JPH07282199A - Picture pattern identifying and recognizing method - Google Patents

Abstract

PURPOSE:To reduce the burden of the correcting work on a user in the case of erroneous recognition, and simultaneously, to easily add a recognized pattern to a dictionary so that the identification and the addition of the said pattern can be executed hereafter. CONSTITUTION:A binary image pattern storage circuit 1 stores the binary image pattern of the character or the graphic of an object to be identified or recognized. A binary mask pattern storage circuit 3 stores the standard dictionary of the binary mask pattern of a character or a graphic category. A normalization processing circuit 2 inputs the image pattern of the binary image pattern storage circuit 1, and executes the normalization of a position or side. An identification.recognition judgement circuit 4 collates the normalized image pattern and the binary mask pattern classified by category with each other, and executes identification, rejection or recognition processing. A binary mask pattern generation circuit 5 generates the binary mask pattern from the image pattern recognized erroneously or not yet learnt, and executes the addition and the alteration of the contents of the binary mask pattern storage circuit 3.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image pattern identification / recognition method for a binary pattern in which an input pattern is represented by white or black. Printed kanji, which is a typical example of a binary pattern,
When targeting many character / graphic categories such as handwritten Chinese characters, alphanumeric characters, symbols, mathematical expressions, and graphics, images with noise such as faintness and crushing, character images that have undergone design processing consisting of striped patterns, etc. Further, the present invention can be applied to a character recognition method for recognizing white characters. It can also be applied to a method of recognizing a plurality of categories of disorders and diseases that occur in a system based on the diagnostic results of many items used in disasters, medical care, and the like.

[0002]

2. Description of the Related Art Traditionally, Kanji OCR (Optical Character)
In a character recognition processing device such as r Reader) or a drawing recognition device, a feature for recognition is extracted in the form of a vector from a character or a graphic pattern, and the feature pattern is compared with a standard pattern vector of each category in a standard dictionary created in advance. A method is known in which a discriminant function such as similarity or dissimilarity is obtained, and the most similar character or figure category is used as the recognition result.
Characteristic quantities that reflect character line structures such as character line direction, connection relationships, and positional relationships are widely used for the elements of conventional feature vectors such as character recognition, but a design with a unique texture. It was almost impossible to obtain sufficient recognition accuracy because these feature values fluctuate greatly with respect to an image with severe characters, faintness, blurring, character background noise, and the like.

In the conventional character recognition, an input / output black-and-white pattern is used as an element of a feature vector, and a standard pattern vector (represented by a binary pattern similar to the feature vector) of each category in a standard dictionary created in advance. A simple similarity is known as a discriminant function based on the number of black pixels between the pixels. However, with this method, there is a problem in that an image having a design texture or a character background noise that is composed of a peculiar texture is erroneously recognized as a category having a large number of black pixels in the standard pattern vector.

[0004]

As described above, the conventional recognition technology such as OCR correctly recognizes characters which are erroneously recognized due to design characters, faintness, crushing, character background noise, and non-white characters. There was a problem that the recognizable method was not fully established. It is an object of the present invention to provide an image pattern identification / recognition method having a recognition function capable of coping with design characters, severe character deformation such as faint and crushed characters, and character strings in which blank characters are mixed.

[0005]

As a first method, a black part of an input binary image pattern is included in a black part of a binary mask pattern of a certain recognition category A, and the black part of the input binary image pattern is included. , The black part of the binary mask pattern of the recognition category A and the other two of the other recognition category B different from that.
There is a method of identifying that the input binary image pattern belongs to the recognition category A when it is not included only in the overlapping portion with the black portion of the value mask pattern. Furthermore, when there is no recognition category in such a procedure, the black part of the input binary image pattern is the black part of the binary mask pattern of the recognition category A and the black part of the other binary mask pattern of the recognition category B. If it is included only in the overlapping portion, it may be rejected. Although these methods focus on the black part, the same procedure may be configured by focusing on the white part.

As a second method, binary image patterns and 2
The value mask pattern has a first degree of matching based on the number of black pixels, and the second degree of matching with a number of white pixels on the binary mask pattern. The first degree of disagreement is based on the number of black pixels in the value mask pattern, and the second degree of disagreement is based on the number of black pixels in the binary image pattern and white pixels in the binary mask pattern. Then, the third degree of coincidence is obtained based on the first degree of coincidence and the second degree of coincidence, and the third degree of disagreement is obtained based on the first degree of disagreement and the second degree of disagreement. The similarity between the binary image pattern and the binary mask pattern is calculated based on the third degree of coincidence and the third degree of disagreement, and the binary image pattern is recognized as belonging to the recognition category having the highest degree of similarity. There is a way to do it. In this method, even if not all positions in the pattern are used, a plurality of positions in the pattern are selected in advance, and the procedure is configured to obtain the degree of coincidence and the degree of disagreement at the plurality of positions. Good.

