JP2000181994A - Character recognition processing method, device therefor and recording medium recording the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a processing method and the device for correctly recognizing a character even in a condition where background noise is present. SOLUTION: A character pattern is divided into local areas in a preprocessing part 102 and the number of black pixels present inside the local area is measured for each local area in a black pixel number extraction part 1-4. In an identification part 1-5, a difference value from the standard black pixel number of the respective categories of a prepared black pixel number standard dictionary 1-7 is obtained, and at the time of calculating a distance value or similarity between a feature vector obtained from the local area of an input pattern in a feature extraction part 1-3 and the standard feature vector of the corresponding category of the feature standard dictionary 1-7, adjustment is performed so as to reduce the distance value or the value of the similarity among the respective local areas by the value proportional to the black pixel number difference value. Thus, the fluctuation of the distance value or the similarity in the local area including a lot of noise is reduced, the fluctuation of the distance value or the similarity in the entire character pattern is reduced and correct character recognition is made possible.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a method and an apparatus for recognizing a character pattern, and more particularly to a method and an apparatus for recognizing a binary character pattern.

[0002]

2. Description of the Related Art As an example of a conventional character pattern recognition method and apparatus, a feature necessary for recognition is extracted in the form of a vector from a character pattern that has been subjected to binarization, position and size normalization ( For example, a character pattern is divided into coarse local areas, a character part existing in each local area is observed from coordinate axes in a plurality of directions, and a black line of a character part that intersects when scanned from the coordinate axis is a character line. No. Hagida, Seiichiro Naito, Isao Masuda, and a method and apparatus for recognizing characters by obtaining the directional contribution (Japanese Patent Application No. 56-46659).
"Handwritten Kanji Recognition Method Using Global / Local Direction Contribution Density Features", IEICE (D), J66-D, No. 6, p
p. 722-729 "), a method of determining a discriminant function such as a distance value or a similarity between the standard pattern vector of each category in a standard dictionary created in advance and determining the most similar character category as a recognition result. (Hereinafter, “first method”).

Further, the binary pattern of the input pattern itself is used as an element of the feature vector, and can be exchanged with a standard pattern vector of each category in the standard dictionary created in advance (expressed as a binary pattern like a feature vector). There is a method of determining the most similar character category as a recognition result by a simple similarity or the like, which is a discriminant function based on the number of black pixels (hereinafter, “second method”).

[0004]

According to the first method of character pattern recognition described above, elements of the conventional feature vector have a wide range of feature values reflecting the character line structure such as the direction of the character line, the connection relationship, and the positional relationship. However, for images with strong background noise, these features vary greatly, making it difficult to recognize correctly due to large variations in distance values or similarities. is there.

Further, the second method of character pattern recognition has a problem that an image having a large background noise or the like is erroneously recognized as a category having a large number of black pixels in a standard pattern vector. In addition, since the determination is made based on the number of overlapping pixels, there is a problem that the recognition performance is greatly reduced even if the position of superimposition of the input pattern and the standard pattern is slightly shifted.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned drawbacks, and has as its object to extract, in the form of a vector, features necessary for recognition from a character pattern whose binarization and position and size have been normalized. In a method in which a discriminant function such as a distance value or a degree of similarity is obtained between a standard pattern vector of each category in a standard dictionary created in advance and a similar character category is determined as a recognition result, a character pattern When noise is present in the background, in a local region where noise is present, the feature value extracted from the noise is superimposed on the feature value extracted from the character line, so the value of the feature vector obtained from the local region varies greatly. As a result, the distance value or similarity value obtained by the conventional discriminant function fluctuates greatly, making it difficult to correctly recognize the character type. Tsu and thus solve the problem, even in situations where background noise as described above exists, it is to provide a correctly recognizing method and apparatus letters.

[0007]

According to the character recognition processing method of the present invention for solving the above-mentioned problems, a character pattern cut out for each character and binarized is normalized with respect to position and size. Performing a process of extracting a feature from the normalized character pattern, and performing a process of dividing the normalized character pattern into local regions separately from the process of extracting the feature. The number of black pixels present in the local area is measured for each of the local areas of the obtained character pattern, and the number of black pixels obtained by the measurement is compared with the standard of each category of the previously created black pixel number standard dictionary. A process of obtaining a difference value of the number of black pixels from the number of black pixels is performed, and a feature vector obtained from a character pattern in the local area and a standard feature vector of a corresponding category of a feature standard dictionary created in advance. Is adjusted according to the magnitude of the black pixel number difference value, and by using the adjusted distance value or similarity, the most similar character category is determined. The recognition result of the character pattern is used.

Alternatively, a distance value or a similarity value obtained between a feature vector obtained from a character pattern in the local area and a standard feature vector of a corresponding category in a previously created feature standard dictionary is calculated. In the process of adjusting according to the size of the black pixel number difference value, if the black pixel number difference value is 0, the distance values or similarities calculated for all the dimensions of the feature vector are accumulated, and If the black pixel number difference value is greater than 0, the distance value or similarity of each dimension of the feature vector is divided by the black pixel number difference value and accumulated, and the accumulated value is adjusted for the local region. It is characterized by a distance value or similarity.

