density may be low or high depending on the size and shape of the original character. , The characteristics of the characters vary. Even in such a case, the above-described character recognition may cause erroneous recognition.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to provide a character feature extraction method and apparatus capable of reducing variations in character feature amounts. is there.

Another object of the present invention is to provide a character recognizing method and a character recognizing method capable of easily performing character recognition by reducing the variation in the characteristic amount of characters extracted for character recognition.

[0007]

According to the first aspect of the present invention, a density value of each small area forming a character data area is calculated, and the density value of each of the calculated small areas is calculated. The average of the density values is similar to a predetermined average density value while maintaining the dynamic range of the density value,
The gist is a character feature extraction method for converting the density value of each small area.

[0008] In the invention according to the second aspect, a step of calculating a density value of each of the small areas based on a ratio of constituent elements of the character data in each of the small areas forming the character data area is provided. In the feature extraction method, the calculating step may be performed such that an average of the density values of the respective small areas is close to a predetermined average density value while maintaining a dynamic range of the density values. Is included.

According to a third aspect of the present invention, there is provided a method for calculating the density value of each of the small areas based on the ratio of constituent elements of the character data in each of the small areas forming the character data area. In the feature extraction method, the calculating step may be performed such that an average of the density values of the respective small areas is close to a predetermined average density value while maintaining a dynamic range of the density values. The predetermined average density value is set to a value that can maintain the converted density value of each of the small areas in a balanced state of the density value of each of the small areas before conversion. The point is that

According to a fourth aspect of the present invention, in the method according to the second or third aspect, the constituent element is black pixel data constituting character data, and the black pixel data exists in the conversion step. The point is that the density value of each small area to be converted is converted.

According to the fifth aspect of the present invention, a first step of calculating a value indicating a ratio of constituent elements of character data in each small area forming a character data area, and a calculated value of each small area And a second step of calculating a density value.
After the first step and before the second step, the average of the density values in each of the small areas is approximated to a predetermined average density value while maintaining the dynamic range of the density values. A gist of the present invention is a character feature extraction method including a step of converting a value indicating a ratio of a component in a small region.

According to a sixth aspect of the present invention, in the method according to the fifth aspect, the constituent element is black pixel data constituting character data, and in the conversion step,
The gist is that the value indicating the ratio of the black pixel data is converted for the small area where the black pixel data exists.

According to the present invention, the density value of each small area forming the character data area is calculated, and the calculated density value of each small area is converted into a density value of a predetermined range width. In the character feature extraction method, the average value of the converted density values of the area occupied by each component of the character data is determined in advance while maintaining the vertical relationship of the density values of each area of the character data before conversion. A gist of the present invention is a character feature extraction method for converting the density value so as to approach a predetermined value.

According to the present invention, the character data area is divided into a plurality of small areas to normalize the size of each small area, and the structure of the character data in each normalized small area. A character extracting step of calculating a density value of each of the small areas based on a ratio of the element to extract a characteristic of the character; and a step of identifying the character based on the characteristic of the extracted character. In the feature extraction method, the extracting step may be performed so that an average of the density values of the respective small areas is close to a predetermined average density value while maintaining a dynamic range of the density values. The gist is a character recognition method including a step of converting a value.

According to the ninth aspect of the present invention, the step of dividing the character data area into a plurality of small areas to normalize the size of each small area, and the structure of the character data in each normalized small area A character extracting step of calculating a density value of each of the small areas based on a ratio of the element to extract a characteristic of the character; and a step of identifying the character based on the characteristic of the extracted character. In the recognition method, in the extracting step, the average of the density values of each of the small areas holds a dynamic range of the density value,
Converting the density value of each of the small areas so as to approximate a predetermined average density value, wherein the predetermined average density value is the density value of each of the converted small areas, The gist is a character recognition method that is set to a value that can maintain the balance of the density value of a small area.

According to a tenth aspect of the present invention, in the method according to the eighth or ninth aspect, the constituent element is black pixel data constituting character data, and the black pixel data exists in the conversion step. The point is that the density value of each small area to be converted is converted.

