JP2797721B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition apparatus, and more particularly to a character recognition apparatus for performing character recognition by using at least two types of character classification means.

[0002]

2. Description of the Related Art In a conventional character recognition method based on a pattern matching method, a similarity between an unknown character to be recognized and a standard character stored in a recognition dictionary is calculated, and the maximum similarity is determined in advance for the standard character. If the difference between the maximum similarity and the next highest similarity is equal to or greater than another predetermined threshold, the unknown character has the maximum similarity. The recognition result of the same character type as the indicated standard character is output. It is known that by setting these two thresholds to appropriate values for each character type, it is possible to suppress the number of misread unknown characters and also reduce the number of unknown characters that cannot be recognized. Yes (Japanese Patent Publication No. 58-178484)
Reference).

In order to improve the recognition rate, it has been considered to use two types of character recognition methods in combination. In this method, the same type of threshold is used, for example, "free handwritten character recognition method". A method of decision theory in 1987 is known.
0). FIG. 6 shows an outline of the determination algorithm in this method. According to this, the pattern matching of characters is arranged in steps 201 and 202, the structural analysis of characters is arranged in steps 204 and 205, and the first threshold value for determining whether or not a recognition result is obtained only by the pattern matching method is set. In step 203, the similarity between the unknown character and the standard character stored in advance is calculated, and if the maximum similarity is equal to or greater than the threshold value previously set for the standard character, the unknown character has the maximum similarity. Steps 203 and 210 for outputting a recognition result of the same character type as the indicated standard character are executed.

If the threshold value is less than the threshold value, it is imperfect that the pattern matching method alone is incomplete. In step 204, feature extraction by structural analysis is performed on the unknown character.
Perform a comparison with the standard. In step 207, it is determined using the second threshold value whether or not the evaluation result based on the evaluation value calculated for the first candidate character code obtained at this time is used as the recognition result. Step 2 is to determine whether to rescue the rejected characters as a recognition result
For example, a total of three thresholds are set for each recognition target character type, such as making a determination using the third threshold in 09,
Recognition results that are considered stable in each determination are sequentially determined based on these thresholds, and those that do not pass the threshold processing are rejected in step 211.

[0005]

The above-described conventional character recognition has a problem in that, by performing threshold processing a plurality of times, misreading is reduced, but unknown characters to be rejected are increased and the overall recognition rate is reduced. is there. Further, in the above-described method of using two methods of pattern matching and structural analysis in combination to prevent a reduction in the recognition rate, three thresholds must be set for each number of character types. As the number increases, there is a problem that the setting of these thresholds becomes complicated, and there is a problem that the number of determination processes increases and the processing amount increases.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems and to solve the problem
The evaluation value is calculated for each of the first candidate character codes of the classification results output by the type and three or more types of classification means, and a comparison is made between the evaluation values and a threshold determination process is performed once for each candidate character code. At the same time as the rejection determination, even if the first candidate of the classification result of the two or three or more classification means does not match, if any of the candidates is correct, the candidate is adopted as the final recognition result. It is an object of the present invention to provide a character recognition device having an evaluation method as described above and having an effect of simultaneously reducing the number of misread and rejected unknown characters.

[0007]

SUMMARY OF THE INVENTION A character recognition apparatus according to the present invention comprises two types of standard patterns for character type K created from a learning pattern and two types of character recognition by calculating the distance between an input unknown character pattern. First and second classifying means for performing character classification of a first candidate character code Ci most similar to a standard pattern as a result of classification by the first classifying means, and a distance value d1 (i) of the first candidate character code; Character code C of the first candidate of the classification result by the second classification means
j and its distance value d1 (j) are input and the character code Ci is input.
When the character codes Ci and Cj match, the character code is used as the final recognition result. When the character codes Ci and Cj are different, the distance value d1 between the first classifying unit and the second classifying unit is used.
A main axis representing the inclination of the distribution of the two-dimensional vector Ck (d1 (k), d2 (k)) of the character code Ck of the recognition candidate on a plane having (k) and d2 (k) as the X axis and the Y axis, respectively. The eigenvectors Vk (p1 (k), p2 () corresponding to the maximum eigenvalues of the two-dimensional variance-covariance matrix formed by using the distance vectors d1 (k) and d2 (k) as variables using the principal component analysis method with the direction vectors. k)),
Further, the main axis direction vector and the two-dimensional vector Ck
(D1 (k), d2 (k)), and an evaluation value Ek as a scale for evaluating the character code Ck as a recognition result is an evaluation value Ei for the character code Ci and an evaluation value for the character code Cj. Ej, and the evaluation value E is determined based on a threshold value and a mutual magnitude comparison.
a determination unit configured to determine whether to select any of the character codes corresponding to i and Ej as the final recognition result or to reject both of them.