Alternatively, a distance value or a similarity value obtained between a feature vector obtained from a character pattern in the local region and a standard feature vector of a corresponding category in a previously created feature standard dictionary is calculated. At the stage of the process of adjusting according to the magnitude of the black pixel number difference value, the distance value or similarity of each dimension of the feature vector is subtracted by the black pixel number difference value, and the subtracted distance value or similarity is calculated. The method is characterized in that the subtracted distance value or similarity in the case of 0 or more is accumulated, and the accumulated value is used as the adjusted distance value or similarity of the local region.

Further, a program for causing a computer to execute the processing steps in the above character recognition processing method is recorded on a computer-readable recording medium.

[0011] Similarly, a character recognition processing apparatus according to the present invention for solving the above-mentioned problem is a pre-processing that performs a normalization process on a position and a size of a character pattern cut out for each character and binarized. Means, feature extracting means for extracting a feature from the normalized character pattern, and dividing the normalized character pattern into local regions, and for each local region of the divided character pattern, And a black pixel number extracting means for measuring the number of black pixels existing in the black pixel number obtained by the measurement and the standard black pixel number of each category of the previously created black pixel number standard dictionary. A black pixel number difference value calculating unit for obtaining a pixel number difference value, and a feature vector obtained from a character pattern in the local area and a standard feature vector of a category corresponding to a previously created feature standard dictionary. A distance value or similarity calculating means for calculating a distance value or a similarity degree in; an adjusting means for adjusting the calculated distance value or similarity value in accordance with the magnitude of the black pixel number difference value; Using the adjusted distance value or similarity to identify the most similar character category as a recognition result of the character pattern.

According to the present invention, a character pattern is divided into local regions, the number of black pixels existing in the local region is measured for each local region, and the number of black pixels in the black pixel number standard dictionary prepared in advance is determined. When calculating a difference value from the standard black pixel number and calculating a distance value or similarity between a feature vector obtained from the local region of the input pattern and a standard feature vector of a corresponding category of the feature standard dictionary, By adjusting so as to reduce the distance value or similarity value obtained between each local region by a value proportional to the black pixel number difference value, the distance value obtained from the local region containing much noise or An object of the present invention is to reduce variation in similarity and reduce variation in distance values or similarities obtained from the entire character pattern.

A feature of the present invention is that, in a conventional character recognition method using features, a black pixel number difference value calculation processing / means, a black pixel number standard dictionary, a black pixel number difference normalization process / means, a distance value The point is that the similarity adjustment processing / means are used together. By using this means / processing together, a feature vector due to noise (such as dirt) which is locally superimposed on the divided area, which has been generated when the character feature is recognized for each divided area of the character to be recognized conventionally. To reduce the negative impact on
Even in such a case, the character recognition rate is improved.

[0014]

Embodiments of the present invention will be described below with reference to the drawings.

[First Embodiment] FIG. 1 is a block diagram for explaining a first embodiment of a character recognition method and apparatus according to the present invention.

The pre-processing unit 1-2 calculates the width and height of the input character pattern 1-1 by using, for example, a position normalization processing method which has been known so far, thereby obtaining the center of the input character pattern. Is calculated, and the parallel movement processing of the entire input character pattern is performed so that the center is located at the center position of the character frame. Also, for example, the input character pattern is enlarged / reduced by using a conventionally known size normalization method so that the width and height of the character are the same as the width and height of the character frame. I do.

FIG. 2 shows an example in which the input character pattern is normalized by the normalization processing of the pre-processing unit for the character "pressure". FIG. 2A is an example of the input character pattern 1-1. FIG. 2B shows the character pattern after the preprocessing unit 1-2 has performed the position and size normalization processing on the input character pattern 1-1.

The feature extraction section 1-3 extracts, from the character pattern subjected to the normalization processing in the preprocessing section 1-2, features necessary for recognition such as the direction of a character line, connection relation, and positional relation in the form of a vector. Perform the extraction process.

FIG. 3 is a block diagram showing an example of the configuration of an apparatus for implementing the feature extracting section 1-3. Where 1-3
-1 is a feature value extraction unit that extracts a feature value required for recognition from a character pattern, and 1-3-2 is a feature value calculation unit that calculates a feature value in the form of a vector.

The number-of-black-pixels extracting section 1-4 includes a preprocessing section 1-2.
, A character pattern subjected to the normalization processing is input, the character pattern is divided into coarse local areas, and the number of black pixels included in the local area is measured from each local area.

FIG. 4 is a block diagram showing an example of the configuration of an apparatus for implementing the black pixel number extracting section 1-4. Where 1
-4-1 is a character pattern dividing unit that divides the character pattern into a plurality of coarse local regions, 1-4-2 is a pixel detecting unit that detects pixels existing in the local region and determines whether the pixel is white or black,
1-4-3: when the pixel detection unit determines that the pixel is a black pixel,
The black pixel number measurement unit that counts the number of pixels is a black pixel number calculation unit that outputs the number of black pixels in the local area.