The invention according to claim 11, wherein the density value of each of the small areas is calculated based on the ratio of the constituent elements of the character data in each of the small areas forming the character data area; Means for converting the density value of each of the small areas so that the average of the density values of each of the small areas calculated by the calculation means is close to a predetermined average density value while maintaining the dynamic range of the density value. The gist is a character feature extraction device including

[0018] According to the invention described in claim 12, in the invention described in claim 11,
In the apparatus according to (1), the predetermined average density value is a value capable of keeping the density value of each of the small areas converted by the conversion unit in a balanced state of the density value of each of the small areas before conversion. It is assumed that it is set to.

According to the thirteenth aspect of the present invention, the character data area is divided into a plurality of small areas to normalize the size of each small area, and the size of each small area normalized by the normalizing means is reduced. Means for calculating the density value of each of the small areas, based on the ratio of the components of the character data in
The density value of each of the small areas is converted such that the average of the density values of each of the small areas calculated by the calculation means is close to a predetermined average density value while maintaining the dynamic range of the density value. A gist of the present invention is a character recognition device comprising: means for identifying the character based on the density value of each of the small areas converted by the conversion means.

According to the fourteenth aspect of the present invention, in the thirteenth aspect,
In the apparatus according to (1), the predetermined average density value is a value capable of keeping the density value of each of the small areas converted by the conversion unit in a balanced state of the density value of each of the small areas before conversion. It is assumed that it is set to.

According to the invention of claim 15, claim 13 is provided.
Or a facsimile device provided with the character recognition device described in 14 above. (Operation) According to the invention described in claims 1, 2, and 11,
The density value of each component is converted such that the average of the density values of each small area approximates a predetermined average density value while maintaining the dynamic range of the density value. Therefore, the average of the converted density values is maintained substantially constant, and the variation in the character feature amount is reduced.

According to the third and twelfth aspects of the invention,
The conversion of the density value of each small area can be performed while maintaining the balance state of the density values of the respective components before conversion.
According to the fourth and tenth aspects of the present invention, only the density value of each small area where the black pixel data exists is efficiently converted.

According to the fifth aspect of the present invention, before the density value of each small area is calculated, the value indicating the ratio of each constituent element is converted. According to the sixth aspect of the present invention, for each small area where black pixel data exists, only the value indicating the ratio of the black pixel data is efficiently converted.

According to the seventh aspect of the present invention, the vertical relationship between the density values before conversion of each area of the character data is maintained,
The conversion of the density values is performed so that the average value of the converted density values of the area occupied by the respective constituent elements of the character data approaches a predetermined value.

According to the invention described in claims 8 and 13,
Character density can be easily recognized in a state where the density value of each component is converted and the variation in the characteristic amount of the character is small.

According to the ninth and fourteenth aspects of the invention,
It is possible to easily and accurately perform character recognition in a state in which the density value of each component is converted, the variation in the characteristic amount of the character is small, and the balance state of the density value of each component before conversion is maintained. Become.

According to a fifteenth aspect of the present invention, there is provided a facsimile apparatus including the character recognition device according to the twelfth or thirteenth aspect.

[0028]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to FIGS. In the present embodiment, a character recognition device that performs character feature extraction is applied to a facsimile device. For a more detailed description of this type of facsimile apparatus, refer to Japanese Patent Application No. 7-29.
No. 9609, it will be briefly described here. FIG. 1 is a block diagram showing an electrical configuration of the facsimile machine 1. As shown in FIG. 1, the facsimile apparatus 1 includes a paper insertion roller 2 in a paper insertion section 1a, and a paper detector 3 and a CCD (Charge Coupled) at its insertion opening.
Device) An image sensor 4 is provided. CCD
The image sensor 4 is connected to a control unit 7 via a binarization circuit 5 and an image memory 6. The control unit 7 includes a paper detector 3, a drive control of a motor 8 for driving the paper insertion roller 2. Drive circuit 9, input unit 10, display unit 1 for performing
1, a display control circuit 12 for performing display control, a tone dial unit 13, and a modem 14 are connected. The tone dial unit 13 and the modem 14 are connected to an NC connected to a telephone line.
U (network control device) 15. The control unit 7
It includes a character cutout processing section 16, a normalization processing section 17, a character recognition section 18, a dictionary 19, a telephone number management processing section 20, and a FAX encoding processing section 21.

