JPH096921A - Method and device for inputting character string - Google Patents

Abstract

PURPOSE: To provide a method and device for character string input which can mechanically select a proper input method corresponding to variation in the front of input characters. CONSTITUTION: As for inputted character image data 105, a character quality evaluation processing block 101 estimates the correct answer rate of a character candidate of 1st order, i.e., a character recognition rate at the time of character recognition and the rate at which the correct answer character candidate is included in character candidates of up to (n)th order, i.e., a correct answer character candidate inclusion rate. Further, a means which has the shortest time required for conversion processing, i.e., the shortest input means is selected out of input means 102-104 which convert the character image data into character codes. By the selected input means, character strings are inputted with high input efficiency corresponding to input character quality at all times.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process for inputting a Japanese character string or the like into a computer system or the like using character recognition technology, and in particular, the efficiency of character string input depends on the quality of the character to be recognized. The present invention relates to a technique for realizing an efficient character string input method by selecting the maximum input means.

[0002]

2. Description of the Related Art A character recognition technique for mechanically converting character image data input from a scanner or a fax into a character code that can be processed by a computer as text data is to reduce manual input operation of characters. Considered a useful technique. In the character recognition technique, the basic processing is to identify which character the input character image data is, based on the similarity with the previously learned character pattern. For this reason, the recognition performance is stable for printed characters whose fonts are relatively less deformed than handwritten characters, and many products using this technology have been announced.

However, for handwritten characters, the character styles are different depending on the person who entered the text, and it is difficult to correctly identify a character style different from the learned character style. Markedly reduced. In spite of the vulnerability of such character recognition technology to handwritten characters, the current character recognition system requires various types of fonts to be input, and correction errors that are less efficient than direct key input methods are required. The situation is strong.

[0004]

As described above, the conventional character string inputting method using the character recognition technology is capable of uniform character recognition regardless of variations in the font deformation (character quality) of the input character. Since it is executed, especially when inputting characters with large font variations, it takes longer time to input than when inputting directly from the keyboard without using character recognition technology, and the effect of character recognition cannot be exhibited. was there.

The present invention has been made in view of the above circumstances, and an object of the present invention is to solve the above-mentioned problems in the prior art and to provide an appropriate input method for a character variation of an input character as a machine. To provide a character string input method and device that can be selectively selected.

[0006]

In order to achieve the above-mentioned object of the present invention, in the present invention, when character recognition is performed on input character image data in a process of converting character image data into a character code. Of estimating the correct answer rate of the first-ranked character candidates, that is, the character recognition rate, and the rate of the correct-character candidates included in the character candidates up to the n-th order, that is, the correct-character candidate inclusion rate, and the estimated character recognition rate and Selecting a conversion means that minimizes the time required for the conversion processing among a plurality of conversion means that converts the character string image data into a character code based on the estimation result of the correct character candidate inclusion rate. A character string input method characterized by

In the above character string input method, the step of estimating the character recognition rate and the correct answer candidate inclusion rate for the input character image data is performed by matching the first-ranked character candidates of the character recognition results by a plurality of character recognition algorithms. The process of estimating the character recognition rate from the degree and a set of character candidates that are similar patterns for each character to be recognized are defined and the degree to which the similar pattern character candidates are included in the actual character recognition result is observed. It is preferable that the configuration includes a step of estimating the correct character candidate inclusion rate.

Further, among the plurality of converting means for converting the character string image data into the character code based on the estimated result of the estimated character recognition rate and the correct answer candidate inclusion rate, the time required for the converting process becomes the shortest. In the step of selecting the conversion means, for each conversion means, the relationship between the character recognition rate, the correct character candidate inclusion rate, and the time required for the conversion process is observed in advance, and the character recognition rate and the correct character in which the merits and demerits of each conversion means are reversed. The process of obtaining the specified value of the candidate coverage, and comparing the estimated character recognition rate and the value of the estimation result of the correct character candidate coverage with the specified value to determine the superiority or inferiority of each conversion means. It is preferable to adopt a configuration including a process of selecting a conversion unit that requires the shortest processing time.