The identification section 1-5 is a main part of the present invention, and includes the value of the feature vector obtained by the feature extraction section 1-3 and the number of black pixels obtained by the black pixel number extraction section 1-4. Using the feature standard dictionary 1-6 for each character and the standard dictionary 1-7 for the number of black pixels, which have already been created, the feature vector and the standard feature of each category of the feature standard dictionary 1-6 are used. In the calculation for each local area in the discriminant function of the distance value or the similarity which has been known so far between the vector and the vector, the number of black pixels in each local area of the input pattern and the black pixel number standard dictionary 1- 7, a difference value of the standard number of black pixels of each category is obtained, and a distance value or a similarity value obtained for each local region is reduced according to the magnitude of the difference value of the number of black pixels. By performing processing on all local regions, By obtaining the distance values or similarity between the category of the input character pattern and the standard dictionary, and identifies the character pattern.

FIG. 5 is a block diagram showing an example of the configuration of an apparatus for implementing the identification section 1-5. Here, 1-5-1
The distance between the feature vector obtained from the input pattern and the standard feature vector of each category of the feature standard dictionary 1-6 that has already been stored is calculated using a conventionally known distance value or similarity. The distance value / similarity calculation unit 1-5-2, which performs calculation for each local region, includes a black pixel number standard dictionary 1- 7, a black pixel number difference value calculation unit for obtaining a difference value of the number of black pixels between the standard black pixel number of each category and 1-5-3 calculates a difference value of the number of black pixels by calculating a distance value / similarity. A black pixel number difference value normalizing unit for normalizing to a numerical value corresponding to the distance value or similarity used in the unit 1-5-1, 1-5
-4 is the distance value or similarity for each local region,
A distance value / similarity adjustment unit that adjusts a distance value or a similarity value based on the normalized black pixel number difference value.

In the distance value / similarity adjustment section 1-5-4, the normalized black pixel number difference value obtained in the black pixel number difference value normalization section 1-5-3 and the distance value / similarity calculation section 1 The distance value or the similarity obtained in -5-1 is input, and a value obtained by reducing the value of the distance value or the similarity by the normalized black pixel number difference value using division or subtraction is output. Accordingly, it is an object to obtain a distance value or a similarity according to a difference value between the number of black pixels of the input pattern and the standard number of black pixels of the black pixel standard dictionary in each local region.

FIG. 6 shows a flowchart of the processing in the distance value / similarity adjustment section 1-5-4. FIG. 6 shows a processing example in the case of using division as a method of adjusting a distance value. In this example, when the normalized black pixel number difference value is 0, the distance value or similarity for each feature vector dimension obtained by the conventional method is accumulated in all dimensions, and the accumulated value is directly used for each local region. Is output as the distance value. When the normalized black pixel number difference value was greater than 0, the distance value or similarity of the conventional method obtained for each feature vector dimension was divided by the normalized black pixel number difference value, and the values were divided for all dimensions. The distance value or the similarity is accumulated, and the accumulated distance value or the similarity is output as the distance value or the similarity in each local region.

Returning to FIG. 5, 1-5-5 is a distance value / similarity calculation for adding a distance value or a similarity for each local area to calculate a distance value or a similarity for each character in the standard dictionary. Department,
1-5-6 is a sorting unit that sorts results in the order of the character type having the highest similarity from the distance value or the value of the similarity obtained from all character types, and 1-5-7 outputs the obtained identification result. This is an identification result output unit.

Next, as a specific embodiment of the character recognition method and apparatus of the present invention, as a feature for recognition, a character pattern is divided into coarse local areas, and black pixels in each local area are determined in advance. 0 °, 45 °, 90 °, 135 °, 1
80 °, 225 °, 270 °, 315 ° (1,
2, 3, 4, 5, 6, 7, 8), the tentacles are extended, the number of black pixels connected in each direction is counted, and the distribution status of the black pixels in each direction component is represented. The character pattern is identified by calculating the distance value by dividing the directional contribution degree (Japanese Patent Application No. 56-46659) by a value corresponding to the difference in the number of black pixels using the Euclidean distance as an identification function. Will be described.

FIG. 7 is a flow chart for explaining this.

First, the input character pattern cut out as an area for each character (step S1) is processed by the preprocessing unit 1-2.
Sent to The preprocessing unit 1-2 normalizes the position and size of the input character pattern (step S2). The normalized character pattern of the N × N mesh obtained by the pre-processing (step S3) includes the character value extraction unit 1-3-1 of the characteristic extraction unit 1-3 and the character pattern of the black pixel number measurement unit 1-4. It is sent to the division unit 1-4-1.