As a third method, a threshold value is set in advance for each binary mask pattern of each recognition category and 2
The similarity between the binary image pattern and each binary mask pattern is obtained, and the similarity between the binary image pattern and the binary mask pattern of the recognition category is the binary mask for only one recognition category. There is a method of identifying a binary image pattern as belonging to the recognition category when it is equal to or larger than the threshold value of the pattern. Further, according to this procedure, the input binary image pattern does not belong to any recognition category, and for two or more recognition categories, the degree of similarity between this binary image pattern and the binary mask pattern of the recognition category is the same. May be configured to reject the input binary image pattern when is greater than or equal to the threshold value of the binary mask pattern.

The first method to the third method may be combined.

In the first method to the third method, a binary mask pattern is newly created by iterative learning using a predetermined similarity based on the information on the black part or the white part of the input binary image pattern. , May be added or updated.

[0010]

According to the above configuration, the input binary image pattern is collated with each binary mask pattern created in advance for each recognition category to efficiently identify the recognition category to which the binary image pattern belongs, Can be recognized or rejected. Further, a binary mask pattern can be newly created, added, or updated by iterative learning based on the information of the input image pattern.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration example of an image pattern identification / recognition processing device to which the method of the present invention is applied. 1 is a binary image pattern storage circuit, 2 is a normalization processing circuit, 3 is a binary mask pattern storage circuit. Reference numeral 4 is an identification / recognition determination circuit, 5 is a binary mask pattern creation circuit, and 6 is an image input / output device. The binary image pattern storage circuit 1 stores binary image patterns of characters or figures to be identified or recognized. The binary image pattern consists of n pixels, for example, white pixels are "0" and black pixels are "1".

The normalization processing circuit 2 inputs a binary image pattern consisting of n pixels extracted from the binary image pattern storage circuit 1, and, for example, the conventionally known center of gravity and 2
Position and size are normalized using the second moment. The binary mask pattern storage circuit 3 stores in advance one or a plurality (generally a plurality) of binary mask patterns of each category used for identification and recognition determination of an input image pattern.

The identification / recognition determination circuit 4 inputs the binary image pattern of n pixels from the normalization processing circuit 2, collates it with the binary mask pattern of each category of the binary mask pattern storage circuit 3, and, for example, The black part (or white part) of the input image pattern is included in the black part (or white part) of the binary mask pattern of a specific category, and the black part (or white part) of the binary mask pattern and the binary of another category If it is not included only in the overlapping part with the mask pattern black part (or white part), it is identified and recognized as that particular category,
If the black part (or white part) of the input image pattern is included only in the overlapping part, it is rejected. In other cases, based on the information of the white part or the black part of the input image pattern, the known similarity degree, etc. The identification function or the identification function for recognizing a white-colored character with high accuracy is recognized or rejected, and the recognition result is output.

The binary mask pattern creating circuit 5 receives the image information output from the image input / output device 6 and the normalized binary image pattern of the addition / update or newly created category from the normalization processing circuit 2, Create a binary mask pattern. The binary mask pattern created by the binary mask pattern creation circuit 5 is newly registered in the binary mask pattern storage circuit 3, or added / updated to the contents of the binary mask pattern storage circuit 3.

The image input / output device 6 inputs the information of the classification result such as the character or figure category number, the discrimination function value, and the binary image pattern output from the identification / recognition determination circuit 4, and displays the information. Further, for example, the image information is input and displayed by a keyboard or a mouse. further,
The image input / output device 6 sends to the binary mask pattern creating circuit 5 information necessary for new creation or update of the binary mask pattern.

First, the input image pattern identification and recognition processing will be described with reference to FIG. After the binary image pattern of the character to be identified / recognized is read from the binary image pattern storage circuit 1, the position and size of the character image pattern are normalized in the normalization processing circuit 2, and then the character image pattern is identified. -It is input to the recognition determination circuit 4. here,
The character image pattern normalized by the normalization processing circuit 2 will be referred to as a normalized character pattern X. The normalized character pattern X can be expressed by a vector having n elements such that X = (x ₁ , x ₂ , ..., X _i , ... X _n ), and each element x _i is an image position i (= 1, 2, 1. , ..., N) pixel values (eg, x _i =
“0” is a white dot, and x _i = “1” is a black dot).

The identification / recognition determination circuit 4 operates on the jth character of the binary mask pattern storage circuit 3 for the normalized character pattern X.
Character category (j = 1, 2, ..., J) kth (k =
1, 2, ..., K _j ) binary mask pattern M _jk . When M _jk is expressed as M _jk = (m ₁ , m ₂ , ... M _i , ..., M _n ) _jk , each m _i takes a binary value, for example, if m _i = “0”, a white point, m _i = “1” indicates a black dot.