The paper detector 3 comprises, for example, an optical sensor and detects that the paper P has been inserted into the paper insertion section 1a. A detection signal from the paper detector 3 is output to the control section 7. You. The CCD image sensor 4 is composed of a number of solid-state imaging devices, and generates an analog image signal AS corresponding to an optical image of a character written on the paper P. The generated image signal AS is output to the binarization circuit 5. The binarizing circuit 5 converts the analog image signal AS sequentially output from the CCD image sensor 4 into a black and white binary digital image signal (hereinafter, referred to as a binary image signal) ID. The converted binary image signal ID is output to the image memory 6.

The image memory 6 is composed of a RAM (random access memory), and stores a binary image signal ID input from the binarization circuit 5 as image data in a memory area corresponding to an address signal from the control unit 7. It is stored in a readable manner. The control unit 7 performs C based on the detection signal from the paper detector 3.
A read timing signal for controlling the read speed is output to the CD image sensor 4 and a drive control signal is output to the motor drive circuit 9 so that the motor 8 is driven in synchronization with the read timing signal.

The input unit 10 includes a number input key for inputting a telephone number, a facsimile transmission key, and the like, and outputs a key input data signal to the control unit 7. As shown in FIG. 2, the control unit 7 stores the image memory 6 in which the read image data of the telephone number TN written in the predetermined area A at the head of the paper P is stored.
Then, the image data is read out from, and character recognition processing is performed by the character cutout processing unit 16, the normalization processing unit 17, and the character recognition unit 18.

The character cutout section 16 obtains vertical component histograms for various inclination angles by assuming the inclination angle of the character string with respect to the pixels of the binary image data read from the image memory 6. Then, the character cutout unit 16 determines a character cutout direction based on the distribution of each histogram value. That is, the inclination angle at which the area where the histogram value is one pixel or more is the largest is defined as the character string direction. Further, the character cutout unit 16 sets an area where a histogram value of one pixel or more continuously appears as one character, and cuts out each character along the determined cutout direction.

The normalization processing section 17 includes a character cutout section 16
A circumscribed rectangle of each character is set based on the connection information of the pixel data forming each character cut out by the above, and a character area is set. The normalization processing unit 17 further sets each character area to 8
It is divided into mesh areas (small areas) of a matrix of × 8 and normalized (so-called size normalization). The normalization processing unit 17 obtains the number of black pixels as the density before normalization existing in each mesh area (64 in this case) of the character area whose size has been normalized. Further, the number of black pixels in each mesh area is divided by the maximum number of black pixels in all the mesh areas, and the value obtained by the division is multiplied by 255 to obtain a mesh density of 256 gradations (so-called density). Normalization). This mesh density represents the characteristics of each character.

Each time the mesh density value of each character is obtained, the normalization processing unit 17 calculates an average value of the mesh density,
Using the obtained average density value K and a preset correction density value C, the value X0 of each mesh density is converted. That is, when the mesh density value X0 satisfies the following condition (i), it is converted to a value X1 according to equation (1), and when it satisfies condition (ii), it is converted to a value X2 according to equation (2).

(I) When X0 ≧ K, X1 = (X0−K) × (255−C) / (255−K) + C (1) (ii) When X0 <K, X2 = C × (1− (K−X0) / K) (2) The average density value K is obtained by calculating the sum of the mesh density values X0 by the number of mesh areas having a density value exceeding 0 (black spots exist). This is the value obtained by the division. Correction density value C
Corrects the current average density value K and calculates the average density value K
Value used to minimize the variation of
This value is determined experimentally. In the present embodiment,
The correction density value C is set to 180. That is, the conversion according to the equations (1) and (2) is performed so that the current average density value K of the mesh density of each character is corrected to a value close to the correction density value C.