The character recognition apparatus of the present invention calculates two types of character classification for character recognition by calculating a distance between two types of standard patterns related to the character type K created from a learning pattern and an input unknown character pattern. First and second classifying means, and a first candidate character code C most similar to a standard pattern as a result of classification by the first classifying means.
When i and its distance value d1 (i) and the first candidate character code Cj of the classification result by the second classification means and its distance value d1 (j) are input and the character codes Ci and Cj match, The character code is used as the final recognition result. If the character codes Ci and Cj are different, the distance values d1 (k) and d2 of the first and second classifying means are used.
A two-dimensional vector Ck (d1 of a character code Ck of a recognition candidate on a plane having (k) as an X axis and a Y axis, respectively.
(K), d2 (k)) using the principal component analysis method to calculate the distance values d1 (k) and d2 (k)
Are the eigenvectors Vk (p1 (k), p2) corresponding to the maximum eigenvalues of the two-dimensional correlation matrix formed by using
(K)), and an evaluation value which is obtained based on the main axis direction vector and the two-dimensional vector Ck (d1 (k), d2 (k)) and is a scale for evaluating the character code Ck as a recognition result. Ek is determined by selecting one of the evaluation value Ei for the character code Ci and the evaluation value Ej for the character code Cj. In the selection, the evaluation value Ei and the evaluation value Ei A determination means is provided for determining whether any character code corresponding to Ej is selected as the final recognition result or whether any character code is rejected.

In addition, the apparatus of the present invention calculates three or more types of all n types of characters for character recognition by calculating a distance between two types of standard patterns relating to the character type K created from the learning pattern and the input unknown character pattern. First to n-th classifying means for classifying, and the distance value dm (h) of the first candidate character code Ch of the classification result of the m-th (m <n) classifying means and n other than the m-th classifying means. -1 classifying means, the n-dimensional vector Ch (d1 (h), d
2 (h),... Dm (h),... Dn (h)) for the first candidate of all the classifiers. When calculating the evaluation value as a scale to evaluate the recognition result for each of the candidate character codes, if the character code of the first candidate in the classification result of all the classification means is the same, the character code is recognized as a minimum. As a result, if they are not the same, the distance values d1 (k), d2 (k),.
(K),..., Dn (k) are coordinate axes, and an n-dimensional vector Ck (d1 (k), d2
(K),..., Dm (k),..., Dn (k)), using principal component analysis, the respective elements d1 (k), d2 (k),. dm (k),
, Dn (k) as a variable, the eigenvector Vk (p
1 (k), p2 (k), ..., pm (k), ..., pn
(K)), and further, the main axis direction vector and the n-dimensional vector Ck (d1 (k), d2 (k),.
dm (k),..., dn (k)), and evaluates the character code Ck as a recognition result as an evaluation value Ek by the n number of first candidate characters by the n classifying means. The character code corresponding to any of the n evaluation values is determined based on a threshold value and a magnitude comparison between the character codes. A character recognition device comprising: a determination unit that determines whether to select as a result or to reject any of them.