The character pattern dividing section 1-4-1 of the black pixel number extracting section 1-4 equally divides the normalized character pattern into K coarse local areas, for example, square local areas (step S4). In order to calculate the number of black pixels in each local region of the locally divided character pattern, the character pattern divided into the local region is sent to the pixel detection unit 1-4-2 (step S5).

FIG. 8 shows a specific processing flow for calculating the number of black pixels in each local area. The pixel detection unit 1-4-2
Pixels in each local area in the character pattern divided into local areas are detected (step S5-1). If the detected pixel is a black pixel (step S5-2), the counter of the number of black pixels is incremented by 1 (step S5-3). The detection and counting of the number of black pixels are performed for all the pixels in the local area, and the number of black pixels in the local area is obtained (step S5-4). The number of black pixels obtained as described above is sent to the black pixel number difference value calculation unit 1-5-1 of the identification unit 1-5.

The feature value extraction unit 1-3 of the feature extraction unit 1-3
1 is a character pattern dividing unit 1-1 of the black pixel number extracting unit 1-4.
The character pattern is divided by the same method as 4-1 (step S6). Next, a feature value is calculated in each local region of the character pattern divided into local regions (step S7).

FIG. 9 shows a specific processing flow for calculating a characteristic value in each local region. The feature value extraction unit 1-3-1
Pixels in each local area in the character pattern divided into local areas are detected (step S7-1). If the detected pixel is a black pixel (step S7-2), the counter of the number of black pixels is incremented by 1 (step S7-3). Next, a process of obtaining a direction contribution is performed on the detected black pixel (step S7-4).

FIG. 10 shows a specific processing flow for calculating the degree of directional contribution at each pixel. Feature value extraction unit 1-3
1 uses the detected pixel as a reference point (step S7-4-
1) Extend the tentacle in each direction and detect adjacent pixels (step S7-4-2). If the pixel adjacent in the scanning direction is black, the connection length counter is increased by 1 (step S7).
-4-3), a new adjacent pixel is set as a reference point (Step S)
7-4-4), the scanning process is repeated. The scanning process is
The processing is performed on the entire normalized character pattern without being limited to the pixels in the block where the detection pixel exists.

If the adjacent pixel is a white pixel or if there is no adjacent pixel, the scanning ends (step S7-4-5).
The above processing is performed for all eight directions (step S7-4)
-6). From the black pixel connection lengths in the eight directions obtained for each black pixel, a black pixel connection length cumulative value is obtained using, for example, a simple sum or a square root of a square sum (step S7-4-).
7). The directional contribution is obtained by dividing the black pixel connection length in each direction by the black pixel connection length accumulated value (step S7-4-8).

The directional contribution f of each black pixel is given by f = (α1, α2, α3, α4, α5, α6, α7, α
8) It is represented by the following eight-dimensional vector. Where α1, α2,
.., Α8 are directional contribution components in eight directions, respectively. Using a black pixel connection length li (i = 1, 2,..., 8) obtained in each direction by extending a tentacle from the black pixel in eight directions, For example, when the square root of the sum of squares is used as the cumulative value of the black pixel connection length, it is represented by αi = li / √ (Σ _{j = 1} ¹⁸ lj ² ).

The directional contribution f obtained in this manner is obtained for all black pixels in each local area, and is accumulated for each direction (step S7-5). The accumulated value of the direction contribution and the number of black pixels are sent to the feature value calculation unit 1-3-2. The feature value calculation unit 1-3-2 averages the accumulated directional contribution values by the number of black pixels in each local region to calculate a feature value in each local region (step S7-6). The feature value fk obtained in the k-th (1, 2,..., K,..., K) local region is represented by fk = (αk1, αk2,..., Αk8). Here, αk1, αk2,..., Αk8 are
Each element of the directional contribution vector obtained by accumulating the directional contribution vectors of all the black pixels existing in the k-th local region for each directional component is averaged by the number of black pixels.

The feature vector f (step S8) of the character pattern in each local region represented in this manner is
It is sent to the distance value / similarity calculation unit 1-5-1 of the identification unit 1-5.

FIG. 11 shows a specific processing flow for calculating the distance value in each local area.

The black pixel number difference value calculation unit 1-5-2 calculates the black pixel number sent from the black pixel number extraction unit 1-4 and the standard black pixel number of each category of the black pixel number standard dictionary 1-7. Then, a black pixel number difference value is calculated (step S9-1). The black pixel number difference value is sent to the black pixel number difference value normalizing section 1-5-3. The black pixel number difference value normalizing unit 1-5-3 normalizes the black pixel number difference value to a value corresponding to the Euclidean distance (step S9-2). The number of black pixels of the input pattern in the k-th local region is ak, the black pixel number standard dictionary 1-7
The standard number of black pixels of each category i (1 ≦ i ≦ M)
Then, the black pixel number difference value C1k is represented by C1k = (ak−bik).

Further, as an example of normalization in which the black pixel number difference value C1k is adjusted to the Euclidean distance, the normalized black pixel number difference value C2k is: C1k <= 0, C2k = 1, C1k> 0, C2k = (C1k / W) +1 (1 <= W <= the number of pixels included in the local area). The normalized black pixel number difference value is sent to the distance value / similarity adjustment unit 1-5-4.