The character recognizing unit 18 stores the dictionary 1 prepared in advance.
9 is compared with the feature amount of the basic character pattern stored in the dictionary 19 and the feature amount of the input converted character pattern to determine similarity. Then, the character having the most similar feature is set as the recognition result. The dictionary 19 is a recognition dictionary in which graphic features of all characters to be recognized are stored. For all numbers and symbols provided on the number input keys of the input unit 10, the typefaces (diagonal characters, habit characters) are used. , Etc.) are stored for each pattern. Characters recognized by the character recognition unit 18 (in this embodiment, telephone numbers or FA
The numbers and symbols used for the X number are output to the telephone number management processing unit 20.

The display unit 11 is, for example, a liquid crystal display (LC)
D), the number input data input from the input unit 10 to the control unit 7 and the character recognition result of the character data read by the control unit 7 are displayed on the screen. This display unit 11
The display is controlled by the control unit 7 via the display control circuit 12. The input unit 10 and the display control unit 12 are connected to a telephone number management processing unit 20.

The telephone number management processing unit 20 stores a telephone number data output from the input unit 10 and a character recognition result output from the character recognition unit 18.
And a processing unit for performing processing on the storage unit based on input data from the storage unit. In the number confirmation mode by character recognition, when the number data is input from the input unit 10 by operating the numeric keys, the telephone number management processing unit 20
Displays the number data on the screen of the display unit 11 via the display control circuit 12. Further, the telephone number data as the character recognition result stored in its own storage unit is replaced with the input number data. Further, the management processing unit 20
Based on a transmission key operation signal from the input unit 10, telephone number data corresponding to the telephone number currently displayed on the display unit 11 is read out from the storage unit and transferred to the tone dial unit 13.

The tone dial unit 13 generates a tone signal corresponding to a push line in accordance with the telephone number data transferred from the telephone number management processing unit 20. The generated tone signal is transmitted to the NCU 15. The NCU 15 selectively switches the connection between the tone dial unit 13 and the modem 14 to the telephone line based on the control signal from the control unit 7, and confirms that the telephone line has been connected to the other party's facsimile machine. The FAX encoding processing unit 21 performs an encoding process on image data sequentially read from the image memory 6 based on an operation signal of a FAX transmission key from the input unit 10 to generate a FAX transmission signal. Generated FA
The X transmission signal is transmitted to the modem 14, NCU 1 after connection to the telephone line.
5 is transmitted from the telephone line to the line connection destination.

Next, the operation of the facsimile apparatus 1 configured as described above up to character recognition will be described with reference to the flowchart of FIG. In step S <b> 1, when the user inserts the transmission sheet P in which the telephone number is handwritten in the predetermined area into the insertion slot of the sheet insertion unit 1 a, the sheet P is detected by the sheet detector 3. Then, the controller 7 operates the CCD image sensor 4 and drives the motor 8 via the motor drive circuit 9 to rotate the roller 2. As a result, the paper P is fed at a predetermined speed in only a predetermined area at the head, and the telephone number is read by the CCD image sensor 4.

In step S2, the analog image signal AS read from the CCD image sensor 4 is converted into a binary image signal ID by the binarization circuit 5, and stored in the image memory 6, and further stored in the image memory 6. The image data is sequentially transferred to the character cutout processing unit 16.
The character cutout processing unit 16 obtains vertical component histograms for various inclination angles for the pixels of the binary image data input from the image memory 6, and determines the character cutout direction based on the distribution of each histogram value. . Thus, each character is cut out along the determined cutout direction.

In step S3, the normalization processing section 17
Sets the character area by taking the circumscribed rectangle of each cut-out character. Further, each character area is divided into 8 × 8 matrix mesh areas, and the size is normalized.

In step S4, the number of black pixels existing in each mesh area of the character area whose size has been normalized is determined, and the mesh density of 256 gradations for each mesh area is calculated. Is normalized. That is, as described in the background art, the mesh density is obtained by dividing the number of black pixels (density before normalization) for each mesh area by the maximum number of black pixels in all mesh areas, and obtaining a value obtained by the division. 255 respectively. FIG. 4 shows each mesh area 30 in the character areas WE1 and WE2 after density normalization has been performed.
Are shown respectively. This character area WE1, W
E2 is obtained by the first and second readings of the same character "1", and corresponds to each of the character areas WE1 and WE2 in FIG. 5 used in the description of the conventional technique.