As an example of collation, a case where the normalized character pattern X and the binary mask pattern M _jk are collated will be described.
FIG. 2 shows an example of a processing flowchart. First,
Between the element x _i of the normalized character pattern X and the element m _i of the binary mask pattern M _jk , (a) x _i = “1”, the number A, and (b) m _i = “1”, B _jk and (c) x _i
The number C _{jk such} that = "1" and m _i = "1", and (d) x _i
The number D _{jk for} which "= 1" and m _i = "0" is counted (step 101).

Next, determination processing is performed in order from the following condition (1) to condition (3), and identification, rejection, and recognition are determined (step 102). Condition (1): If there is only one character category j satisfying C _jk ≦ B _jk and A = C _jk , it is determined that the input character can be identified as the character category j (step 110). Condition (2): If there are two or more character categories j satisfying C _jk ≦ B _jk and A = C _jk , it is determined that the input character is rejected (step 120). Condition (3): When there is no character category that satisfies the conditions (1) and (2), characters with a lot of background noise or between the normalized character pattern X and the binary mask pattern M _jk or A recognition process is performed using a discriminant function for recognizing white characters or a known discriminant function (step 103), and recognition or rejection is determined (step 104). If recognized, the character category is set as a candidate category (step 130).

Here, as an example of the recognition processing (step 103) in the case of the condition (3), the case where the new similarity S _H (X, M _jk ) according to the present invention is used for the discriminant function will be described. That is, the number of black pixels A of the normalized character pattern X
Then, the similarity S _H (X, M _jk ) represented by the following equation is calculated.

[0021]

[Equation 1]

As the recognition determination result, the character category j having the maximum similarity is output as the first recognized character category. In this case, since the similarity S _H (X, M _jk ) does not need to extract the feature amount related to the character line structure such as the direction of the character line and the connection relationship, the feature extraction process is simplified, and the blurring Even if these character line structures are deformed due to crushing, noise in the character background, etc., the recognition performance is not easily affected. In the recognition processing in the case of the condition (3) is the similarity S _H (X, M _jk) the absolute value of _{| S H (X, M jk} ) | may be used to identify the function.

Next, a specific process of the identification / recognition / rejection determination process will be described with reference to FIG. In FIG. 3, n = 16
(Pixel), two-character category (j = 1, 2), K ₁ = 1,
It shows an example of a normalized character pattern X and a binary mask pattern M _jk when K ₂ = 2. Reference numeral 11 denotes a binary mask pattern M ₁₁ (first binary mask pattern of the first character category), and the number B ₁₁ = 8 for mi = “1”. 12 is a binary mask pattern M ₂₁ (first binary mask pattern in the second character category), and m _i = “1”
Therefore, the number B ₂₁ = 7. Reference numeral 13 is a binary mask pattern M ₂₂ (second binary mask pattern of the second character category), and the number B ₂₂ = 8 where m _i = “1”. Two
1 to 26 are six pattern examples of the normalized character pattern X. The normalized character pattern 25 is a white-out character pattern obtained by inverting the normalized character pattern of the normalized character pattern 24 in black and white.

First, each of the normalized character pattern X of the normalized character pattern 21 and each of the binary mask patterns 11 to 13 is used.
A case of matching with the value mask pattern M _jk will be described. x
_{When the} number A is 5 where _i = “1”, x _i = “1” and m _i =
The numbers of "1" are C ₁₁ = 5, C ₂₁ = 4 and C _{22 respectively.}
= 3 and only the binary mask pattern M _{11 of the} binary mask pattern ₁₁ satisfies the condition (1), so the normalized character pattern X of the normalized character pattern 21 is identified in the first character category.

Next, in the case of the normalized character pattern X of the normalized character pattern 22, A = 6, x _i = “1” and m _i
The numbers for which “= 1” are C ₁₁ = 5, C ₂₁ = 6, and C, respectively.
_{Since 22} = 5 and only the binary mask pattern M ₂₁ satisfies the condition (1), the normalized character pattern X of the normalized character pattern 22 is identified as the second character category.

Next, in the case of the normalized character pattern X of the normalized character pattern 23, A = 4, x _i = “1” and m _i
= ₁₁ are the numbers of C ₁₁ = 4, C ₂₁ = 4, C
₂₂ = 3, and the binary mask pattern M _{11 of the} binary mask pattern ₁₁ and the binary mask pattern M _{21 of the} binary mask pattern 12 satisfy the condition (2). Therefore, the normalized character pattern 23 is the normalized character pattern 23. X is rejected.

In the case of the normalized character pattern X of the normalized character pattern 24, A = 7, x _i = "1" and m _i.
The numbers for which “= 1” are C ₁₁ = 5, C ₂₁ = 6, and C, respectively.
_{Since 22} = 6 and none of the binary mask patterns 11 to 13 satisfy the conditions (1) and (2), the discrimination function of the condition (3) is executed.

