JPH0962773A - Character recognition device - Google Patents

Abstract

(57) 【Abstract】 PROBLEM TO BE SOLVED: To improve the efficiency of error detection accuracy of a recognition result to improve the efficiency of the correction work of the recognition result of a device user. A final recognition result is obtained by a plurality of processes such as a pattern matching process for an unknown character, and a certainty of the recognition result is comprehensively determined based on information obtained by a specific process result among the plurality of processes. Determines the certainty factor. Next, prior to the comparison between the determined certainty factor and the threshold value, the threshold value is set to a predetermined value at any time according to a predetermined condition (for example, the character type of the recognition result) (S3).
1). Next, the certainty factor is compared with a threshold value set at any time (S
32), based on the result of the comparison, it is determined that "the recognition result is suspicious" or "the recognition result is correct" (S33, S3).
4). The information of the determination result is added to the character of the recognition result and displayed or printed (S35).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device such as an optical character reading device (OCR), and more particularly to improving the accuracy of error detection of the recognition result of the device. It is also applicable to processing.

[0002]

2. Description of the Related Art Conventionally, as this type of apparatus, for example,
The character recognition device described in Japanese Patent Application Laid-Open No. 5-101216 and the character recognition device described in Japanese Patent Application Laid-Open No. 4-218183 are known. The character recognition device described in Japanese Patent Laid-Open No. 5-101216 discloses a similarity between the feature amount of each character cut out from a document image and the feature amount of each recognizable character stored in the recognition dictionary. To calculate. And
First candidate character, second candidate character in descending order of similarity ...
When the difference in similarity between the first candidate character and the second candidate character is less than a predetermined value, the first candidate character is highlighted because the recognition result has a high probability of erroneous recognition.

The character recognition device described in Japanese Patent Application Laid-Open No. 4-211883 performs character recognition processing as shown in the flowcharts of FIGS. 10 and 11. That is,
First, a character image is cut out from a binary image obtained by reading an original image (step 1), and then a normalization process is performed to correct the distortion of the cut out character image (step 2). . Next, step 3
In the pattern matching process, the feature amount of the normalized character image is extracted, the extracted feature amount is compared with the standard feature amount registered in the dictionary, and the similarity or difference is calculated. Alternatively, one or more recognition candidate characters having a small dissimilarity are determined. Then, when the processing from step 1 to step 3 is completed for all characters (step 4; Y), the process proceeds to the next step 5.

In the path selection process of step 5, for example, in the case of Japanese, an optimum combination of characters such as whether it is one full-width character or two half-width characters is selected. In the rule process of step 6, character size information (which is normalized so that it is related to case sensitivity such as “C” and “c”) and character type information (eg, katakana in a katakana character string) If there is a "one", the recognition result candidate by the pattern matching process is corrected according to a rule such as that which is corrected to the katakana long-sound symbol "-"). In the language processing of step 7, the character string of the corrected recognition result is morphologically analyzed by the rule processing, and the recognition error is corrected by collating with the dictionary. The result of this language processing is the final result of character recognition, and is the target of correction of the recognition result by the device user.

In the confidence determination process of step 8, the first candidate evaluation value of the pattern matching process (for example, a value obtained by dividing the distance from the dictionary by the total number of contours) or the first candidate and the second candidate.
The difference of the evaluation value with the candidate, the evaluation value when determining the path in the path selection process, the information of the rule applied for the correction in the rule process, the information showing the result of the language correction in the language process,
The certainty factor of the recognition result of the final recognition is comprehensively determined by using the above-mentioned information as evidence and using the probability theory of Dempster-Schafer described later. And for every character,
When the processes from step 5 to step 8 are completed (step 9; Y), the character recognition process is completed. Next, as shown in the flowchart of FIG. 11, the determined certainty factor is compared with a preset threshold value (step 10), and as a result, when the certainty factor is smaller than the threshold value (step 10; Y). Is determined to be "suspicious" (step 11), and when the certainty factor is greater than the threshold value (step 10; N), "recognition result is correct" (step 12). Then, this determination result is output to a display device or the like together with the recognition result (step 1
3).