Further, in order to achieve the above-mentioned object of the present invention, in the present invention, in a device for converting character image data into a character code, the first place in the case of performing character recognition on input character image data. Means for estimating a correct answer rate of the character candidates, that is, a character recognition rate; and means for estimating a rate of correct character candidates in the character recognition included in the character candidates up to the nth, that is, a correct answer candidate inclusion rate.
Based on the estimation results of the estimated character recognition rate and the correct answer candidate inclusion rate, a conversion means that minimizes the time required for the conversion processing is selected from a plurality of conversion means that converts the character string image data into a character code. And a means for performing the character string input device.

[0010]

In the character string input method and the apparatus thereof according to the present invention, the character quality of the input character image data is evaluated based on the estimation results of the character recognition rate and the correct character candidate inclusion rate to correspond to the input character quality. By selecting the conversion means that minimizes the time required for converting the character string image data into the character code, the character string can be input with high input efficiency according to the input character quality.

Specifically, when the input character quality is high, a conversion means for directly using the character recognition result is selected, and
If the input character quality is medium, the efficiency of the work of selecting the conversion means that corrects the character recognition error by matching with the word dictionary corresponding to the input character string and further correcting the character recognition error is low. When the input character quality is low, by selecting the conversion means for directly inputting from the keyboard or the like, the character string can be input with the input efficiency higher than that of the method of directly inputting from the keyboard or the like.

[0012]

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

FIG. 1 is a processing block diagram of a character string input method and apparatus according to an embodiment of the present invention. In FIG. 1, 101 is a character quality evaluation processing block. 1
02, 103, and 104 are means for converting character image data into a character string code, that is, input means for specifically inputting a character string. In FIG. 1, for convenience of explanation, an example in which there are three types of input means is shown, but it is sufficient if there is an effect from the viewpoint of character string input, and the number of types of input means is the same as the gist of the present invention. Is irrelevant. Also,
Although the type thereof is not limited, in order to specifically proceed with the description, hereinafter, 102 is used as an input means for character recognition and recognition correction, and 103 is used for character recognition and matching with a word dictionary (word matching). 1 as input means
Reference numeral 04 is an input means for manually inputting directly from the keyboard without using character recognition.

A specific character string input method according to this embodiment will be described below. First, the character quality evaluation processing block 1
In 01, the character quality is evaluated for the input character image data 105. That is, the correct answer rate (character recognition rate) of the first-ranked character candidates when character recognition is performed on the input character image data 105 and the rate of correct answer character candidates included up to the nth position in the character candidate ranking ( Estimate the correct character candidate coverage). The method of estimating the character recognition rate and the correct answer candidate inclusion rate in the character quality evaluation processing block 101 will be described later in detail with reference to FIG.

The input of the character image data 105 is as follows.
For example, it is performed for each form or page for which it can be assumed that the character quality is constant, and the input means is selected for that range. Of course, it is also possible to specify an arbitrary range and select the input means within that range. In addition, the character recognition rate and the correct answer candidate inclusion rate can be estimated on a character-by-character basis. In that case, for example, the average of estimated values for each character in the range of input character image data is used as the estimation result. To do.

Next, in the character quality evaluation processing block 101, of the selectable input means 102, 103, 104,
The input means suitable for the estimated result of the estimated character recognition rate and the correct answer candidate inclusion rate is selected. Specifically, for example, when the specified character recognition rate is X% or more, the input unit 102 is selected by the activation information 106, the character recognition rate is less than X%, and the correct character candidate inclusion rate is specified. If it is Y% or more, the input means 103 is selected by the start information 107, and if the character recognition rate is less than X% and the correct character candidate inclusion rate is less than Y%. A method such as selecting the input means 104 according to the activation information 108 can be adopted. The above startup information includes character image data when necessary, such as when performing character recognition. In this way, the selected input means converts the character image data into the character string code 109 and inputs it to, for example, a computer system or the like.

