JPH09204490A - Method for selecting character string input means - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a highly efficient character string input means selecting method for freely incorporating various estimation algorithms without restriction, for improving the reliability of the estimation of character recognition precision and for inputting a character string with maximum efficiency for the quality of various input characters. SOLUTION: Image data 113-117 of the character strings are inputted to plural character recognition precision estimation processing blocks 101-105 and plural character recognition precision estimation processing results 107-111 are derived. The estimation processing results 107-111 are inputted to a character recognition estimation processing result derivation processing block 106 and a single character recognition precision estimation result is derived. In the derived processing block 106, an operator where an operation rule is made to correspond to the derivation logic of the estimation processing block is defined and a rule for designating ID which is previously allocated for the individual character recognition precision estimation result and a processing is provided as the factor of the operator. Then, a derivation processing is executed based on an operation expression mentioned based on the rule.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a processing technique for inputting a Japanese character string or the like into a computer system or the like using a character recognition technique, and more particularly, to an arbitrary number of character recognition accuracy estimation processes after performing the character recognition process. Can be freely incorporated or their recognition accuracy estimation processing can be performed in an arbitrary order to derive a plurality of recognition accuracy estimation results by an arbitrary method,
The present invention relates to a processing technique that enables selection of an efficient character string input means using the recognition accuracy estimation result.

[0002]

2. Description of the Related Art Character recognition technology is recognized as a useful technology for reducing the number of manual character input operations, and is gradually spreading as a practical technology due to many years of research and development. However, when targeting handwritten characters, it is difficult to perform highly accurate character recognition for all characters with various recognition accuracies (quality), from carefully written characters to inferior characters. If the character recognition accuracy (quality) is poor, there will be many recognition errors, resulting in a correction work that is less efficient than the manual character string input method using a keyboard or the like.

Therefore, the recognition accuracy of handwritten characters is estimated,
Solving the problem of recognition accuracy estimation, such as selecting an appropriate input means according to the recognition accuracy and efficiently inputting a character string, is an essential condition.

In the conventional recognition accuracy estimation, a method of independently executing a specific recognition accuracy estimation algorithm has been adopted.

[0005]

In the method for independently executing the character recognition accuracy estimation algorithm described above, when an estimation algorithm that depends on the recognition result, such as a rejection judgment process, is used, the recognition result is reflected almost as it is. Therefore, it may be unreliable. When an estimation algorithm that is independent of the recognition result is used, it is possible to output an estimation result that does not depend on the recognition result. However, if the estimation result and the recognition result differ greatly, which result should be used. Since the key is not clear, it was difficult to flexibly estimate the recognition accuracy.

The present invention has been made in view of the above circumstances, and an object thereof is to solve the above-mentioned problems in the conventional art and to rely on a recognition algorithm for character recognition estimation. A highly accurate character string input means selection method having a character recognition accuracy estimation result derivation method capable of deriving a highly reliable single character recognition estimation result by freely incorporating an independent estimation algorithm etc. into To provide.

[0007]

SUMMARY OF THE INVENTION The above objects of the present invention are as follows.
In the character string input means selection method for converting character image data into a character code and selecting one having a high input efficiency from a plurality of character string input means based on the result of character recognition accuracy estimation, regarding the input character image data Performing a plurality of character recognition accuracy estimation, deriving a single character recognition accuracy estimation result from the plurality of character recognition accuracy estimation results, based on the derived single character recognition estimation result And a step of selecting a character string input means having a high input efficiency from among the plurality of character string input means.

In the above character string input means selecting method,
The step of deriving a single character recognition accuracy estimation result from a plurality of character recognition accuracy estimation results defines an operator that corresponds the operation of the derivation process to an operation rule, and an arithmetic expression constructed using these operators. Based on the above, a processing method for deriving the single character recognition accuracy estimation result and a process for deciding the order of deriving the single character recognition accuracy estimation result have a single character recognition accuracy estimation from a plurality of character recognition accuracy estimation results. This is preferable because it facilitates the process of deriving the result.