In addition, the character recognition apparatus of the present invention calculates three or more types of character recognition by calculating the distance between two types of standard patterns related to the character type K created from the learning pattern and the input unknown character pattern. Distance value dm of the first candidate character code Ch as a classification result of the first to n-th classification means for performing the character classification of (m) (m <n).
(H) and an n-dimensional vector Ch (d1) in which these distance values are arranged in the order of the classifying means with reference to the distance values for the character code Ch in the (n-1) th classifying means other than the m-th classifying means.
(H), d2 (h),... Dm (h),... Dn (h)) for the first candidate of all the classifying means, and classifying all n classifying means from the n-dimensional vector. When calculating the evaluation value as a scale to evaluate the recognition result for each of the first candidate character codes of the result, if the character code of the first candidate of the classification result of all classification means is the same, the character If the codes are not the same, the distance values d1 (k), d2 (k),
, Dm (k),..., Dn (k) as coordinate axes, an n-dimensional vector Ck (d1
(K), d2 (k), ..., dm (k), ..., dn
Using the principal component analysis method, the principal axis direction vector of the distribution (k)) is used to calculate each element d1 (k), d2 of the n-dimensional vector.
, Dm (k),..., Dn (k) are eigenvectors corresponding to the maximum eigenvalues of an n-dimensional correlation matrix formed as variables. (P1 (k), p2 (k),.
(K),..., Pn (k)), and further, the main axis direction vector and the n-dimensional vector Ck (d1 (k),
d2 (k),..., dm (k),..., dn (k)), and the evaluation value Ek as a scale for evaluating the character code Ck as a recognition result is calculated by n classification means. The selection is made by selecting any one of the n evaluation values for the character codes of the first candidates, and in the selection, any one of the n evaluation values is determined based on a threshold value and a mutual magnitude comparison. And a determination unit for determining whether to select a character code corresponding to or as a minimum recognition result or to reject any of them.

[0011]

The operation of the present invention will be described with reference to FIG. FIG. 5 shows an example in which two types of classification means are used together.

An evaluation value Ei for the first candidate character code Ci of the classification result of the first classification means and an evaluation value Ei for the first candidate character code Cj of the classification result of the second classification means.
When calculating j, a character pattern for learning belonging to the character type k is placed on a two-dimensional plane whose axis is the distance value between the first classification means and the second classification means for each character type k to be classified. When the distance values d1 (k) and d2 (k) of the first classifier and the second classifier obtained for Ck are plotted, they are distributed as shown in a closed curve S in FIG. The main axis Vk (p1 (k), p2 (k)) representing the gradient of this distribution is as shown in FIG. 5, and is obtained by a known principal component analysis method.

As shown in FIG. 5, a straight line L perpendicular to the main axis Vk and tangent to the distribution indicated by the curve S is shown in FIG.
Set. When the two-dimensional vector Ck (d1 (k), d2 (k)) of the candidate character code Ck is located closer to the origin, the character code Ck belongs to the character type k. One having a small distance value as indicated by a mark is also determined to belong to the character type k. In addition, the two-dimensional vector C ^′ k (d1 ^′ (k), d2 ^′ (k)) corresponding to the candidate character code C _′ k indicated by a triangle in FIG. 5 is outside the shaded area and belongs to the character type k. Reject without judgment.

In this example, Ekmax is the main axis Vk with respect to the intersection of the aforementioned straight line L and the main axis Vk obtained earlier.
The above value is used as a determination threshold value for determining whether or not the candidate character code Ck belongs to the character type k.

On the other hand, consider a conventional case in which the first classifying means and the second classifying means independently set thresholds, respectively, and a threshold for the first classifying means and the second classifying means at that time. The values are t1 and t2, respectively. Assuming that t1 and t2 are the maximum values of the distance values with respect to the learning pattern, the area that does not exceed either of them is a rectangular area surrounded by the dashed line and the two-dot chain line in FIG. Is set, and the recognition rate decreases.

As is clear from this comparison, according to the present invention, the area which is not rejected can be widened, and the character code of the first candidate of one of the classification methods outputs a very close distance value. If is less than or equal to the threshold value, it is possible to determine the character type.

The first candidate character code C of the first classifying means
In the means for determining which one of the character codes Cj and i is to be output or rejected, the character codes Ci and Cj of the first candidate of the respective classification means Evaluation values Ei, E calculated for
If both j exceed the threshold values Eimax and Ejmax, the character codes Ci and Ci of the first candidates output by the first classifying unit and the second classifying unit are rejected. In other cases, since the smaller evaluation value is closer to the origin, when the smaller evaluation value does not exceed the judgment threshold value for the character type, the first candidate for the evaluation value is determined. The character code is used as the final recognition result. Otherwise, the first candidate character codes Ci and Cj output by the first classification means and the second classification means are rejected. The same effect is obtained when three kinds of classification means are used together.