In this case, the similarity S _H between the normalized character pattern X of the normalized character pattern 24 and each of the binary mask patterns M _jk of the binary mask patterns 11 to 13 is

[0029]

[Equation 2]

Then, the binary mask pattern M _{21 of the} binary mask pattern 12 has the highest similarity, and the second character category is the first recognition result.

Finally, in the case of a normalized pattern X of the normalized pattern 25, that is, a white character pattern in which the normalized character pattern 24 is reversed in black and white, A = 9 and x _i =
The numbers of "1" and m _i = "1" are C ₁₁ =
3, C ₂₁ = 1 and C ₂₂ = 2, and binary mask pattern 1
Since none of the binary mask patterns M _{jk of} 1 to 13 satisfy the conditions (1) and (2), the discriminant function of the condition (3) is executed. Here, as the discriminant function, instead of the similarity S _H (X, M _jk ), the absolute value | S
_{When H} (X, M _jk ) | is used, the normalized character pattern 2
5 normalized character pattern X and binary mask patterns 11 to 11
Each binary mask pattern M _jk and similarity of 13 | S _H |
Is

[0032]

[Equation 3]

Then, the binary mask pattern M _{21 of the} binary mask pattern 12 has the highest degree of similarity, and similarly for the normalized character pattern 25, which is the black-and-white inverted pattern of the normalized character pattern 24, the second The character category is the first recognition result.

Next, another embodiment of the recognition process (step 103) for the above-mentioned condition (3) will be described. In the following example, as the discriminant function in the recognition process under the condition (3), another novel similarity S according to the present invention shown in the following formula
_c (X, M _jk ) is used.

[0035]

[Equation 4]

The numerator term of this new similarity S _c (X, M _jk ) is the number C _jk and x where x _i = “1” and m _i = “1” in the input and the binary mask pattern. _i = "0" and m _i =
From the product with the number of “0” (n−B _jk −D _jk ), x _i
= "0" and the value obtained by subtracting the product of the m _i = "1" and becomes the number (B _jk -C _jk) and x _i = "1" and m _i = "0" becomes the number D _jk, C _jk (n−B _jk −D _jk ) − (B _jk −C _jk ) · D _jk = n ·
It is derived from C _jk- A · B _jk . Normalized character pattern 2 shown in FIG.
4 normalized character pattern X and each binary mask pattern 11
The similarity S _c with each binary mask pattern M _{jk of}

[0037]

[Equation 5]

As in the case of using the similarity S _H (X, M _jk ), the binary mask pattern M _{21 of the} binary mask pattern 12 has the highest similarity and the second character category is the first. It becomes the recognition result of rank.

Further, a normalized character pattern 25, which is an outline character pattern obtained by inverting the normalized character pattern 24 in black and white.
In the case of the normalized character pattern X of, the similarity S _c (X, M
_jk ) absolute value | S _c (X, M _{j k} ) | is used, the normalized character pattern 25 and each binary mask pattern 11 to 13
And the similarity | S _c |

[0040]

[Equation 6]

Then, the binary mask pattern M _{21 of the} binary mask pattern 12 has the highest similarity, and the normalized character pattern 25 is the same as the case of the normalized character pattern 24.
Also, the second character category is the first recognition result.

Next, another method of the identification / rejection determination processing will be described. In this method, in the determination of the above conditions (1) to (3), the new similarity S described above for the discriminant function is used.
Processing is performed using _c (X, M _{j k} ). In the following description, the conditions corresponding to the above conditions (1) to (3) will be referred to as conditions (1a) to (3a), respectively.

In this identification / recognition / rejection determination processing, the element x _i of the normalized character pattern X and the binary mask pattern M _jk
A, B _jk , and C _jk are counted with the element m _i of the above (see step 101), and the similarity S _c (X, M _jk ) is calculated, and stored in the binary mask pattern storage circuit 3 in advance. Using the stored identification threshold value t _jk of each binary mask pattern, the processes of the following conditions (1a) to (3a) are sequentially performed (see step 102). The details of the method of setting the identification threshold value t _jk of the binary mask pattern will be described later.

Condition (1a): When there is only one character category j satisfying S _c (X, M _jk ) ≧ t _jk , it is determined that the input character can be identified in the character category j (step 110). reference).

Condition (2a): When there are two or more character categories j satisfying S _c (X, M _jk ) ≧ t _jk , it is determined that the input character is rejected (see step 120).

Condition (3a): When there is no character category satisfying condition (1a) or condition (2a), between the normalized character pattern X and the binary mask pattern M _jk ,
The recognition process is performed using the similarity S _c as a discriminant function (see step 103), and when there is a character category j that satisfies the similarity S _c (X, M _jk ) ≧ u _jk , the similarity is highest. The character category of the binary mask pattern M _jk is output as the first recognition result (see step 130). On the other hand, when there is no character category j that satisfies the similarity S _c (X, M _jk ) ≧ u _jk , that is, the similarity S _c (X, M _jk ) <u _jk for all the binary mask patterns M _jk. If so, it is determined to be recognition rejection (see step 120). The variable u _jk is a preset threshold value for rejection determination for each binary mask pattern M _jk , and is a value smaller than the identification threshold value t _jk (u _{j k} <
t _jk ).