In step S5, the normalization processing section 17
Is obtained by calculating the average value of the mesh density every time the mesh density value of each character is obtained, and using the obtained average density value K and the preset correction density value C, the above equation (1),
According to (2), the value X0 of each mesh density is converted into X1 and X2. For example, as shown in FIG. 4, for the character area WE1, the average density value K of the mesh density
When it is calculated as 150, the average density value K and the preset correction density value C = 180 are used to calculate the equations (1) and (2).
Is converted into X1 and X2. By this conversion, the average density value K of the mesh density of the character “1” becomes K = 162, and the density is corrected to a value close to the corrected density value C = 180.

Further, when the average density K of the mesh density of the character area WE2 is calculated to be K = 167, the mesh density values X0 are similarly calculated according to the equations (1) and (2).
Is converted into X1 and X2. By this conversion, the average density K = 174 of the mesh density of the character “1” is obtained.
The correction density value is corrected to a value close to 180. Thus, after the conversion, the variation in the average value of the mesh density (K = 150, K = 167) is reduced (K =
162, K = 174). In order to make the explanation easy to understand, the normalization of the density in step S4 and
Although the density value conversion in step S5 is performed in a separate step, these may be naturally combined into one step.

Here, the corrected density value C = 180 is a value obtained as a result of an experiment. The reason that this value is appropriate is that the variation of the average density value K can be minimized. It is presumed that the balance of the mesh density values can be maintained. For example, when the correction density value C is set to a high value (a value close to 255), there is a merit that the mesh density value on the relatively low side is shifted upward by the conversion to emphasize the feature, but conversely, the high density side There is a disadvantage that the difference between the mesh density values is lost (that is, the change in the value is not preserved), and the balance of the mesh density values is lost. Therefore, it is estimated that the correction density value capable of maintaining the balance of the mesh density value while preserving the change of the mesh density value on the high side while shifting the mesh density value on the relatively low side is about 180. Is done. Therefore, if the dispersion of the average density value K can be minimized and the balance of the mesh density values can be maintained, the corrected density value C can be changed to a value other than 180, such as 170 or 160.

In step 6, the character recognition unit 18 performs character recognition by comparing and calculating the characteristic amount of the input character pattern and the characteristic amount of the basic character pattern stored in the dictionary 19. In this way, each character described in the predetermined area of the paper P is recognized, and the character recognition result is stored as telephone number data in the telephone number management processing unit 20 and is displayed via the display control circuit 12 on the display unit 11.
Is displayed on the screen G.

As described in detail above, the facsimile apparatus 1 of the present embodiment has the following effects. (I) Each mesh density value is converted by using the above equations (1) and (2) so that the average density value K of the mesh density is close to the corrected density value C. By making the value K substantially uniform, the fluctuation of the mesh density can be reduced. As a result, the variation in the feature amount of each character is reduced, and character recognition using the dictionary 19 is facilitated, and occurrence of erroneous recognition can be reduced.

(Ii) By setting the correction density value C used in the above equations (1) and (2) to a value (180) capable of maintaining the balance of the mesh density value before conversion,
Accurate character recognition can be performed while retaining the feature amount before conversion.

(Iii) The width of the change in the mesh density of the character (the width of the dynamic range) is from 0 to 255, which is the same as the conventional one, and the vertical relationship of the mesh density of the character is maintained. That is, the feature amount of each character at the time of recognition does not decrease so much. In addition, it is possible to suppress variations in the feature amounts when the same character type is recognized. Therefore, accurate character recognition can be performed.

The present invention is not limited to the above embodiment, but may be implemented as follows. (A) The mesh density value may be converted using a conversion formula other than the above formulas (1) and (2).