[0018]

Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention, FIG. 2 is a flowchart showing the operation of the first embodiment of FIG. 1, and FIG. 3 is a block diagram of the second embodiment of the present invention. FIG. 4 is a block diagram showing the configuration, and FIG. 4 is a flowchart showing the operation of the second embodiment of FIG.

In the first embodiment shown in FIG. 1, a character pattern input unit 1 for inputting a character pattern, a first classifying unit 14 constituting a first classifying unit, and a second classifying unit constituting a second classifying unit are provided. A second classification unit 15, a determination unit 1 for determining a character pattern based on the classification results of the first classification unit 14 and the second classification unit 15;
6 and a final recognition result output unit 13 for outputting a final recognition result
And a control unit 17 for controlling the operation of each unit.

The character pattern input unit 1 is composed of a normal image scanner that takes in characters on a form into a recognition system.
The captured character image is supplied to the first classification unit 14 and the second classification unit 15 and is classified by two types of means. If the character codes of the first candidates of the classification results are the same as the classification results of the first classification unit 14 and the second classification unit 15, the determination unit 16 sets the final recognition result as the final recognition result output unit. 13 is transferred.

If the first candidate character code of each classification result is different, the first candidate character code of the first classification unit 14 and the second classification unit 15 is adopted as the final recognition result. Or reject the first candidate character code of each classification result. The control unit 17 controls the character pattern input unit 1, the first classifying unit 1
4, operation control of the second classification unit 15, the pair determination unit 16, and the final recognition result output unit 13.

The first classification unit 14 includes a first character recognition unit 2, a first classification dictionary memory 3 referred to by the first character recognition unit 2, and a first classification result storage memory for storing classification results output from the first character recognition unit 2. The second classification unit 15 includes a second character recognition unit 5, a second classification dictionary memory 6 referred to by the second character recognition unit 5, and a second classification that stores a classification result output from the second character recognition unit 5. The result storage memory 7 is provided.

The first classifying dictionary memory 3 and the second classifying dictionary memory 6 store in advance the feature amounts forming the different standard patterns handled by the first character recognition unit 2 and the second character recognition unit 5 for learning. Created from patterns and stored. The first classifying unit 14 and the second classifying unit 15 calculate the distance between the standard and unknown character patterns for the classification target character type k stored in the respective classification dictionary memories, and match them in ascending order of distance. Are stored in the first classification result storage memory 4 and the second classification result storage memory 7 together with the corresponding character code.

When the processing of the first classifying unit 14 and the second classifying unit 15 is completed, the judging unit 16 is started by the control unit 17 and stores the first candidate of the first classifying unit 14 from the first classification result storage memory 4. And a distance value corresponding to the character code of the first candidate in the second classification unit 15 from the second classification result storage memory 7 in the two-dimensional vector format. 10 is stored. At the same time, the character code and the distance value of the first candidate of the second classification unit 15 are extracted from the second classification result storage memory 7 of the second classification unit 15, and are extracted from the first classification result storage memory 4.
4 is an integrated result storage register 1 in a two-dimensional vector format to which a distance value corresponding to the character code of the first candidate is added.
Store to 0. At this time, the first classification unit 14, the second classification unit 1
If the first candidate character code of each of the classification results is the same, the character code is output to the final recognition result output unit 13 as the final recognition result.