Here, the specific processing of the identification / rejection determination processing will be described with reference to FIG. 3 again. _First , a case where the normalized character pattern X of the normalized character pattern 22 is compared with each binary mask pattern M _{jk of the} binary mask patterns 11 to 13 will be described. The similarity S _c between the normalized character pattern 22 and each of the binary mask patterns 11 to 13 is

[0048]

[Equation 7]

Since only the binary mask pattern M ₂₁ satisfies the condition (1a), the normalized character pattern X of the normalized character pattern 22 is identified as the second character category.
Next, in the case of the normalized character pattern 26, the similarity S with each binary mask pattern of the binary mask patterns 11 to 13 is S.
_c is

[0050]

[Equation 8]

Since the binary mask pattern M _{11 of the} binary mask pattern ₁₁ and the binary mask pattern M _{21 of the} binary mask pattern 12 satisfy the condition (2a), 26 normalized character patterns X are rejected. . Further, in the case of the normalized character pattern 23, each of the binary mask patterns 11 to 13
And the similarity S _c is

[0052]

[Equation 9]

Since none of the binary mask patterns 11 to 13 satisfy the condition (1a) and the condition (2a), the similarity S _c of the condition (3a) is satisfied.
The discriminant function of is executed. Under the condition (3a), since the binary mask pattern M _{21 of the} binary mask pattern 12 has the highest similarity, the second character category is the first recognition result.

Next, regarding the processing for newly creating a binary mask pattern from a character that has been erroneously recognized or an unlearned character, and adding / updating the contents of the binary mask pattern storage circuit 3,
A few examples will be described.

<Embodiment 1> This is a binary mask pattern obtained by iterative learning using the similarity S shown in the following equation as a known discriminant function based on the information on the black portion or the white portion of the input image pattern. This is an example of creating.

[0056]

[Equation 10]

First, the image input / output device 6 inputs information on a character category (jth character category) to be added / updated or newly created (added here) to the binary mask pattern creation circuit 5. To be done. The binary mask pattern creation circuit 5 receives N input image patterns (normalized character patterns) X ₁ belonging to the j-th character category to be learned, which are obtained from the normalization processing circuit 2 for the input image to be added. , X ₂ , ..., X _N , a new binary mask pattern M _{j (k + 1)} is created in the following steps.

(Step 1) N input image patterns X
_1, X _2, ..., the X _N, which counts the total number A _i of the input image pattern X as the x _i = "1" a total of A _T and the pixel position i of the number A of the x _i = "1" . When A _i is rearranged in descending order and the columns are set as A ′ ₁ , A ′ ₂ , ..., A ′ _l , ..., A ′ _n, with respect to a preset threshold value a (0 <a <1) hand

[0059]

[Equation 11]

L that satisfies the above condition is determined. However, when N is small, for example, when N = 1, a = 1 is set.

(Step 2) For the image position i corresponding to each rank of l = 1, 2, ..., L, the element of the binary mask pattern M _{j (k + 1)} is set to m _i = "1", and The image position i other than is set to m _i = “0”.

(Step 3) N input image patterns X
Each image X _{t of 1} , X ₂ , ..., X _N and 2 created in step 2
The similarity S (X _t , M ₎ with the value mask pattern M _{j (k + 1)
j (k + 1)} ) is calculated to _{find the} image X _t having the maximum similarity,
The image X _t is replaced with a new binary mask pattern M _{j ( k + 1)} .

(Step 4) Again, for each of the N input images X ₁ , X ₂ , ..., X _N , each image X _l and all the two images already stored in the binary mask pattern storage circuit 3 are input.
The similarity S between the value mask pattern M _jk and the binary mask pattern M _{j (k + 1)} created in step 3 is calculated and recognized. If all N images are recognized in the jth character category, go to step 6, otherwise go to step 5.

(Step 5) For M (<N) images correctly recognized in the jth character category, the binary mask pattern M is processed in the same procedure as in Steps 1 and 2.
_{j (k + 1)} is created and stored in the binary mask pattern storage circuit 3. For the remaining (N−M) images, k = k +
1, N = NM, and the process returns to step 1.

(Step 6) A binary mask pattern M _{j (k + 1)} is created in the same procedure as in Steps 1 and 2, and is stored in the binary mask pattern storage circuit 3 to end the process.

<Embodiment 2> This is a new binary mask for each character category by iterative learning using the similarity S _H based on the information of the black part or white part of the input image pattern (sample pattern). It is an example of creating a pattern and an identification threshold.