Next, a specific example in which the certainty factor is comprehensively determined by using the above-described Dempster-Schaefer probability theory will be described. This Dempster-Schafer theory of probability has the characteristic that it can represent ignorance and can collect and synthesize fragmentary information (independent evidence), and the probability of "trust", "ignorance", and "untrust" It is divided into areas and "credibility" is called the lower bound probability, and "credit + ignorance" is called the upper bound probability. Specifically, the hypothesis H1 that this character is correct,
The hypothesis H2 that this character is wrong is set, and the pseudo-potential distribution (the distribution of the probability that the hypothesis is true assigned to each hypothesis) is obtained for each piece of evidence gathered. For example, the pseudo-possibility distribution regarding the evaluation value of the first candidate, which is the evidence gathered by the pattern matching process, also the first possibility
Pseudo-possibility distribution regarding difference in evaluation value between candidate and second candidate, pseudo-possibility distribution regarding difference in evaluation value at the time of path selection which is evidence collected by path selection processing, pseudo-possibility regarding evidence content collected from rule processing Generates a pseudo-possibility distribution related to the correction content collected from the sex distribution and language processing.

When the evidence is numerical information, this pseudo-possibility distribution is obtained by, for example, the following linear equation. y = -x / 3 + 100 In the formula, y is a pseudo-possibility distribution and x is evidence of numerical information. If the evaluation value is 100, the probability 67% of the hypothesis H1 and the probability 33% of the hypothesis H2 are calculated by this formula.
If the evidence is not numerical, give it a heuristic. For example, in the case of evidence that a certain rule is not satisfied, the probability of the hypothesis H1 is 20% and the probability of the hypothesis H2 is 80%.

From the pseudo-potential distribution for each piece of evidence thus collected, the basic probability allocation is obtained. That is, the pseudo possibility distribution is assigned to each probability of “correct”, “ignorance”, and “error”. For example, if the probability of hypothesis H1 is 67% and the probability of hypothesis H2 is 33%, 34% (= 67-33): correct 66% (= 100-34): ignorance 0%: like To do. Also, for example, the probability of hypothesis H1 is 20
%, If the possibility of hypothesis H2 is 80%, 0%: correct 30% (= 100-70): ignorance 70% (= 90-20): make an error.

Next, Dempster
The basic probability is synthesized by the integration rule of. Furthermore, lower bound probability and upper bound probability are calculated. Take the following jury model as an example. In this example, the lower bound probability of guilty is 15% and the upper bound probability is 74.
% (= 15 + 59), the innocence lower bound probability is 26%, and the upper bound probability is 85% (26 + 59). Finally, the confidence level is determined from the lower bound probability and the upper bound probability. For example, from the upper bound probability of hypothesis H2, A, B, C
Decide the rank of. The upper bound probability of H2 is 0 to 19 correct (A rank) The upper bound probability of H2 is 20 to 79 Suspicious (B rank) The upper bound probability of H2 is 80 to 100 False (C rank)

[0010]

SUMMARY OF THE INVENTION Incidentally, Japanese Patent Application Laid-Open No.
As described above, the character recognition device described in Japanese Patent No. 101216 determines that the recognition result has a high probability of erroneous recognition when the difference in the similarity between the first candidate character and the second candidate character is less than a predetermined value. Only the first candidate character can be highlighted, and the certainty factor, which is the certainty of the recognition result, cannot be compared with the threshold value. On the other hand, the character recognition device described in Japanese Patent Application Laid-Open No. 4-211883 can detect an error in the recognition result by comparing the certainty factor with a threshold value, but the threshold value is fixed and not necessarily appropriate. For this reason, in addition to the increase in the typefaces and character types to be recognized, in recent years the number of manuscripts to be recognized has been diversified, the accuracy of error detection in recognition results is insufficient, and the recognition results of device users There was a problem that the efficiency of the correction work of was insufficient.