Further, in the above-mentioned character string input means selecting method, a processing method for deriving a single character recognition accuracy estimation result and an deriving order are determined based on an arithmetic expression formed by using operators. The process is a process of deciding the order of deriving a single character recognition accuracy estimation result of the operator described by the user, a process of deciding the type of derivation process associated with the operator in advance, and a process of using ID number of the character recognition accuracy estimation result described by the operator as an argument of the operator,
Alternatively, an arithmetic expression formed by using an operator has a process of determining a character recognition accuracy estimation result or an estimation process to be processed based on an ID number of the character recognition accuracy estimation process. This is preferable because it facilitates the method of deriving a single character recognition accuracy estimation result based on the above, and the process of determining the deriving order.

In the method of deriving a character recognition accuracy estimation result according to the present invention, a plurality of character recognition accuracy estimations are performed on input character image data, and a single character recognition accuracy estimation is performed from the plurality of character recognition accuracy estimation results. The derivation method that derives the result is adopted. In order to facilitate such derivation processing, specifically, a plurality of logics prepared for deriving individual character recognition accuracy estimation processing results are associated with calculation rules for the respective derivation logics. A rule is defined in which an operator is defined, and an ID assigned in advance to each character recognition accuracy estimation result and processing is specified as an argument of the operator, and the derivation method is described by an arithmetic expression based on this rule. By doing so, the derivation process is performed. By adopting such a derivation method, it is possible to flexibly derive by freely incorporating the process (algorithm) for estimating the character recognition accuracy, and no matter what quality character type is included in the handwritten character, Even if the recognition accuracy of the input character type is biased, an optimized derivation method can be configured to improve the reliability of the character recognition accuracy estimation process,
A form input method or the like for selecting the most efficient character string input means depending on the character recognition accuracy is realized with high accuracy.

[0011]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram including a recognition accuracy estimation result derivation method according to an embodiment of the present invention. In FIG. 1, 101 to 105 are n blocks for performing character recognition accuracy estimation processing. These character recognition accuracy estimation processing blocks 101 to 105 perform processing for estimating whether or not the result of character recognition is correct.
It is possible to arbitrarily incorporate either a dependent estimation process using a character recognition result or a stand-alone estimation process using no recognition result. Further, if it is feasible, any number of different estimation processes may be prepared, and the character recognition rate, the correct character candidate inclusion rate, and other measures can be used as the recognition accuracy, and the number of types of estimation processing and The scale of recognition accuracy is
It is optional as long as the gist of the present invention is not impaired.

These character recognition accuracy estimation processing block 1
For 01 to 105, the recognition accuracy estimation process is performed based on the input of the result data of the preprocessing for the image data of the character strings represented by 113 to 117. These inputs 1
Numerals 13 to 117 are result data performed as a pre-process of the recognition accuracy estimation process such as the character recognition process and have no relation to the gist of the present invention. The estimation result data 107 to 111 estimated by the recognition accuracy estimation processing blocks 101 to 105 are output to the character recognition accuracy estimation result derivation processing block 106.

Character recognition accuracy estimation result derivation processing block 1
In 06, the character recognition accuracy estimation result derivation method description data 11 is stored in advance from the character recognition accuracy estimation result derivation method description file or the like.
2 is read, and the derivation process of the character recognition accuracy estimation result is performed using various derivation logics according to the description.
Finally, one character recognition accuracy estimation result is output from the plurality of character recognition accuracy estimation results.

The character recognition accuracy estimation result derivation processing block 106 can actually perform not only the derivation processing but also the individual character recognition accuracy estimation processing. An exemplary embodiment of such a designation method will be described with reference to FIG. Further, although the specification of the character recognition accuracy estimation result derivation method is performed using a file, a method of directly inputting from the keyboard may be used, and the means for inputting the derivation method specification does not impair the gist of the present invention. Is optional.

Next, the flow of processing of the character recognition accuracy estimation result derivation processing block 106 of the embodiment shown in FIG. 1 will be described with reference to FIG.