First, the image input / output device 6 inputs information on a new character category (j-th character category) to be created to the binary mask pattern creation circuit 5.
The binary mask pattern creation circuit 5 obtains from the normalization processing circuit 2 for each input image, for each character category j = 1, 2, ..., J to be newly created,
In the following steps, K _j binary mask patterns M _j1 , M
Create _j2 , ..., M _jk , ..., M _{j k j} .

(Step 1a) Let K _j = 1 be the number of binary mask patterns of the jth character category to be newly created. For N _j input image patterns (normalized character patterns) X ₁ , X ₂ , ... X _Nj belonging to the jth character category,
In x _i = "1" Piece A _T and the pixel position i x _i = "1"
Then, the number A _i is calculated. Then, at the pixel position i, x _i =
The ratio of the number of “1” s A _i / N _j and x _{i of} all pixels =
The binary mask pattern M _j1 is created by comparing the ratio _AT / (nN _j ) of the number of “1” as in the following equation. If A
_{If i} / N _j ≧ A _T / (nN _j ), set the element of M _j1 to m _i = "1", otherwise set the element of M _j1 to m _i = "0" (Normalized character pattern) X ₁ ,
A label j1 is assigned to X ₂ , ..., X _Nj as the pattern used to create the binary mask pattern M _j1 . The above process is performed for all character categories to be newly created.

(Step 2a) For all input image patterns X _t to be newly created, the similarity S _H (X _t , M _jk ) with all binary mask patterns M _{jk of} all character categories is calculated, _Find M _lh with the maximum similarity. Now, it is assumed that the label jk is already assigned to the input image pattern X _t . If l = j, X _t is determined to be correctly recognized in the j-th character category, and a new label jh is assigned. Otherwise, X _t is determined to have been erroneously recognized, and a new label je is assigned. After this process is completed, if for all input image patterns X _t, unless erroneous recognized and determines X _t is, the flow proceeds to step 4a. Otherwise, go to step 3a.

(Step 3a) Set h = 1 for the jth character category to be newly created. For all input image patterns to which the label jk is assigned, if there is one or more input image patterns, a new binary mask pattern M _jh is created by the same procedure as in step 1a, and h = h + 1 is set. . This determination process is performed for all K _j labels in the j-th character category. Next, if there is one or more input image patterns to which the label je is assigned, the temporary binary mask pattern M is processed for all of these input image patterns X _{t in} the same procedure as in step 1a. create _je and all X _t
The similarity S _H (X _t , M _je ) is calculated between M _je and M _je, and the input image pattern X _S having the maximum similarity is set as a new binary mask pattern M _jh, and h = h + 1. Let K _j = h be the number of new binary mask patterns for the jth character category.
far. This process is repeated for all the new character categories to be created, and new binary mask patterns are stored in the binary mask pattern storage circuit 3. Then, the process returns to step 2a.

(Step 4a) Each binary mask pattern M
_jk and between all the input image patterns X _t other than the j character category, the similarity _{S H (X t, M jk} ) was calculated to obtain the maximum similarity t _jk for each binary mask pattern M _jk , The maximum similarity is the identification threshold t of the binary mask pattern M _jk.
It is assigned to _jk and stored in the binary mask pattern storage circuit 3. When these processes are completed for all binary mask patterns, the process ends.

<Third Embodiment> This is similar to the second embodiment in that the similarity S _H is used, but the binary mask pattern and the identification threshold are determined by using the image pattern of the erroneously recognized character or the unlearned character. It is an example of updating a value. First, the image input / output device 6 inputs information about a character category to be updated (referred to as a j-th character category) to the binary mask pattern creation circuit 5. The binary mask pattern creation circuit 5 performs a binary mask on the input image in the following steps for each of the character categories j = 1, 2, ..., J to be updated, which are obtained from the normalization processing circuit 2. Update the pattern.

(Step 1b) Let K _j be the number of binary mask patterns in the j-th character category before updating. For all input image patterns X _t to be updated, all binary mask patterns M _jk (j = 1, ..., J) (k =
1, 2, ..., K _j ) and the similarity S _H (X _t , M _jk ) is calculated to obtain M _lh having the maximum similarity. Now, it is assumed that the input image pattern X _t belongs to the j-th character category. If l ≠ j, it is determined that X _t is erroneously recognized, and a label je is newly assigned. For all character categories to be updated, after this process is completed, if for all input image patterns X _t, unless erroneous recognized and determines X _t is,
Go to step 3b. Otherwise, go to step 2b.

(Step 2b) With respect to all the input image patterns X _t to which the label je is assigned, a temporary binary mask pattern M _je is created by the same procedure as in step 1a of the second embodiment, and all X _t. and M similarity _{S H (X t, M je} ) between _je was calculated, binary mask pattern should be added to the input image pattern X _s having the maximum similarity M _{j (Kj + 1)}
Is stored in the binary mask pattern storage circuit 3. In addition, K _j = K _j +1 Distant, returns to step 1b.