The distance / similarity calculation unit 1-5-1 calculates the correspondence between the feature vector sent from the feature extraction unit 1-3 and the standard feature vector of each category of the feature standard dictionary 1-6 created in advance. Then, the square value of the difference is calculated (step S9-3). The above processing is performed for all eight directions (step S9-4). The feature vector of the input character pattern in the k-th local region is represented by fk = (α1k, α2k, α
k3, αk4, αk5, αk6, αk7, αk8), and the standard feature vector of each category i (1 ≦ i ≦ M) of the feature standard dictionary 1-6 is represented by si = (βk1, βk2, βk3, βk
4, βk5, βk6, βk7, βk8), the square value of the difference between the feature values is (αk1-βk1) ² , (αk2-βk2) ² ,.
k8-βk8) ² . The calculation result is a distance value / similarity adjustment unit 1-5.
-4.

The distance value / similarity adjusting section 1-5-4 converts the square value of the difference sent from the distance value / similarity calculating section 1-5-1 into a black pixel number normalizing section for every eight directions. Divide by the normalized number of black pixels sent from 1-5-2 and accumulate (Step S9)
-5). The values accumulated for all eight directions are calculated as distance values (step S9-6). The value obtained by dividing the square value of the difference for each direction by the number of normalized black pixels is (αk1-βk1) ² / C2k, (αk2-βk2) ² / C2k,..., (Αk8-βk8) ² / C2k ... 1), and the value Dik obtained by accumulating the above values is Dik = (αk1-βk1) ² / C2k + (αk2-βk2) ² / C2k +... + (Αk8−βk8) ² / C2k... Can be represented by

The distance value calculated for each local area is sent to the distance / similarity calculation section 1-5-5. The distance value / similarity calculation section 1-5-5 accumulates the distance values calculated from all the local areas and sends the accumulated distance values to the sorting section 1-5-6 as distance values in each category (step S10).

The distance value Di obtained between the input character pattern and each category i (1 ≦ i ≦ M) of the feature standard dictionary 1-6 is the distance value obtained from all the 1st to Kth local areas. .. + Dik +... + DiK can be expressed as Di = Di1 + Di2 +.

The sorting unit 1-5-6 sorts the distance values of all categories obtained by performing the above series of processing on all categories of the standard dictionary in ascending order (depending on other distance values / similarity, Rearrange them (in descending order) (step S11). The rearranged result is output to the identification result output unit 1
-5-7.

The identification result output section 1-5-7 outputs the character having the smallest distance value (in descending order depending on other distance values / similarity) as an identification result (step S1).
2).

In the above description, the distance value / similarity adjustment unit 1-
In 5-4, as a method of adjusting the distance value / similarity, the distance value was divided by the normalized black pixel number difference value.
Of course, any method that can reduce the distance value / similarity according to the difference value of the number of black pixels, such as subtraction, is applicable.

The effect of this embodiment in a specific character pattern will be described with reference to FIG. 17A shows an example of a normal character pattern, and FIG. 17B shows an example of a character pattern having noise in the background. K-th local region 17- (a)-
In k, 17- (b) -k, when the directional contribution feature described in the embodiment is used as a feature, 17- (a)
Table 1 shows the characteristic values Fa, 17- (b) obtained from −k and the number of black pixels B in each local region.
Table 2 shows “a” and “Bb”, and shows a conventional Euclidean distance D1 as a discriminant function and a distance value D when the method of the present embodiment is used.
2 is shown in Table 3.

[0050]

[Table 1]

[0051]

[Table 2]

[0052]

[Table 3]

From Table 1, it can be seen that the characteristic amount obtained when noise is present varies from the characteristic amount obtained when noise is not present. From Table 3, in the above situation, the Euclidean distance, which is a conventional discriminant function, obtains a value as large as about 120/1000 as a distance value obtained between local regions, but is 8/100 in the method of the present invention.
It can be seen that the variation of 0 and the distance value is suppressed. Hereinafter, by using the distance values obtained from all the local regions by the same processing, the original Euclidean distance is obtained in the region where no noise is present, and the distance is reduced according to the amount of noise in the local region where noise is present. By obtaining the value, even when noise is present in the background, the fluctuation of the distance value obtained from the entire character pattern can be reduced, and the recognition can be performed correctly.

[Embodiment 2] As a second embodiment of the present invention, the feature is the same as that of the first embodiment, and the difference in the number of black pixels is determined by using the city block distance as a discrimination function. When a character pattern is identified by calculating a distance value by subtracting a distance value by a value corresponding to
FIG. 12 shows a specific processing flow of calculating the distance value of each local region.
Shown in

The black pixel number difference value calculation unit 1-5-2 calculates the number of black pixels sent from the black pixel number extraction unit 1-4 and the standard black pixel number of each category in the black pixel standard dictionary 1-7. Then, a difference value of the number of black pixels is calculated (step S9-11). The black pixel number difference value is sent to the black pixel number difference value normalizing section 1-5-3. The black pixel number difference value normalizing unit 1-5-3 normalizes the black pixel number difference value to a value corresponding to the city block distance (step S9-12). For example, the number of black pixels of the input pattern in the k-th local area is dk, and the number of black pixels of each character i (1 ≦ i ≦ M) in the black pixel number standard dictionary 1-7 is ei.
Assuming that k, the black pixel number difference value G1k is represented by G1k = (dk-eik).