It should be noted that the prescribed value X is the value when the first-ranked character candidate in the case of character recognition is the prescribed value X% or more,
This is a value that is determined as the efficiency of character string input by the input unit 102 being higher than the efficiency of the other input units 103 and 104 (time required for conversion processing, that is, input time is short). Further, when the specified value Y is the character recognition rate is less than Y% and the correct character candidate inclusion rate is not less than the specified value Y%, the efficiency of the character string input by the input means 103 is
This is a value determined to be higher than the efficiency of the other input means 102, 104.

As described above, according to the character string input method shown in FIG. 1, the input means is switched according to the quality of the characters in the input image data, so that the character string input is always performed with the maximum character string input efficiency. It becomes possible.

FIG. 2 is a diagram for explaining an example of the relationship between the quality of input characters and the character recognition rate and the correct answer inclusion rate. In FIG. 2, 201 shows the character recognition rate with respect to the input character quality, and 202 shows the coverage rate up to the nth position of the correct character candidate with respect to the input character quality.
When n is 1, the character recognition rate 201 and the correct character candidate inclusion rate 202 have the same value.

As described above, the character recognition rate is the accuracy rate of the first-ranked character candidate. The higher the rate, the smaller the correction amount for the subsequent recognition error. Increase. However, at the current level of character recognition technology, especially for handwritten characters, the character recognition rate is 2 in FIG.
As shown in 01, the input character quality is greatly influenced, and in order that the method of character recognition and recognition correction shown in the input means 102 in FIG.
The recognition rate should be close to%. On the other hand, as shown in FIG. 2, when the same character recognition algorithm is used, the correct answer candidate inclusion rate 202 is higher than the character recognition rate for all character qualities. Therefore, even if the character recognition rate is low, if the correct character candidate inclusion rate is high, the input unit 103 that corrects the character recognition error by matching with the word dictionary becomes the most efficient character string input unit. Is likely to be.

FIG. 3 is a diagram showing an example in which the relationship between the quality of input characters and the relative character string input efficiency of various character string input means is assumed. In FIG. 3, 301
Indicates the input efficiency when a character string is directly input from the keyboard, and the character string input efficiency of other input means is 30.
It is a normalized reference value for calculating the assumed efficiency of 2,303. As the unit indicating the character string input efficiency, for example, the number of characters that can be input per unit time can be used.
In the case of a direct input from the keyboard, if the character string is readable by the input person, the input efficiency does not depend on the quality of the input character. Therefore, the graph 301 in FIG. 3 is a straight line regardless of whether the character quality is high or low. Reference numeral 302 shows the relative character string input efficiency of the character string input means assuming character recognition and matching with a word dictionary.

Reference numeral 302 is a relative value when the value of 301 is used as a normalization reference value. Reference numeral 303 indicates the relative input efficiency of the character string input means assuming character recognition and recognition correction of the recognition result. Note that 303 is also a relative value when the value of 301 is used as a normalization reference value. Reference numeral 304 denotes an input unit of the input efficiency 301 (104 in FIG. 1) and the input efficiency 3
02 indicates the value of the character quality at which the superiority or inferiority of the input means (103 in FIG. 1) is reversed, and 305 indicates the input means 103 of the input efficiency 302 and the input means of the input efficiency 303 (10 in FIG. 1).
2) Shows the value of character quality in which superiority and inferiority are reversed.

As shown in FIG. 3, the input means that maximizes the character string input efficiency differs depending on the change in the input character quality. In the example of FIG. 3, in the character input quality 304, the efficiency of the direct input means 104 from the keyboard and the efficiency of the input means 103 by character recognition and matching with the word dictionary intersect, and thus the efficiency of the two input means is reduced. The merits and demerits are reversed. Further, in the input character quality 305, the efficiency of the input means 103 by character recognition and matching with a word dictionary and the efficiency of the input means 102 by confirmation and correction of character recognition and recognition confirmation intersect, and these two input means are used. The merits and demerits of are reversed. As shown in FIG. 3, the character quality value (specified value) 304 at which the merits and demerits of the three types of input means are reversed.
Also, by observing 305 and 305 in advance, it is possible to perform the character string input with the maximum input efficiency under any character quality condition.