(Step 3b) Each binary mask pattern M
If there is _jk and all the input image patterns X _t other than the j character category, the similarity _{S H (X t, M jk} ) was calculated between the X _t, the maximum for each binary mask pattern M _jk When the similarity t ′ _jk is obtained and the maximum similarity is greater than the identification threshold t _jk (t _jk = −n when newly created) stored in the binary mask pattern storage circuit 3, Is the identification threshold value t _jk = t ′ of the binary mask pattern M _jk.
Updated to _jk and set new value to binary mask pattern storage circuit 3
Remember. When these processes are completed for all binary mask patterns, the process ends.

Here, in the second and third embodiments, the case where the similarity S _H is used as the discriminant function is described, but it is also possible to use the absolute value of the similarity S _H as the discriminant function. .

As a method of creating a binary mask pattern, other known methods by statistical inference (for example, "Statistics" by Tetsuo Suzuki (Asakura Shoten) and "Pat") are used.
tern classification and scence analysis ”R.I. O.
Duda and P.D. E. By Hart (John Wiley & Sons I)
ns.)) method can also be used. Also,
Also in the first embodiment, even if there is an image pattern that is not included in any character category, it is possible to create a new binary mask pattern of the character category using the image pattern.

Further, in the above description, characters are mainly shown, but general binary images such as figures can be processed in the same manner. As an example, a case where the method is applied to a method for recognizing a category of a disease based on the diagnostic results of many items used in medical treatment will be briefly described below. For example, the results of a plurality of diagnostic items are displayed on a check sheet as a binary image pattern as shown in FIG. 3 (for example, for a blood pressure item, "1" is given if the blood pressure is high, and "0" if it is low. "Black" or "white" is displayed at the position corresponding to that item.) The pattern on this check sheet is input as a binary image pattern and stored in advance based on typical diagnostic data of each disease category. The category of the disease name is recognized by matching with the binary mask pattern that has been set. In this way, the recognition performance of the disease name category can be improved by converting the recognition target into a binary image and then recognizing it as in the above-described embodiment.

[0079]

As described above, according to the present invention,
By identifying, rejecting, or recognizing the input image by comparing the input image pattern with the binary mask pattern stored in advance in the standard dictionary, the OCR user sequentially re-inspects the erroneous recognition and reject pattern by visual inspection. However, it is not necessary to re-examine the identified images in the correction work, which is advantageous in reducing the work load.
In the conventional character recognition method, one element of the recognition feature is often stored with a depth of n (n> 1) bits. However, in the present invention, each element of the binary mask pattern can be represented by one bit. The storage capacity can be reduced to 1 / n. Further, by using the known image encoding method together, the storage capacity of the binary mask pattern can be further reduced, the processing efficiency is high, and the dictionary capacity is compact.

In addition, according to any one of the first method, the second method, and the third method described in the section of means for solving the problem, the characteristics relating to the character line structure such as the direction of the character line and the connection relation. Since there is no need to extract the amount, there is an advantage that the feature extraction process is simple and the recognition performance is not easily affected even if these character structures are deformed due to blurring, crushing, noise in the character background, etc. . Not only faintness, crushing, and noise in the character background, but also using a binary mask pattern created from an image pattern composed of the character part as black dots and the background as white dots, the character part is configured as white dots and the background as black dots. There is an advantage of having a high recognition performance even with respect to so-called white characters, such as fading, crushing, and background noise.

According to the first method, the input image pattern is compared with the binary mask pattern stored in advance in the standard dictionary to identify, reject or recognize the input image. In the work of successively re-inspecting and correcting misrecognition and reject patterns,
There is also an advantage that it is not necessary to re-examine the recognized image, and the work load is reduced.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of an image pattern identification / recognition processing device according to the present invention.

FIG. 2 is a flowchart of an example of identification / rejection / recognition determination processing according to the present invention.

FIG. 3 is a diagram showing a pattern example for explaining a specific determination process of identification / rejection / recognition.