First, in processing block 201, FIG.
The input processing of reading the individual character recognition accuracy estimation result data indicated by 107 to 111 is performed. Next, in processing block 202, a description representing a derivation method rule is input from a character recognition accuracy estimation result derivation method description file or the like. Next, in processing block 203, processing block 202
The rules of the derivation method (deriving order, derivation processing method, etc.) are interpreted and determined from the description input in. Next, processing block 2
In 04, the individual character recognition accuracy estimation result is derived based on the designation of the derivation method of the processing block 203. In the derivation process in the processing block 204, not only the individual estimation result is derived, but also the character recognition accuracy estimation process which is not performed in the estimation process blocks 101 to 105 in FIG. 1 may be performed in some cases. Estimation processing block 101 of FIG.
It is also possible to re-execute the character recognition accuracy estimation processing performed in steps 105 to 105. Next, in processing block 205,
It is determined whether or not all the processing of the processing block 204 has been completed, and the processing is returned to the processing block 204 and repeated until it is completed. Finally, processing block 2
If it is determined at 05 that all the processing of the processing block 204 has ended, at processing block 206,
The character recognition accuracy derivation estimation result is output, and the processing ends.

FIG. 3 is an explanatory diagram of the block 204 for deriving the character recognition accuracy estimation result of the embodiment shown in FIG. In this figure, the process of inputting the rule of the derivation method from the character recognition accuracy estimation result derivation method description file or the like is omitted.

In FIG. 3, reference numerals 305 to 309 denote those in FIG.
Nos. 107 to 111 are the estimation results of the individual character recognition accuracy, in other words, the estimation results of whether the character recognition result is correct. Reference numerals 301 to 304 are blocks representing derivation processing. In the embodiment example shown in FIG. 3, four derivation processes will be described.

The derivation process 301 represents a majority vote. In other words, if the character recognition result is estimated to be correct among a plurality of estimation results that exceeds a separately specified threshold value, it is derived as the result that the character recognition result is estimated to be correctly recognized, and the result is exceeded. If it is not, it is a derivation method that derives it as the result of estimating that it is not correctly recognized. The threshold value as to how many results the character recognition result is estimated to be correct can be arbitrarily set within a range not changing the gist of the present invention.

The derivation process of 302 represents a process of limiting the estimation target and performing the estimation process. That is, the estimation target is limited according to the derivation (estimation) result output in the previous process, and the estimation process in the estimation process blocks 101 to 105 in FIG. 1 or another estimation process is performed. As described above in the description of the exemplary embodiments shown in FIGS. 1 and 2, in the processing block 204, not only the derivation process but also the individual character recognition accuracy estimation process can be actually performed. .

The derivation process of 303 represents a logical product. That is, only when it is estimated that all the character recognition results are correct among the plurality of estimation results, the result is estimated as the result of the correct recognition.

The derivation process of 304 represents a logical sum. That is, if even one of the plurality of estimation results is estimated to be correctly recognized, it is derived as the result of the estimation that it is correctly recognized.

In the above, processing blocks 301 to 304
Although the description has been limited to the four derivation methods (including the estimation process) described above, other derivation methods (including the estimation process) are also possible. This can be extended within the scope of the present invention.

Next, the description method of the above derivation will be described with an example.

For simplicity of explanation, 305 to 309
As the individual estimation result indicated by, whether or not the result of character recognition is correct is represented by a numerical value of 1 or 0. That is, in the estimation processing blocks 101 to 105 of FIG. 1, 1 is set as the estimation result when the character recognition result is estimated to be correct, and 0 is set as the estimation result when the character recognition result is estimated to be incorrect. And As the result of the character recognition processing, it is assumed that the correct character candidates are output up to the fifth place, and the estimation result is output up to the fifth place correspondingly. The left end is the first place and the right end is the fifth place. Therefore,
For example, "10000" or "011" for one character
A sequence of five 1s or 0s is associated with each other, such as "00". Therefore, as an example of the present embodiment, 30
The data format from 5 to 309 is "10000" or "0"
1100 ″ is used. As the method of expressing the estimation result, various expressions other than 1/0 can be used, and any expression can be used as long as the gist of the present invention is not impaired.