FIG. 4 shows the character string input efficiency according to the present invention when the maximum efficiency of character string input can be achieved according to the character quality. 401 is 3 shown in FIG.
It shows the efficiency of integrating the regions that maximize the character string input efficiencies of each type of input means. As shown in FIG. 3, the character quality value 304 in which the superiority or inferiority of each input means is reversed,
Estimating 305 is important because it affects the effect of the present invention. As shown in the explanation of FIG. 1, the character quality is
It is determined by the character recognition rate and the correct answer candidate inclusion rate.
For example, 305 is defined by the character recognition rate, and 304 is defined by the correct character candidate inclusion rate. That is, in the area where the input unit 102 for character recognition and recognition correction is most effective, the character recognition rate has the greatest influence, and therefore the lower limit character recognition rate 305 needs to be set. on the other hand,
In the area where the input means 103 by character recognition and matching with the word dictionary is most effective, the influence from the coverage rate of correct character candidates is the greatest, so it is necessary to determine the lower limit correct content candidate coverage rate 304.

FIG. 5 is a drawing for explaining the details of the character quality evaluation processing block 101 in FIG. In FIG. 5, reference numeral 501 denotes a processing block for estimating the correct answer rate (character recognition rate) of the first-ranked character candidates, and 502 is processing for estimating a correct answer character candidate inclusion rate (correct answer character candidate inclusion rate) up to the nth position. It is a block. Here, n is a constant that can be set arbitrarily, and in general, it is generally an integer value of about 2 to 32. 503 is processing blocks 501 and 502.
It is a processing block that selects the input means that maximizes the character string input efficiency based on the character recognition rate and the correct answer candidate inclusion rate.

The estimation of the character recognition rate in the processing block 501 can be realized by estimating from the matching degree of the first-ranked character candidates of the character recognition results by a plurality of character recognition algorithms. Further, in the estimation of the correct character candidates in the processing block 502, a set of character candidates having similar patterns is determined for each character, and the degree to which these similar pattern character candidates are included in the actual character recognition result is determined. It can be estimated by a method such as observation. Processing block 503
The selection of the input means in
This can be realized by observing the relationship between the character recognition rate and the correct character candidate inclusion rate and the input time, and obtaining the character recognition rate and the correct character candidate inclusion rate at which the merits and demerits of each input means are reversed. For example, assuming three kinds of character string input means, A, B, and C, when the character recognition rate is X%, the character string input efficiency of the means A is the highest, the character recognition rate is less than X%, and the correct answer is given. When the character coverage is Y% or more, the character string input efficiency of the means B is the highest, and when the character recognition rate is less than X% and the correct character candidate coverage is less than Y%, the character string input of the means C is performed. It is assumed that the observation result that the efficiency is highest is obtained.
Under such an observation result, processing block 503
For the above, the estimation results of the character recognition rate and the correct answer candidate inclusion rate may be applied to each of the above conditions, and the input means that maximizes the character string input efficiency may be selected.

[0027]

As described above in detail, the present invention estimates a character recognition rate and a correct answer candidate inclusion rate for an input character image, and converts a plurality of character string image data into character codes. Among these, the method that minimizes the time required for conversion processing is selected, so it is possible to mechanically select an appropriate input method with high input efficiency that corresponds to the font variation and quality of input characters. This has the remarkable effect of realizing the column input method and its apparatus.

Further, the character recognition rate is estimated from the degree of coincidence of the character recognition results obtained by a plurality of character recognition algorithms, and the correct character candidate inclusion rate is estimated from the degree to which similar pattern character candidates are included in the character recognition results. If or
For each conversion means, by observing the time required for the conversion processing, the prescribed values of the character recognition rate and the correct character candidate inclusion rate at which the superiority or inferiority of each conversion means reverses are obtained, and each conversion means is compared with the estimation result. When selecting the conversion method that determines the time required for the conversion process by determining the superiority or inferiority of the It is possible to easily realize a character string input method that can be selected.