[Explanation of symbols]

 1 binary image pattern storage circuit 2 normalization processing circuit 3 binary mask pattern storage circuit 4 recognition determination circuit 5 binary mask pattern creation circuit 6 image input / output device

Claims (9)

[Claims] 1. For an input binary image pattern,
An image pattern identification / recognition method for determining a recognition category to which a binary image pattern belongs by collating with a binary mask pattern for each previously created recognition category, wherein a black part of the input binary image pattern is The black part of the input binary image pattern included in the black part of the binary mask pattern of the first recognition category is 2 of the first recognition category.
The input binary image pattern belongs to the first recognition category if the input binary image pattern is not included only in the overlapping portion of the black part of the value mask pattern and the black part of the binary mask pattern of any other second recognition category. An image pattern identification / recognition method having a first procedure of identifying 2. The input binary image pattern does not belong to any recognition category according to the first procedure, and the black part of the input binary image pattern is a binary mask of the first recognition category. A second step of rejecting the input binary image pattern when it is included only in an overlapping portion of a black portion of a pattern and a black portion of a binary mask pattern of the second recognition category that is different from any one of the other; The image pattern identification / recognition method according to claim 1. 3. With respect to the input binary image pattern,
An image pattern identification / recognition method for determining a recognition category to which a binary image pattern belongs by collating with a binary mask pattern for each previously created recognition category, wherein a white part of the input binary image pattern is The white part of the input binary image pattern included in the white part of the binary mask pattern of the first recognition category is 2 of the first recognition category.
The input binary image pattern is not included only in the overlapping part of the white part of the value mask pattern and the white part of the binary mask pattern of any other second recognition category, and the input binary image pattern is the first recognition category. An image pattern identification / recognition method characterized by having a third step of identifying that the image pattern belongs to. 4. The input binary image pattern does not belong to any recognition category according to the third procedure, and the black portion of the input binary image pattern is a binary mask of the first recognition category. A fourth step of rejecting the input binary image pattern when it is included only in the overlapping part of the white part of the pattern and the white part of the binary mask pattern of the second recognition category of any other The image pattern identification / recognition method according to claim 3, characterized in that 5. With respect to the input binary image pattern,
An image pattern identification / recognition method for determining a recognition category to which the binary image pattern belongs by collating with a binary mask pattern for each previously created recognition category, wherein the binary value is used at a plurality of predetermined positions in the pattern. A first degree of matching based on the number of black pixels in both the image pattern and the binary mask pattern, and the binary image pattern and the binary mask pattern both become white pixels at the plurality of predetermined positions. A second degree of coincidence based on a number, and a first degree of disagreement based on the number of white pixels in the binary image pattern and black pixels in the binary mask pattern at the plurality of predetermined positions. , 2 at the plurality of predetermined positions
A second degree of disagreement is obtained based on the number in which the value image pattern becomes a black pixel and the binary mask pattern becomes a white pixel, and a second degree of disagreement is obtained based on the first degree of coincidence and the second degree of coincidence. The third degree of coincidence is obtained, the third degree of disagreement is obtained based on the first degree of disagreement and the second degree of disagreement, and based on the third degree of coincidence and the third degree of disagreement. It has a fifth procedure for obtaining a similarity between the binary image pattern and the binary mask pattern, and a sixth procedure for recognizing that the binary image pattern belongs to a recognition category having the maximum similarity. Image pattern identification and recognition method. 6. The image pattern identification / recognition method according to claim 5, wherein all positions in the pattern are used as a plurality of predetermined positions in the pattern. 7. The input binary image pattern,
An image pattern identification / recognition method for determining a recognition category to which the binary image pattern belongs by collating with a previously created binary mask pattern for each recognition category. A seventh procedure for setting a threshold value for each pattern, and a first procedure based on the number of black pixels in both the binary image pattern and the binary mask pattern at a plurality of predetermined positions in the pattern. The degree of coincidence, a second degree of coincidence based on the number of white pixels in both the binary image pattern and the binary mask pattern at the plurality of predetermined positions, and the binary image at the plurality of predetermined positions The first degree of disagreement based on the number of patterns that become white pixels and the binary mask pattern that becomes black pixels,
A second degree of disagreement is obtained based on the number in which the value image pattern becomes a black pixel and the binary mask pattern becomes a white pixel, and a second degree of disagreement is obtained based on the first degree of coincidence and the second degree of coincidence. The third degree of coincidence is obtained, the third degree of disagreement is obtained based on the first degree of disagreement and the second degree of disagreement, and based on the third degree of coincidence and the third degree of disagreement. A fifth step of obtaining the degree of similarity between the binary image pattern and the binary mask pattern; and the fifth step of comparing the binary image pattern and the binary mask pattern of the recognition category with respect to only one recognition category. Image pattern identification, which comprises an eighth step of identifying the binary image pattern in the recognition category when the degree of similarity in the fifth step is equal to or greater than the threshold value of the binary mask pattern. -Recognition method. 8. The input binary image pattern does not belong to any recognition category according to the eighth procedure, and the binary image pattern and its recognition category are classified into two categories for two or more recognition categories. A ninth step of rejecting the input binary image pattern when the similarity to the value mask pattern in the fifth step is equal to or more than the threshold value of the binary mask pattern. The image pattern identification / recognition method according to claim 7. 9. A tenth procedure for newly creating, adding, or updating a binary mask pattern by iterative learning using a predetermined similarity based on information on a black part or a white part of the input binary image pattern. 1 to 8 are included.
The image pattern identification / recognition method according to any one of 1.