Therefore, an object of the present invention is to provide a character recognition device which can improve the efficiency of error detection accuracy of the recognition result, thereby improving the efficiency of the correction work of the recognition result of the device user. .

[0012]

According to a first aspect of the present invention, there is provided a character recognizing device capable of obtaining a recognition result of an input unknown character and determining a certainty factor which is the certainty of the recognition result. Detecting means for detecting the error in the recognition result by comparing the determined certainty factor with a threshold value, threshold value changing means for changing the threshold value at any time according to a predetermined condition when the detecting means detects the error, The object is achieved by providing an output means for outputting an error in the recognition result detected by the detection means together with the recognition result. According to a second aspect of the present invention, in the character recognition apparatus according to the first aspect, the threshold changing means achieves the object by changing the threshold according to the type of the recognized character. According to a third aspect of the present invention, in the character recognition apparatus according to the first aspect, the threshold value changing means is configured to change the recognized character according to the degree of ease of overlooking by a corrector.
The object is achieved by changing the threshold value at any time. According to a fourth aspect of the present invention, in the character recognition device according to the first aspect, the threshold value changing means changes the threshold value at any time according to the degree of easiness of erroneous recognition of a character to be recognized. Achieve the purpose.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION Preferred embodiments of the character recognition device of the present invention will be described in detail below with reference to FIGS. FIG. 1 shows a configuration of a character recognition device according to a first embodiment of the present invention. As shown in FIG. 1, this character recognition device includes a scanner 1 for optically reading an image of an original document, an input device 2 such as a keyboard used for correcting a recognized character, and a character composed of each element described later. A recognition unit 3, a display 4 such as a CRT, and a printing device 5 such as a laser printer are provided, and these components are connected to a bus. Character recognition unit 3
Is an image memory 31 that stores images, a central processing unit 32 that controls each unit to perform a character recognition process, which will be described later, in a predetermined procedure, and a character recognition process that will be described later performed by the central processing unit 32. A ROM (Read Only Memory) 33 that stores various programs such as the following in advance, a dictionary 34 in which standard feature amounts of characters are registered, and various data during character recognition processing are temporarily stored. A work area RAM (random access memory) 35 is provided.

Next, the operation of the character recognition apparatus having such a configuration will be described with reference to the flowcharts of FIGS. Now, when the document image is read by the scanner 1, it is converted into a binary image (step 21), stored in the image memory 31, and then a cutting process for cutting out a character image from the binary image stored in the image memory 31 is performed. Perform (step 22). Next, a normalization process is performed to correct the distortion of the cut out character image (step 23). In the pattern matching process of step 24, the feature amount of the normalized character image is extracted, the extracted feature amount is compared with the standard feature amount registered in the dictionary 34 to calculate the similarity or the difference, and the similarity is calculated. One or more recognition candidate characters having a high degree or a low degree of difference are determined. And
When the processing from step 22 to step 24 is completed for all characters (step 25; Y), the process proceeds to the next step 26.

In the path selection process of step 26, for example, in the case of Japanese, an optimum combination of characters such as one full-width character or two half-width characters is selected. In the rule process of step 27, character size information (normalized, so it is related to case sensitivity such as “C” and “c”) and character type information (eg, katakana in a katakana character string). If there is a “one” in the above, this is corrected to the katakana long-sound “–”)
The recognition result candidate by the pattern matching process of No. 4 is corrected. In the language processing of step 28, the corrected recognition result character string is subjected to morphological analysis by the rule processing of step 27 and collated with the dictionary 34 to correct the recognition error. The result of this language processing is the final result of character recognition, and is the target of correction of the recognition result by the device user.