[Brief description of drawings]

FIG. 1 is a diagram illustrating processing blocks of a character string input method according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining the relationship between the quality of input characters, the character recognition rate, and the correct answer character inclusion rate in the above embodiment.

FIG. 3 shows the quality of input characters in the above embodiment,
It is a figure for demonstrating the relationship with the relative character string input efficiency of various character string input means.

FIG. 4 is a diagram illustrating character string input efficiency according to the above embodiment.

FIG. 5 is a diagram for explaining details of a character quality evaluation processing block in the embodiment.

[Explanation of symbols]

101 ... Character quality evaluation processing block 102, 103, 104 ... Input means for inputting character string 201 ... Character recognition rate for input character quality 202 ... Inclusion rate (correct answer character) up to the nth position of correct answer candidate for input character quality Candidate inclusion ratio) 301 ... Input efficiency when a character string is directly input from a keyboard 302 ... Relative character string input efficiency of a character string input means assuming character recognition and matching with a word dictionary 401 ... Of three kinds of input means Input efficiency that integrates areas of maximum character string input efficiency 501 ... Processing block for estimating character recognition rate 502 ... Processing block for estimating correct character candidate inclusion rate 503 ... Input means for maximizing character string input efficiency Processing block to select

Front Page Continuation (72) Inventor Yoshimasa Kimura 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corp. (72) Shoji Kurakake 1-1-6 Uchiyuki-cho, Chiyoda-ku, Tokyo Nihon Telegraph Phone Co., Ltd.

Claims (4)

[Claims] 1. In the process of converting character image data into a character code, the correct answer rate of the first-ranked character candidates, that is, the character recognition rate and the correct answer character candidate, when the character recognition is performed on the input character image data, is first. Based on the step of estimating the rate included in the character candidates up to the nth position, that is, the correct character candidate inclusion rate, and the character string image data is converted into a character code based on the estimated result of the character recognition rate and the correct character candidate inclusion rate. And a step of selecting a conversion means that minimizes the time required for the conversion processing from among a plurality of conversion means for converting into a character string input method. 2. The step of estimating the character recognition rate and the correct answer candidate inclusion rate for the input character image data is the first step of character recognition results by a plurality of character recognition algorithms.
The process of estimating the character recognition rate from the degree of matching of the character candidates and the set of character candidates that are similar patterns for each character to be recognized are defined, and the similar pattern character candidates are included in the actual character recognition result. 2. The method of inputting a character string according to claim 1, further comprising: 3. The time required for the conversion process is the shortest among a plurality of conversion means for converting the character string image data into a character code based on the estimated results of the estimated character recognition rate and the correct answer candidate inclusion rate. In the step of selecting the conversion means, the relationship between the character recognition rate and the correct answer candidate inclusion rate and the time required for the conversion process is observed for each conversion means in advance, and the character recognition rate and the correct character in which the superiority and inferiority of each conversion means reverses. The process of obtaining the specified value of the candidate coverage, and by comparing the estimated character recognition rate and the value of the estimation result of the correct answer character candidate coverage and the specified value to determine the superiority or inferiority of each conversion means, the conversion The character string input method according to claim 1 or 2, further comprising: a step of selecting a converting means that minimizes a processing time. 4. An apparatus for converting character image data into a character code, and means for estimating a correct answer rate, that is, a character recognition rate of a first-ranked character candidate when character recognition is performed on input character image data, Based on a means for estimating a rate of correct character candidates included in the character candidates up to the nth position in the character recognition, that is, a correct character candidate inclusion rate, and an estimated result of the estimated character recognition rate and correct character candidate inclusion rate. A character string input device comprising: a plurality of converting means for converting the character string image data into a character code; and a means for selecting a converting means that minimizes the time required for the conversion processing.