(B) In the above embodiment, after the density is normalized (concentration is set within a predetermined range of 0 to 255), the above equations (1) and (2) are converted. But,
This may be reversed. That is, first, density correction is performed. In this case, the corrected density value C ′ is: C ′ = ((highest value of density before normalization (the number of black pixels)) − (lowest value of density before normalization)) × (corrected density value C) / 256 Since the minimum value of the density before normalization is 0, C ′ = the maximum value of the density before normalization × the corrected density value C / 256.

Then, equation (1) becomes Z1 = (concentration before normalization Z0−K) × (highest value of density before normalization−C ′) / (highest value of density before normalization−K) + C ′. , Equation (2)
Z2 = C'.times. (1- (K-Z0) / K). In addition,
As the density Z0 before normalization, only the mesh area where the black point exists is subject to conversion. Thereafter, density normalization is performed using the converted pre-normalization densities Z1 and Z2, and a mesh density value is calculated.

(C) In the above embodiment, character recognition is performed using the dictionary 19. Alternatively, character recognition may be performed using a neural computer or a Neumann computer as a neural network type character recognition means. Regarding neural network type character recognition, JP-A-4-177485, JP-A-4-260987,
This is a well-known technique described in JP-A-6-44409.

(D) In the above embodiment, the characters to be recognized are numbers and symbols used in the telephone number. However, other characters to be recognized may include, for example, hiragana, katakana, English characters, and kanji.

(E) The present invention is not limited to a facsimile apparatus, but can be applied to other apparatuses.

[0057]

As described above in detail, according to the present invention, the average of the density values of the respective small areas forming the character data is approximated to a predetermined average density value while maintaining the dynamic range of the density values. As described above, by performing the conversion of the density value of each small area, it is possible to reduce the variation in the characteristic amount of the character.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a facsimile apparatus.

FIG. 2 is an explanatory diagram showing telephone numbers written on a sheet.

FIG. 3 is a flowchart showing the operation of the facsimile machine.

FIG. 4 is an explanatory diagram showing a conversion process of a mesh density value.

FIG. 5 is an explanatory diagram showing normalization and feature extraction processing of a conventional example.

[Explanation of symbols]

17 ... Normalization processing unit as calculation means, conversion means and normalization means, 18 ... Character recognition unit as character recognition means, K
... average density value, C ... correction density value (predetermined average density value),
X0 ... density value (before conversion), X1, X2 ... density value (after conversion).

──────────────────────────────────────────────────続 き Continuing on the front page (72) Hiroshi Horii 2-5-5 Keihanhondori, Moriguchi-shi, Osaka Sanyo Electric Co., Ltd. (72) Inventor Junji Tanaka 3-201 Minamiyoshikata, Tottori-city, Tottori (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/62 630

Claims (15)