When the classification results of the first classification unit 14 and the second classification unit 15 are different after the processing of the classification result integration unit 8 is completed, the evaluation value calculation unit 9 is started by the control unit 17 and stores the integration result. Referring to the register 10 and the dictionary memory 11 for calculating an evaluation value, the evaluation values Ei and Ej for the candidate character codes output by the first classification unit 14 and the second classification unit 15 are respectively expressed by, for example, the following equations (1) and (2). Is calculated by Ei = d1 (i) P1 + d2 (i) P2 (i) (1) Ej = d1 (j) P1 + d2 (j) P2 (j) (2) The variance-covariance matrix is stored in the evaluation value calculation dictionary memory 11. Seed axis Vk (p1 (k), p2 (k)) obtained from
And a threshold value Ekmax for determination are stored for each character type k.
Here, the calculation of the main axis Vk for each character type k uses a two-dimensional variance-covariance matrix when the distances (d1 (k), d2 (k)) are variables, but the first classification unit 14, If the scales of the distance values output by the second classification unit 15 are significantly different, a correlation matrix is used instead. Accordingly, when calculating the evaluation value, it is possible to eliminate the effect that the evaluation value of the classification means having the larger variance of the distance value is weighted, and the distance value output by the first classification unit 14 and the second classification unit 15 can be eliminated. Can be evaluated equally by the threshold value determination unit 12.

The threshold value judging unit 12 is started by the control unit 17 when the processing of the evaluation value calculation unit 9 is completed, and the first classification unit 14 and the second classification unit 15 calculated by the evaluation value calculation unit 9 output The comparison of the evaluation value for the candidate character code is performed by comparing the evaluation value with the threshold value stored in the evaluation value calculation dictionary memory 11, and it is determined which classification result is to be adopted or both classification results are rejected. I do. An example of the algorithm will be described with reference to FIG.

If the character candidates extracted in steps 101 and 103 are the same as the character codes extracted in steps 102 and 104, step 105 is executed in which the candidate character code is adopted as the final recognition result in step 105. . This processing is performed by the classification result integration unit 8.

The threshold value judging section 12 executes the processing from step 105 onward. Evaluation values Ei, for candidate character codes output by the two classification units obtained by the evaluation value calculation unit 9
If both Ej exceed the threshold value of the evaluation value calculation dictionary memory 11, a step 111 for rejecting both candidate character codes is executed. Otherwise, the character code of the candidate that does not exceed the threshold value is selected, and if the evaluation value is equal to or smaller than the other evaluation value in step 109, the character code of the corresponding candidate is selected as the final recognition result. And sends the character code to the final recognition result output unit 13; otherwise, rejects the corresponding candidate character code and sends a preset rejection code corresponding to the rejection to the final recognition result output unit 13. send.

The final recognition result output unit 13 outputs the final recognition result output from the determination unit 16 as a character code in response to an instruction from the control unit 17.

Next, a second embodiment of the present invention will be described with reference to FIG. The difference between the second embodiment of FIG. 3 and the first embodiment of FIG.
5 outputs the same candidate character code, it is immediately used as the final recognition result.
In the first embodiment, the threshold stabilizing process is performed by the determination unit 16 in the same manner as in the first embodiment in which the first classification unit 14 and the second classification unit 15 output character codes of different candidates. Therefore, the pair determination processing avoidance route R1 set in FIG. 1 is deleted.

FIG. 4 shows a determination algorithm in the pair determination section 16 in the second embodiment. The determination unit 16 performs the same evaluation up to the point at which the determination process of step 105 is performed, but also performs an evaluation value calculation when the character codes of the candidates of the first classification unit 14 and the second classification unit 15 are the same. ,
If the result of step 112 is equal to or less than the threshold value set in step 113, the result is adopted as the final recognition result, and the character code is sent to the final recognition result output unit 13,
Otherwise, a step 111 for rejecting the character code of the corresponding candidate is executed, and a preset rejection code corresponding to the rejection is sent to the final recognition result output unit 13.

In this way, even if the candidate character codes output by the first classification unit 14 and the second classification unit 15 are the same,
If both of the distance values are larger than those calculated using the learning pattern, the distance values are rejected, so that the first classifying unit 14 and the second classifying unit 15 may misread at the same time. Therefore, it is possible to suppress a decrease in the recognition rate to be equal to or less than the first embodiment.

Although two embodiments have been described for the case where two types of classification means according to the present invention are used together, the distance scale calculated by the first classification unit 14 and the second classification unit 15 is used in character recognition. It is not limited to those usually used, and any distance scale or similarity scale can be used, and a mixture of them is also possible. Further, even when three or more kinds of classification means are used together,
It can be configured according to one embodiment. Note that the above 2
In one embodiment, an example in which hardware is used as a basis has been described, but each of these functions may be realized by any means of hardware or software.