As an example of normalization of the black pixel number difference value G1k according to the city block distance, the normalized black pixel number difference value G2k is: G1k <= 0, G2k = 0 G1k> 1, G2k = G1k / V (1 <= V <= the number of pixels included in the local region). The normalized black pixel number difference value is sent to the distance value / similarity adjustment unit 1-5-4.

The distance value / similarity calculation unit 1-5-1 calculates the distance between the feature vector sent to the feature extraction unit 1-3 and the feature standard vector of each category of the feature standard dictionary 1-6 created in advance. To calculate the absolute value of the difference (step S9)
-13). The above process is performed for all eight directions (step S9-14). The feature vector of the input character pattern in the k-th local region is defined as hk = (γ1k, γ2k, γ
k3, γk4, γk5, γk6, γk7, γk8), and the feature vector of each character i (1 ≦ i ≦ M) in the feature standard dictionary 1-6 is ti = (δk1, δk2, δk3, δk4, δk
5, δk6, δk7, δk8), the absolute value of the difference between the feature values is | γk1−δk1 |, | γk2−δk2 |,.
8-δk8 |. The calculation result is a distance value / similarity adjustment unit 1-5.
-4.

The distance value / similarity adjusting section 1-5-4 calculates the absolute value of the difference sent from the distance value / similarity calculating section 1-5-1 into the black pixel number normalizing section 1 for every eight directions. The absolute value of the value subtracted by the normalized black pixel number sent from -5-2 is accumulated (step S9-15). The values accumulated for all eight directions are calculated as distance values (step S9-16).

The value obtained by subtracting the absolute value of the difference for each direction by the number of normalized black pixels is: || γk1−δk1 | −G2k |, || γk2−δk2 | −G2k |, ..., || γk8−δk8 | -G2k | ... Expression (3), and the value Jik obtained by accumulating the above values is Jik = (|| γk1-δk1 | -G2k |) + (|| γk2-δk2 | -G2k |) + .. + (|| γk8−δk8 | −G2k |) Expression (4), and the distance value in each local region is obtained.

Distance value / similarity adjustment section 1 in this embodiment
-5-4 uses subtraction as a method of adjusting the distance value / similarity. A specific processing example in this case is shown in the flowchart of FIG. In this example, subtraction is performed between the distance value or similarity obtained by the conventional method and the normalized black pixel number difference value obtained for each feature vector dimension. If the obtained distance value or similarity is 0 or more, the subtracted distance value or similarity is accumulated. When the obtained distance value or similarity is smaller than 0, the process proceeds to the calculation of the next dimension without accumulating the distance value or the similarity. Perform the above processing for all dimensions,
The finally obtained cumulative value is output as a distance value or a similarity in each local region.

In the above description, the distance value / similarity adjustment section 1-
In 5-4, as a method of adjusting the distance value / similarity, the distance value was subtracted by the normalized black pixel number difference value.
Of course, any method that can reduce the distance value / similarity according to the black pixel number difference value, such as division, can be applied.

FIG. 14 shows a K-shaped local area 14-1, 14-2,..., 14-k,.
The figure at the time of dividing into 4-K is shown.

FIG. 15 shows eight directions (15-) for extending the tentacle in order to obtain the black pixel connection length of the character pattern.
1, 15-2, 15-3,..., 15-7, 15-8).

FIG. 16 shows the k-th local region 1 in FIG.
In order to obtain the directional contribution in the 4-k black pixel,
7 shows how a tentacle is extended to obtain a black pixel connection length.

Table 4 shows the effects of the second embodiment of the present invention. Table 4 uses features similar to those of the present embodiment as features of the conventional method, uses a city block distance (conventional method 1) and a weighted city block distance (conventional method 2) as discriminant functions of the conventional method, and uses a feature dictionary. Uses the average value and standard deviation of each category, and ranks 1st and 2nd for the 3449 patterns containing much background noise as input patterns.
The cumulative classification rates for the fifth, fifth, and tenth positions are obtained.

[0066]

[Table 4]

For a pattern containing a lot of background noise,
Since the conventional method 2 using the weighted city block distance cannot cope with the fluctuation of the feature value due to the noise, the recognition rate is lower than that of the conventional method 1 using the city block distance. According to the second embodiment of the present invention in which the adjustment is made, it can be understood that the fluctuation rate due to noise can be suppressed effectively, so that the classification rate can be considerably improved as compared with the method using the conventional discriminant function.