In the confidence determination process of step 29, for example, the first pattern matching process of step 24 is performed.
An evaluation value of a candidate (for example, a value obtained by dividing the distance from the dictionary by the total number of contours) or a difference between the evaluation values of the first candidate and the second candidate, an evaluation value when a path is determined by the path selection processing in step 26, Information on the rules applied for correction in the rule processing in step 27 and information representing the result of the language correction in the language processing in step 28 are collected, and the information is used as evidence to comprehensively use the probability theory of Dempster-Schaefer. , Determining the certainty factor of the recognition result of the final recognition. Since the specific method of obtaining the certainty factor has been described in the related art, it will be omitted. Here, the certainty factor indicates how likely the recognition result is, for example, 0 to 10
Expressed as an integer value of 0, the higher the number, the more likely it is. Then, when the processing from step 26 to step 29 is completed for all characters (step 30; Y), the character recognition processing is ended.

Next, a recognition result error detection process of comparing the certainty factor determined as described above with a threshold value and detecting an error in the recognition result from the comparison result will be described with reference to the flowchart of FIG. To do. In this processing, first, prior to the comparison with the certainty factor threshold value, the threshold value is set at any time to a predetermined value in accordance with a predetermined condition such as a character recognition result (step 31). In other words, the threshold value is changed at any time according to a predetermined condition such as a character recognition result. Then, the determined certainty factor is compared with a threshold value set at any time (step 3
2) As a result of the comparison, when the certainty factor is smaller than the threshold value (step 32; Y), it is determined that the recognition result is doubtful (step 33), and when the certainty factor is larger than the threshold value (step 32; In N), it is determined that the recognition result is correct (step 34). Then, the information of the determination result is added to the character of the recognition result and displayed on the display 4 or printed by the printing device 5 (step 35).

As a display example of the display 4 or a printing example by the printing device 5, a character when "recognition result is correct" is black, and a character when "recognition result is suspicious" is black with a white background. , The brightness of the character is changed depending on whether the recognition result is correct or not. With such a display, the device user can easily recognize the certainty factor of the recognition result, quickly and accurately find the correctable character, and efficiently perform the correction work of the recognition result.

As described above, according to the first embodiment, the threshold value for comparing the certainty factors is set to a predetermined value at any time in accordance with a predetermined condition such as a character recognition result.
The accuracy of error detection of the recognition result can be improved, and the efficiency of the work of correcting the recognition result of the device user can be improved.

Next, a character recognition device according to a second embodiment of the present invention will be described with reference to FIG. FIG.
FIG. 8 is a flow chart showing a main part of the operation of the character recognition device according to the second embodiment of the present invention. In the second embodiment, the error detection process of the recognition result shown in FIG. 3 of the first embodiment is further embodied as the error detection process of the recognition result shown in FIG.

That is, in the second embodiment, as shown in FIG. 4, prior to the comparison between the certainty factor and the threshold value, the threshold value is changed according to the type of the character obtained from the recognition result.
The value as shown in (4) is set at any time (step 41). In other words, the threshold is changed as needed according to the type of character obtained from the recognition result. Then, the certainty factor is compared with a threshold value set at any time (step 42), and when the certainty factor is smaller than the threshold value as a result of the comparison (step 42; Y), it is determined that the “recognition result is suspicious” (step 42). 43), when the certainty factor is larger than the threshold value (step 42; N), it is determined that "the recognition result is correct" (step 44). And
Information of the determination result is added to the character of the recognition result and displayed on the display 4 or printed by the printing device 5 as in the first embodiment (step 45). The second
Since the character recognition processing and its configuration of the second embodiment are similar to those of the first embodiment, the description thereof will be omitted.

As described above, according to the second embodiment, the threshold value for comparing the certainty factors is set at any time according to the character type obtained from the recognition result. It is possible to prevent the error detection accuracy from being varied due to the recognition result, and to improve the efficiency of the correction work of the recognition result by the device user.

Next, a character recognition device according to a third embodiment of the present invention will be described with reference to FIG. Figure 6
FIG. 9 is a flowchart showing a main part of the operation of the character recognition device according to the third embodiment of the present invention. In the third embodiment, the error detection process of the recognition result shown in FIG. 3 of the first embodiment is further embodied as the error detection process of the recognition result shown in FIG.