Here, a unique ID number is assigned to the estimation result output from the character recognition accuracy estimation processing block. That is, ID = 1 is added to the estimation result 305 output from the estimation processing block 101, and the estimation processing block 102
ID = 2 in the estimation result 306 of Eq.
ID = 3 in the estimation result 307 of 3 and the estimation processing block 1
The estimation result 308 of 04 is ID = 4, the estimation result 309 of the estimation processing block 105 is ID = 5, and so on. The ID numbers do not have to be assigned in order from 1 and may be symbols or alphabets other than numbers. It is arbitrary as long as it can distinguish the individual estimation processing results.

Rules representing that the derivation processing of the processing blocks 301 to 304 are performed are M, R, A and O, respectively.
Expressed by the operator symbol represented by. That is, M is a derivation process based on majority decision, R is a derivation process based on limitation of estimation target, and A is
Is a derivation process by a logical product, and O is a derivation process by a logical sum. Further, which character recognition accuracy estimation result is to be targeted is expressed by an argument surrounded by “[” and “]”, and the argument can be separated by commas and an arbitrary number can be designated. The argument can include the operator itself. For example, when deriving the estimation results 1, 2, and 3 by a majority decision, “M [1,
2, 3] ”, and when the estimation process 4 is performed only on the result of deducing the character recognition result out of the results obtained by logically deriving the estimation results 1, 2, and 3,“ R [O [ 1, 2,
3], 4] ”. When the operator R is specified, the argument is interpreted as representing the estimation process, not the estimation process result.

In the example embodiment shown in FIG. 3, ID = 6 is assigned to the estimation process of processing block 302. Also,
Due to the characteristics of the operator R, there are always two arguments. The derivation process shown in FIG. 3 is expressed as an arithmetic expression using the above-mentioned operators and arithmetic rules, as follows.

O [A [R [M [1,2,3], 6], 4], 5] That is, the estimation result ID = 1 is “10110”, 2 is “01100”, 3 is “10001”, 4 for "1110
When 1 ”and 5 are set to“ 11000 ”and the derivation process of FIG. 3 is sequentially written, M [1,2,3] →“ 10110 ”R [M [1,2,3], 6] →“ 10110 ” A [R [M [1,2,3], 6], 4] → “10100” Therefore, finally, O [A [R [M [1,2,3], 6]] , 4], 5] → “11100”.

Here, in the operator M, the number of "1" is 1 /
The estimation result was derived as "1" if it was 2 or more and "0" if it was less than 1/2. Further, in the operator R, an example in which the recognition result is estimated to be correct again as a result of performing the estimation process 6 by limiting the estimation target only when the derivation result output in the previous process is “1” showed that.

The method other than the derivation method shown in FIG. 3 can be incorporated, and the order is not limited to the example shown in FIG. Also, not all operators need to be used. Furthermore, it is possible to use the same operator multiple times. Based on these, as another example, the estimation results 1, 2, and 3 are derived by taking a majority vote, and as a result, the estimation process 8 is performed on the result of which the recognition result is estimated to be correct. An arithmetic expression derived by taking the three majority decisions of the result obtained by taking the logical sum of the results 4, 5, 6 and the estimated result 7 is expressed as follows.

M [R [M [1,2,3], 8], O
[4, 5, 6], 7] Further, an arithmetic expression derived by sequentially performing the processing of limiting the estimation target and performing the estimation processing 2 only when the first estimation result 1 is "1". Is represented as follows.

R [R [R [R [R] 1,2], 2], 2], 2] The character string input means is selected based on the single character recognition accuracy estimation result obtained by the above derivation method. An example of the method is shown. First, as an example of the character string input means, a character string input means assumed to directly input a character string using a keyboard or the like, a character string input means assumed to perform character recognition and match with a word dictionary, etc. , A character string input means that is supposed to perform character recognition and correct the recognition result, or a character string input means that uses these in combination (for example, a homophone input method). Draw a relative input efficiency curve for the character recognition accuracy for each of these character string input means, the character recognition accuracy value at the turning point of the input efficiency between each character string input means as a threshold, the above character recognition accuracy estimation result The most efficient character string input means is determined and selected.