The feature of the present invention described above is that a method using a directional contribution density feature which is a conventional character recognition method (distance value / similarity calculation unit 1-5-1, feature standard dictionary) 1-6, a method including the distance value / similarity calculation unit 1-5-5 as a constituent element), the black pixel number difference value calculation unit 1
−5-2, black pixel number standard dictionary 1-7, black pixel number difference value normalizing section 1-5-3, and distance value / similarity adjusting section 1-5-4. By these combined use, noise (such as dirt) which is locally superimposed on the divided area, which has been generated in the case of recognizing the character feature for each area obtained by dividing the character for which the direction contribution density characteristic is to be recognized in the related art. Adversely affects the feature vector, and in such a case, the character recognition rate can be improved.

To further explain the features of the present invention, the fact that the divided area has a large amount of noise means that the black pixel number standard dictionary 1-7 and the divided area are converted into the black pixel number difference value calculation unit 1- It is determined by comparing at 5-2. If you determine that there is a lot of noise,
The black pixel number difference value normalizing unit 1-5-3 performs an adjustment process, obtains a difference value C2k as an output, and uses C2k as an input to the distance value / similarity adjusting unit 1-5-4. The central processing of the present invention is the processing of the distance value / similarity adjustment section 1-5-4. As a specific method of the processing, a case where division is used as a distance value adjusting method shown in the first embodiment (FIG. 6), and a subtraction is used as a distance value adjusting method shown in the second embodiment example (FIG. 13). Details of “division of the distance value (or similarity) of each dimension by the normalized black pixel number difference value” in FIG. 6 are described in Expression (1). The details of “accumulate the divided distance value” in FIG.
It is described in. The process of “accumulating distance values of each dimension” on the right side in FIG. 6 uses a conventional method. The details of “subtract the distance value (or similarity) of each dimension by the normalized black pixel number difference value” in FIG. 13 are described in Expression (3). FIG.
The details of “accumulate the divided distance value” in 3 are given by equation (4).
It is described in.

It should be noted that some or all of the means shown in FIGS. 1, 3, 4, and 5 can be made to function using a computer, or the steps of the processing shown in FIGS. It goes without saying that a program for causing a computer to function as the means or a program for causing a computer to execute the steps of the processing can be stored in a computer-readable recording medium such as an FD. (A floppy disk), an MO, a ROM, a memory card, a CD, a DVD, a removable disk, and the like.

[0071]

As described above, according to the present invention,
The number of black pixels in each local area is measured, the difference between the number of black pixels in the standard dictionary created in advance, and the value of the difference in the number of black pixels is used to calculate the distance value or similarity value. In the case where noise is present in the background, the feature value greatly fluctuates in accordance with the amount of noise, and the distance value greatly fluctuates accordingly. The value can be reduced, thereby reducing the variation of the distance value obtained from the entire character pattern, so that a pattern that would be erroneously recognized by the conventional identification function can be correctly recognized.

In the present invention, when noise does not exist, the same distance value and similarity as those of the conventional discriminant function can be obtained by appropriately performing the normalization of the difference value of the number of black pixels. It is possible to obtain the same discrimination ability as before without giving an adverse effect to a character pattern that does not exist.

The present invention is not particularly limited to a unique feature extraction technique, and any technique that divides a character pattern into coarse local areas and extracts features used for recognition from each of the local areas is applicable. Therefore, we applied it to various feature extraction methods that have been proposed as character recognition methods,
By making it possible to recognize a character pattern containing noise, it is possible to improve the performance of various methods.

In the conventional method of performing recognition based on the degree of black pixel overlap in a character pattern, the position of a pixel at the time of superposition is slightly shifted or the font of a character is different.
Although the recognition performance has deteriorated significantly, the present invention uses a method of extracting features from character patterns, so that the recognition performance is not significantly improved even for pixel displacements and differences in character fonts within a coarse local area. A pattern in which noise exists can be recognized without lowering.

[Brief description of the drawings]

FIG. 1 shows a first example of a character recognition method and apparatus according to the present invention.
FIG. 2 is a configuration diagram showing an example of the embodiment.

FIGS. 2A and 2B are diagrams showing a normalization process in a preprocessing unit according to the first embodiment. FIG.

FIG. 3 is a block diagram illustrating an example of a configuration of an apparatus that implements a feature extracting unit according to the first embodiment.

FIG. 4 is a block diagram illustrating an example of a configuration of an apparatus that implements a black pixel number extracting unit according to the first embodiment.

FIG. 5 is a block diagram illustrating an example of a configuration of a device that implements an identification unit according to the first embodiment.

FIG. 6 is a diagram illustrating a processing example in a case where division is used as a distance value adjustment method in the distance value / similarity adjustment unit according to the first embodiment.

FIG. 7 is a flowchart for explaining processing in the first embodiment.

FIG. 8 is a flowchart for explaining a specific processing example of calculating the number of black pixels in a local region in the first embodiment.

FIG. 9 is a flowchart for explaining a specific processing example of feature value calculation in each local region in the first embodiment.