(57) [Claims] 1. A density value (X0) of each small area forming a character data area is calculated, and an average (K) of the calculated density values (X0) of each small area is calculated as the density value (X0). A) converting the density value (X0) of each of the small areas so as to approximate a predetermined average density value (C) while maintaining the dynamic range of (1). 2. A character feature comprising a step (S4) of calculating a density value (X0) of each small area based on a ratio of constituent elements of character data in each small area forming a character data area. In the extraction method, the calculating step (S4) may include determining a density value (X0) of each of the small areas.
Converting the density value (X0) of each of the small areas such that the average (K) of the small areas approximates a predetermined average density value (C) while maintaining the dynamic range of the density value (X0) ( S5) A character feature extraction method characterized by including: 3. A character feature comprising a step (S4) of calculating a density value (X0) of each small area based on a ratio of constituent elements of character data in each small area forming the character data area. In the extraction method, the calculating step (S4) may include determining a density value (X0) of each of the small areas.
Converting the density value (X0) of each of the small areas such that the average (K) of the small areas approximates a predetermined average density value (C) while maintaining the dynamic range of the density value (X0) ( S5), the predetermined average density value (C) converts the converted density value (X1, X2) of each small area into a balance state of the density value (X0) of each small area before conversion. A character feature extraction method characterized by being set to a value that can be maintained. 4. The method according to claim 2, wherein
The component is black pixel data constituting character data, and in the conversion step, the density value (X0) of each small area where the black pixel data exists is converted. Method. 5. A value (Z0) indicating a ratio of constituent elements of character data in each small area forming a character data area.
, And a second step of normalizing the calculated value (Z0) of each small area to calculate a density value. Before the second step, the average (K) of the density values in each of the small areas is close to a predetermined average density value (C ′) while maintaining the dynamic range of the density value (X0). And a step of converting a value (Z0) indicating the ratio of the constituent elements in each of the small areas. 6. The method according to claim 5, wherein the component is black pixel data constituting character data;
In the conversion step, a value (Z0) indicating a ratio of the black pixel data is converted for a small area in which the black pixel data exists, wherein the character is extracted. 7. A character feature extraction method for calculating a density value of each small area forming an area of character data and converting the calculated density value of each small area into a density value of a predetermined range width. The average value of the converted density values of the area occupied by the constituent elements of the character data approaches a predetermined value (C) while maintaining the vertical relationship of the density values of each area of the character data before conversion. In this manner, the character feature extraction method is characterized in that the conversion of the density value is performed. 8. A step of dividing the character data area into a plurality of small areas and normalizing the size of each small area, and based on the ratio of the constituent elements of the character data in each normalized small area. Extracting a character feature by calculating a density value (X0) of each of the small areas; and identifying the character based on the extracted character feature. The extracting step is such that the average (K) of the density values (X0) of each of the small areas is close to a predetermined average density value (C) while maintaining the dynamic range of the density value (X0). A character recognition method comprising a step (S5) of converting a density value (X0) of each of the small areas. 9. A step of dividing a character data area into a plurality of small areas to normalize the size of each small area, and based on a ratio of a component of character data in each normalized small area. Extracting a character feature by calculating a density value (X0) of each of the small areas; and identifying the character based on the extracted character feature. The extracting step is such that the average (K) of the density values (X0) of each of the small areas is close to a predetermined average density value (C) while maintaining the dynamic range of the density value (X0). A step (S5) of converting the density value (X0) of each small area, wherein the predetermined average density value (C) is obtained by converting the converted density value (X1, X2) of each small area before conversion. Of the density value (X0) of each of the small areas Character recognition method characterized in that it is set to the possible values keeping. 10. The method according to claim 8, wherein the component is black pixel data constituting character data, and the density value of each small area where the black pixel data exists in the conversion step. (X0) is converted. 11. A means (1) for calculating a density value (X0) of each of the small areas based on a ratio of constituent elements of the character data in each of the small areas forming the character data area.
7), and the average (K) of the density values (X0) of the small areas calculated by the calculation means (17) is a predetermined average density value while maintaining the dynamic range of the density value (X0). (C)
Means (17) for converting the density value (X0) of each of the small areas so as to approximate 12. The apparatus according to claim 11, wherein the predetermined average density value (C) is a density value (X1, X2) of each of the small areas converted by the conversion means (17).
A character feature extraction apparatus characterized in that the density value (X0) of each of the small areas before the conversion is set to a value capable of maintaining a balance state. 13. A means (17) for dividing a character data area into a plurality of small areas to normalize the size of each small area, and a means for normalizing each small area by said normalizing means (17). Means (1) for calculating the density value (X0) of each of the small areas based on the ratio of the constituent elements of the character data
7), and the average (K) of the density values (X0) of the small areas calculated by the calculation means (17) is a predetermined average density value while maintaining the dynamic range of the density value (X0). (C)
Means (17) for converting the density value (X0) of each of the small areas so as to approximate to the following: based on the density values (X1, X2) of each of the small areas converted by the conversion means (17). And a means (18) for identifying the character. 14. The apparatus according to claim 13, wherein the predetermined average density value (C) is a density value (X1, X2) of each of the small areas converted by the conversion means (17).
A character recognition device characterized in that the character recognition device is set to a value that can maintain the balance of the density value (X0) of each of the small areas before conversion. 15. A facsimile device comprising the character recognition device according to claim 13.