[0035]

As described above in detail, according to the present invention, a plurality of character recognition accuracy estimations are performed on input character image data, and a single character recognition is performed from the plurality of character recognition accuracy estimation results. Since the accuracy estimation result is derived, an arbitrary estimation algorithm can be freely incorporated, and a highly accurate character string input means selection method having a highly reliable character recognition accuracy estimation result derivation method can be realized. It is effective.

Further, when the derivation processing method and the derivation order are determined based on the arithmetic expression described in the derivation process using the operator corresponding to the arithmetic rule, the derivation process can be freely performed in any order. It is possible to realize a character string input means selecting method which has a character recognition accuracy estimation result derivation method capable of designating a character recognition accuracy estimation result derivation method, and which is easy to process and has high accuracy.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a processing block configuration according to an exemplary embodiment of the present invention.

FIG. 2 is a diagram illustrating a processing flow of a character recognition accuracy estimation result derivation processing block in the above processing block configuration.

FIG. 3 is a diagram illustrating an example of deriving a character recognition accuracy estimation result using an operator in the flow of the above processing.

[Explanation of symbols]

101, 102, 103, 104, 105 ... Processing block for estimating character recognition accuracy 106 ... Processing block for deriving character recognition accuracy estimation result 107, 108, 109, 110, 111 ... Output from character recognition accuracy estimation processing block Estimation result data 112 ... Character recognition accuracy estimation result derivation method descriptive data 113, 114, 115, 116, 117 ... Result data 201 pre-processed for recognition accuracy estimation processing such as character recognition processing 201 ... Individual character recognition accuracy estimation Processing block for inputting result 202 Processing block for inputting character recognition accuracy estimation result derivation method description file 203 Processing block for interpreting and determining rule for deriving character recognition accuracy estimation result from description file 204 A processing block that derives the individual character recognition accuracy estimation results based on the calculation rules. 205 ... Processing block for determining end of derivation processing 206 ... Processing block for outputting derivation result of character recognition accuracy estimation result 301 ... Processing block for deriving a plurality of derivation (estimation) results by majority decision 302 ... Previous derivation (estimation) ) A processing block 303 that performs estimation processing by limiting the estimation target according to the derivation (estimation) result output by the processing ... Processing block that derives a plurality of derivation (estimation) results by logical product 304 ... Plural derivation (estimation) ) Processing block deriving a result by logical sum 305, 306, 307, 308, 309 ... Estimation result data output from the character recognition accuracy estimation processing block

Claims (3)

[Claims] 1. A character string input means selecting method for converting character image data into a character code and selecting one having a high input efficiency from a plurality of character string input means based on a result of character recognition accuracy estimation. Performing a plurality of character recognition accuracy estimations for the character image data, deriving a single character recognition accuracy estimation result from the plurality of character recognition accuracy estimation results, and the derived single character recognition And a step of selecting a character string input means having a high input efficiency from a plurality of character string input means based on the estimation result. 2. The step of deriving a single character recognition accuracy estimation result from a plurality of character recognition accuracy estimation results defines operators that correspond the calculation process to the operation rules, and uses the operators. 2. The method for selecting character string input means according to claim 1, further comprising a processing method for deriving the single character recognition accuracy estimation result and a step of deciding the order of deriving based on the constructed arithmetic expression. . 3. A processing method for deriving a single character recognition accuracy estimation result based on an arithmetic expression formed by using an operator and a process of deciding the order of derivation are performed by an operator described by a user. , The process of deciding the order of deriving a single character recognition accuracy estimation result, the process of deciding the type of derivation process associated with the operator in advance, and the characters described by the user as arguments of the operator. ID number of recognition accuracy estimation result or ID of character recognition accuracy estimation process
3. The character string input means selection method according to claim 2, further comprising: a step of determining a character recognition accuracy estimation result or an estimation process to be processed based on the number.