That is, in the third embodiment, as shown in FIG. 6, the character obtained by the recognition result prior to the comparison between the certainty factor and the threshold value is a character that is easily overlooked by a human when correcting the recognition result. A threshold value is set at any time depending on whether or not (step 61). In other words, the threshold value is changed at any time according to the degree of ease of overlooking when correcting the character obtained from the recognition result. Here, the meaning of whether or not a character is easily missed by a person when correcting the recognition result is, for example, “O (oh)”.
And 0 (zero), "Ta and Yu (Yu)", "He (Hiragana) and He (Katakana)", which are hard to see the difference visually. These are overlooked when the recognition result is corrected. Because it is easy. Therefore, unlike other characters, this kind of character should be detected as an error as easily as possible.

Therefore, for example, as shown in FIG. 7, the threshold is changed according to the category. Category "1" is "O
(Oh) and 0 (zero) "," 1 (one) and l (el) "
English-numeric relations such as, category "2" is Kanji-katakana relations such as "Ta and Yu", "To and me", and category "3" is "He (Hiragana)" Hiragana-Katakana relations such as "and f (katakana)", "ri (hiragana) and ri (katakana)", and other characters are other categories. Then, the certainty factor is compared with the threshold value set at any time as described above (step 6).
2) As a result, when the certainty factor is smaller than the threshold value (step 62; Y), it is determined that the recognition result is doubtful (step 63), and when the certainty factor is larger than the threshold value (step 62; N). Is determined to be "correct" (step 64). Next, the information of the determination result is added to the character of the recognition result and displayed on the display 4 or printed by the printing device 5 as in the first embodiment (step 65). Since the character recognition processing and its configuration of the third embodiment are the same as those of the first embodiment, the description thereof will be omitted.

As described above, according to the third embodiment, the threshold value used for comparing the certainty factors is determined according to whether the character obtained from the recognition result is a character that is easily overlooked by humans when correcting the recognition result. Since it is set as needed, the accuracy of error detection of the recognition result can be improved for characters that are easily overlooked by humans when correcting the recognition result, and the efficiency of the work of correcting the recognition result of the device user can be improved.

Next, a character recognition device according to a fourth embodiment of the present invention will be described with reference to FIG. FIG.
FIG. 9 is a flow chart showing a main part of the operation of the character recognition device according to the fourth embodiment of the present invention. In the fourth embodiment, the error detection process of the recognition result shown in FIG. 3 of the first embodiment is further embodied as the error detection process of the recognition result shown in FIG.

That is, in the fourth embodiment, as shown in FIG. 8, it is determined whether the recognized character is a character that is likely to be erroneously recognized by the character recognition unit 3 prior to the comparison of the certainty factor and the threshold value. Accordingly, the threshold value is set at any time (step 81). In other words, the threshold is changed at any time according to the degree of ease of erroneous recognition of the recognized character in the character recognition unit 3. Here, the meaning of whether the recognized character is a character that is easily erroneously recognized is that it is close in the feature amount space. Therefore, the content of the category corresponding to the threshold naturally changes depending on what kind of feature amount the character recognition adopts. Whether or not they are close to each other in the feature amount space is determined by clustering the feature amounts of all characters and determining a threshold value as a category of characters that is easily erroneously recognized as a sufficiently large one among several determined clusters. For example, as shown in FIG. 9, the threshold is changed depending on the category.

Then, the certainty factor is compared with the threshold value set at any time as described above (step 82). As a result, when the certainty factor is smaller than the threshold value (step 82; Y), "the recognition result is suspicious". When the certainty factor is larger than the threshold value (step 82; N), it is determined that the "recognition result is correct" (step 84). Next, the information of the judgment result is added to the character of the recognition result and displayed on the display 4 or printed by the printing device 5 as in the first embodiment (step 85). Since the character recognition processing and its configuration of the fourth embodiment are the same as those of the first embodiment, the description thereof will be omitted.