[0035]

As described above, according to the present invention, the number of erroneously read characters is reduced by using at least two types of character classifying means, and at the same time, the candidate character codes output by each character classifying means are reduced. By performing rejection determination by comparing the calculated evaluation values with each other and performing one rejection determination process as compared with the case where rejection determination thresholds are individually set for two types of character classification means. The number of characters to be rejected can be greatly reduced, and the recognition system can be significantly improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the first embodiment of FIG.

FIG. 3 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

FIG. 4 is a flowchart showing the operation of the second embodiment of FIG. 3;

FIG. 5 is an explanatory diagram of the operation of the present invention.

FIG. 6 is a flowchart showing the operation of a conventional character recognition device.

[Explanation of symbols]

 Reference Signs List 1 Character pattern input unit 2 First character recognition unit 3 First classification dictionary memory 4 First classification result storage memory 5 Second character recognition unit 6 Second classification dictionary memory 7 Second classification result storage memory 8 Classification result integration unit 9 Evaluation value Calculation unit 10 Integrated result storage register 11 Evaluation value calculation dictionary memory 12 Threshold value judgment unit 13 Final recognition result output unit 14 First classification unit 15 Second classification unit 16 Judgment unit 17 Control unit

Continuation of the front page (56) References JP-A-63-263588 (JP, A) JP-A-57-25082 (JP, A) JP-A-63-79191 (JP, A) JP-A-62-74188 (JP, A) , A) Proceedings of the IEICE National Convention, VOL. 1991, NO. SPRING PT 7 PAGE. 7-244 IEICE Technical Report, VOL. 89, NO. 436 (PRU89 118-125) PAGE. 15-22 (58) Field surveyed (Int. Cl. ⁶ , DB name) G06K 9/62 G06K 9/03

Claims (4)