FIG. 10 is a flowchart for explaining a specific example of processing for calculating a directional contribution at each pixel according to the first embodiment.

FIG. 11 is a flowchart illustrating a specific process of calculating a distance value in each local region in the first embodiment.

FIG. 12 is a flowchart for explaining a specific processing example of calculating a distance value of each local region in the second embodiment of the present invention.

FIG. 13 is a diagram illustrating a processing example in a case where subtraction is used as a distance value adjustment method in the distance value / similarity adjustment unit of the second embodiment.

FIG. 14 is a diagram illustrating a state in which a character pattern is divided into coarse local regions in a feature extraction unit according to the embodiment of the present invention.

FIG. 15 illustrates a direction in which a tentacle is extended to obtain a black pixel connection length in the feature extraction unit according to the embodiment of the present invention.
It is a figure showing the case where it makes it eight directions.

FIG. 16 is a diagram illustrating a state where a black pixel connection length is obtained in a feature extraction unit according to the embodiment of the present invention.

FIGS. 17A and 17B are diagrams illustrating a character pattern when noise is present in the background as an effect of the present invention.

[Description of Signs] 1-1: Input character pattern 1-2: Pre-processing unit 1-3: Feature extraction unit 1-4: Black pixel number extraction unit 1-5: Identification processing unit 1-6: Feature standard dictionary 1 -7: black pixel number standard dictionary 1-8: identification result 14-1, 14-2 to 14-k to 14-K: each local area in the case where a character pattern is divided into coarse K square local areas 15 -1, 15-2, 15-3, 15-4, 15-5,
15-6, 15-7, 15-15 ... Direction of extending tentacles

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Toshiaki Sugimura 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo F-term in Nippon Telegraph and Telephone Corporation (reference) 5B064 AA01 BA01 CA11 DC19 DC27 DC27 DC28 DC39 DC42

Claims (5)

[Claims] 1. A normalization process is performed on a position and a size of a character pattern cut out and binarized for each character, and a process of extracting a feature from the normalized character pattern is performed. Apart from the process of extracting the feature, a process of dividing the normalized character pattern into local regions is performed. For each local region of the divided character pattern, the number of black pixels present in the local region is calculated. Performing a process of measuring a black pixel number difference value between the black pixel number obtained by the measurement and the standard black pixel number of each category of the black pixel number standard dictionary prepared in advance; The distance value or the similarity value obtained between the feature vector obtained from the character pattern in the above and the standard feature vector of the corresponding category of the previously created feature standard dictionary is calculated as the large value of the black pixel number difference value. Adjusted if the, by using the adjustment distance values or similarity,
A character recognition processing method, wherein the most similar character category is used as a recognition result of the character pattern. 2. A distance value or similarity value obtained between a feature vector obtained from a character pattern in the local area and a standard feature vector of a corresponding category of a previously created feature standard dictionary is calculated. In the process of adjusting according to the size of the black pixel number difference value, if the black pixel number difference value is 0, the distance values or similarities calculated for all dimensions of the feature vector are accumulated, and If the black pixel number difference value is larger than 0, the distance value or similarity of each dimension of the feature vector is divided by the black pixel number difference value and accumulated, and the accumulated value is adjusted for the local region. 2. The character recognition processing method according to claim 1, wherein the distance value or the similarity is used. 3. A distance value or a similarity value obtained between a feature vector obtained from a character pattern in the local area and a standard feature vector of a corresponding category of a previously created feature standard dictionary is calculated. In the process of adjusting according to the magnitude of the black pixel number difference value, the distance value or similarity of each dimension of the feature vector is subtracted by the black pixel number difference value, and the subtracted distance value or similarity is calculated. 2. The character recognition processing method according to claim 1, wherein the subtracted distance value or similarity in the case of 0 or more is accumulated, and the accumulated value is used as an adjusted distance value or similarity of the local region. . 4. A pre-processing means for normalizing a position and a size of a character pattern cut out and binarized for each character, and a feature for extracting a characteristic from the normalized character pattern. Extracting means for dividing the normalized character pattern into local areas, and for each local area of the divided character pattern, black pixel number extracting means for measuring the number of black pixels existing in the local area; Black pixel number difference value calculating means for obtaining a black pixel number difference value between the black pixel number obtained by the measurement and the standard black pixel number of each category of the black pixel number standard dictionary created in advance; A distance value or similarity calculator for calculating a distance value or similarity between a feature vector obtained from a character pattern in a local area and a standard feature vector of a corresponding category in a previously created feature standard dictionary. When the value of the calculated distance values or similarity, and adjusting means for adjusting in accordance with the size of the black-pixel count difference value, by using the distance value or the degree of similarity the adjustment,
A recognition unit that determines a most similar character category as a recognition result of the character pattern. 5. A program for causing a computer to execute a processing step in the character recognition processing method according to claim 1 on a recording medium readable by the computer. A recording medium that records a character recognition processing method as a feature.