As described above, according to the fourth embodiment, the threshold value for comparing the certainty factors is set at any time according to whether the recognized character is a character that is easily misrecognized. It is possible to improve the accuracy of the error detection of the recognition result of the character that is easily erroneously recognized, so that the efficiency of the correction work of the recognition result of the device user can be improved.

In the second embodiment described above, the threshold is changed as needed according to the type of character.
In this embodiment, the character obtained from the recognition result is changed at any time depending on whether or not the character is a character that is easily missed by a person when correcting the recognition result. However, the threshold changing condition may take these two conditions into consideration.

[0032]

According to the first aspect of the present invention, the certainty factor of the recognition result is compared with the threshold value, and when the error in the recognition result is detected, the threshold value is changed according to a predetermined condition. Therefore, the error detection accuracy of the recognition result is improved, and the efficiency of the work of correcting the recognition result by the device user can be improved. According to the second aspect of the present invention, the threshold value used when comparing the certainty factors is changed at any time according to the character type obtained from the recognition result. Therefore, the error detection accuracy of the recognition result depending on the character type can be improved. It is possible to prevent variations, and thus to improve the efficiency of the work of correcting the recognition result of the device user.

According to the third aspect of the present invention, the threshold value for comparing the certainty factors is changed at any time according to the degree of ease of overlooking by the corrector when correcting the recognized character. It is possible to improve the accuracy of error detection of the recognition result for a character that is easily overlooked when correcting the recognition result, and thus to improve the efficiency of the work of correcting the recognition result by the device user. In the invention according to claim 4, the threshold value for comparing the certainty factors is changed at any time according to the degree of easiness of erroneously recognizing the recognized character. It is possible to improve the accuracy of error detection,
Therefore, the efficiency of the work of correcting the recognition result of the device user can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of a character recognition device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a character recognition process by the character recognition device.

FIG. 3 is a flowchart showing a recognition result error detection process by the character recognition device.

FIG. 4 is a flowchart showing a recognition result error detection process by the character recognition device according to the second embodiment of the present invention.

FIG. 5 is a table showing an example of threshold values for each character type.

FIG. 6 is a flowchart showing a recognition result error detection process by the character recognition device according to the third embodiment of the present invention.

FIG. 7 is a table showing an example of threshold values in each category as viewed from a human.

FIG. 8 is a flowchart showing a recognition result error detection process by the character recognition device in the fourth embodiment of the present invention.

FIG. 9 is a table showing an example of thresholds in each category as seen from character recognition.

FIG. 10 is a flowchart showing a character recognition process of a conventional device.

FIG. 11 is a flowchart showing an error detection process of a recognition result of a conventional device.

[Explanation of symbols]

 1 Scanner 2 Input Device 3 Character Recognition Unit 4 Display 5 Printing Device 31 Image Memory 32 Central Processing Unit 33 ROM 34 Dictionary 35 Work Area RAM

Claims (4)

[Claims] 1. A character recognition device capable of obtaining a recognition result for an unknown character that has been input and determining a certainty factor, which is the certainty of the recognition result, by comparing the determined certainty factor with a threshold value. A detection unit that detects an error in the recognition result, a threshold value changing unit that changes the threshold value at any time according to a predetermined condition when the detection unit detects the error in the recognition result detected by the detection unit, A character recognition device comprising: an output unit that outputs the recognition result together with the recognition result. 2. The character recognition device according to claim 1, wherein the threshold value changing means changes the threshold value at any time according to the type of the recognized character. 3. The character recognition device according to claim 1, wherein the threshold value changing means changes the threshold value at any time in accordance with the degree of easiness of being missed by the corrector when correcting the recognized character. . 4. The character recognition device according to claim 1, wherein the threshold value changing unit changes the threshold value at any time according to the degree of easiness of erroneous recognition of a character to be recognized.