(57) [Claims] 1. A first and second classification for performing two types of character classification for character recognition by calculating a distance between two types of standard patterns related to the character type K created from a learning pattern and an input unknown character pattern. Means, the first candidate character code Ci most similar to the standard pattern and the distance value d1 (i) of the classification result by the first classification means, and the first candidate of the classification result by the second classification means. A character code Cj and its distance value d1 (j) are input. If the character codes Ci and Cj match, the character code is used as a final recognition result. If the character codes Ci and Cj are different, the first code is used. A two-dimensional vector Ck (d of the character codes Ck of the recognition candidates on a plane having the distance values d1 (k) and d2 (k) of the classification means and the second classification means as the X axis and the Y axis, respectively. 1 (k), d2 (k)) is represented by a principal axis direction vector representing the gradient of the distribution using a principal component analysis method, and the two-dimensional variance covariance formed by using the distance values d1 (k) and d2 (k) as variables. An eigenvector Vk (p1 (k), p2 (k)) corresponding to the maximum eigenvalue of the variance matrix is obtained, and further obtained based on the main axis direction vector and the two-dimensional vector Ck (d1 (k), d2 (k)). And an evaluation value Ek as a scale for evaluating the character code Ck as a recognition result is an evaluation value Ei for the character code Ci.
And any one of the evaluation values Ej for the character code Cj. In the selection, any one of the character codes corresponding to the evaluation values Ei and Ej is determined based on a threshold value and a mutual magnitude comparison. A character recognition device comprising: a determination unit that determines whether to select as a final recognition result or to reject any of them. 2. A first and second classification for performing two types of character classification for character recognition by calculating a distance between two types of standard patterns related to the character type K created from a learning pattern and an input unknown character pattern. Means, the first candidate character code Ci most similar to the standard pattern and the distance value d1 (i) of the classification result by the first classification means, and the first candidate of the classification result by the second classification means. A character code Cj and its distance value d1 (j) are input. If the character codes Ci and Cj match, the character code is used as a final recognition result. If the character codes Ci and Cj are different, the first code is used. A two-dimensional vector Ck (d of the character codes Ck of the recognition candidates on a plane having the distance values d1 (k) and d2 (k) of the classification means and the second classification means as the X axis and the Y axis, respectively. 1 (k), d2 (k)) is a two-dimensional correlation matrix formed by using principal component analysis as a principal axis direction vector representing the gradient of the distribution and using the distance values d1 (k) and d2 (k) as variables. As the eigenvector Vk (p1 (k), p2 (k)) corresponding to the maximum eigenvalue of the main axis direction vector and the two-dimensional vector Ck
(D1 (k), d2 (k)), and an evaluation value Ek as a scale for evaluating the character code Ck as a recognition result is an evaluation value Ei for the character code Ci and an evaluation value for the character code Cj. Ej, and the evaluation value E is determined based on a threshold value and a mutual magnitude comparison.
A character recognition device comprising: a determination unit that determines whether any one of the character codes corresponding to i and Ej is selected as a final recognition result or both are rejected. 3. All three or more types for character recognition by calculating a distance between two types of standard patterns related to the character type K created from a learning pattern and an input unknown character pattern.
First to n-th classifying means for classifying the types of characters, and the distance value dm (h) of the first candidate character code Ch of the classification result of the m-th (m <n) classifying means and the m-th classifying means Other than n
−1, the n-dimensional vectors Ch (d1 (h), d2 (h),.
(H),... Dn (h)) are generated for the first candidates of all the classifying means, and the n-dimensional vectors are used as the first candidate character codes of the classification results of all the n classifying means. When calculating the evaluation value as a scale to be evaluated as the recognition result, if the character code of the first candidate of the classification result of all classification means is the same, the character code is regarded as the minimum recognition result, and if not, Are the distance values d1 (k), d2 (k),..., Dm (k),.
An n-dimensional vector Ck (d1 (k), d2 (k),..., Dm) in an n-dimensional space when (k) is a coordinate axis
(K),..., Dn (k)), the principal axis direction vectors representing the gradients of the distributions are calculated using principal component analysis, and the respective elements d1 (k), d2 (k),. ),
, Dn (k) as a variable, the eigenvector Vk (p
1 (k), p2 (k), ..., pm (k), ..., pn
(K)), and further, the main axis direction vector and the n-dimensional vector Ck (d1 (k), d2 (k),.
dm (k),..., dn (k)), and evaluates the character code Ck as a recognition result as an evaluation value Ek by the n number of first candidate characters by the n classifying means. The character code corresponding to any of the n evaluation values is determined based on a threshold value and a magnitude comparison between the character codes. A character recognition device comprising: a determination unit that determines whether to select as a result or to reject any of them. 4. All three or more types for character recognition by calculating the distance between two types of standard patterns related to the character type K created from a learning pattern and an input unknown character pattern.
First to n-th classifying means for classifying the types of characters, and the distance value dm (h) of the first candidate character code Ch of the classification result of the m-th (m <n) classifying means and the m-th classifying means Other than n
−1, the n-dimensional vectors Ch (d1 (h), d2 (h),.
(H),... Dn (h)) are generated for the first candidates of all the classifying means, and the n-dimensional vectors are used as the first candidate character codes of the classification results of all the n classifying means. When calculating the evaluation value as a scale to be evaluated as the recognition result, if the character code of the first candidate of the classification result of all the classification means is the same, the character code is used as the final recognition result, and if not, Are the distance values d1 (k), d2 (k),..., Dm (k),.
An n-dimensional vector Ck (d1 (k), d2 (k),..., Dm) in an n-dimensional space when (k) is a coordinate axis
(K),..., Dn (k)), the principal axis direction vectors representing the gradients of the distributions are calculated using principal component analysis, and the respective elements d1 (k), d2 (k),. ),
.., Dn (k) as a variable, an eigenvector Vk (p1) corresponding to the maximum eigenvalue of an n-dimensional correlation matrix.
(K), p2 (k), ..., pm (k), ..., pn
(K)), and further, the main axis direction vector and the n-dimensional vector Ck (d1 (k), d2 (k),.
dm (k),..., dn (k)), and evaluates the character code Ck as a recognition result as an evaluation value Ek by the n number of first candidate characters by the n classifying means. The character code corresponding to any one of the n evaluation values is determined based on a threshold value and a magnitude comparison between each other. A character recognition device comprising: a determination unit that determines whether to select as a minimum recognition result or to reject any of them.