JPH0540853A - Post-processing system for character recognizing result - Google Patents

Abstract

PURPOSE:To correct the wrong character recognizing result into the grammatically correct words. CONSTITUTION:The characters are recognized and the difference of resemblance is calculated between a candidate character and its original character pattern in a step 1. In a step 2, the candidate characters are combined together into a candidate word and also the assurance of the candidate is calculated from the difference of resemblance. In a step 3, the contents of a grammar dictionary 4 and a part-of-speech dictionary 5 are referred to for the words whose maximum assurance values of their candidates are less than a prescribed threshold level. At the same time, the grammatical relations are utilized between those words having the maximum assurance values of their candidates less than the threshold level and the words having the maximum assurance values of their candidates more than the threshold level. Thus the wrong words can be corrected into the grammatically correct words.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for recognizing and outputting an optically read character, and more particularly to a character recognition for automatically correcting and outputting an error in the character recognition result. It relates to the post-processing method of the result.

[0002]

2. Description of the Related Art Conventionally, as a technique in this field, for example, there is one disclosed in Japanese Patent Laid-Open No. 2-267670.
The technology disclosed in the above document, when an unrecognizable character (reject character) is present in a word, automatically detects candidate characters from the character table based on the characters before and after this reject character, the character array before and after, and the position. The character recognition result was corrected by calling and searching for a word in which this was replaced with a reject character.

[0003]

However, since the above-mentioned conventional technique is a method of making corrections on a character-by-character basis and does not utilize a grammar that defines the relationship between words, the correction result cannot be meaningfully understood. , There was a problem in that there is a possibility of correcting a word that is grammatically incorrect. SUMMARY OF THE INVENTION It is an object of the present invention to solve the above problems and provide a post-processing method of a character recognition result that can correct a grammatically correct word even if the character recognition result is incorrect.

[0004]

In order to solve the above problems, the present invention creates a candidate word for each word of a character recognition result and calculates the certainty factor, and the maximum value of the certainty factor is a predetermined value. Using a grammatical relationship between a word smaller than a threshold value and a word whose maximum confidence value of a candidate word close to the word is larger than a predetermined threshold value, the word is grammatically correct. It is to correct it into words. In the present invention, the certainty factor is the degree to which the candidate word is surely correct.

[0005]

According to the present invention, since the post-processing method of the character recognition result is configured as described above, the candidate word for each word of the character recognition result is created and the certainty factor is calculated, and the maximum value of the certainty factor is calculated. For a word whose is less than a predetermined threshold, use the grammatical relationship with the word whose maximum confidence value of candidate words near that word is greater than the predetermined threshold to Correct words. Therefore,
Even if the result of character recognition is incorrect, it can be corrected to a grammatically correct word.

[0006]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a flow chart showing a post-processing method of a character recognition result according to an embodiment of the present invention, and FIG. 2 is a block diagram showing a character recognition device implementing a post-processing method of a character recognition result according to the embodiment of the present invention. 3 is a diagram showing an example of a candidate word and a certainty factor in the embodiment of the present invention, and FIG. 4 is an explanatory diagram of a candidate word selection process in the embodiment of the present invention.

In FIG. 2, 11 is a CPU that controls the entire apparatus, 12 is a part-of-speech dictionary that describes part-of-speech information that gives a part-of-speech for a word, and 13 is a grammar dictionary that describes information that gives a relationship between parts-of-speech. Reference numeral 14 is a document reading means for reading a word on the document, 15 is a character recognizing means for recognizing the read character and outputting a candidate character and its distance, 16 is a character recognition result for storing the candidate character as a recognition result and its distance Storage means, 17 is a candidate word creating means for creating a candidate word by combining candidate characters, 18 is a confidence factor calculating means for calculating the certainty factor from the distance between the candidate characters, and 19 is a part-of-speech dictionary when selecting a candidate word. 12 is a dictionary search means for searching the grammar dictionary 13, 20 is an output word determining means for selecting a candidate word based on the contents of the dictionary, 21 is a result display / output for displaying / outputting the output word It is a stage.

Hereinafter, the processing operation of the post-processing method of the character recognition result according to the embodiment of the present invention will be described with reference to FIGS. (1) Character Recognition (Step 1) The word on the document is read by the document reading means 14, the character recognition means 15 recognizes each character of the read word, and the distance between them is calculated and stored in the recognition result storage means 16. To do. Here, the distance represents the similarity between each candidate character and the original character pattern, and the smaller the numerical value, the more similar the candidate character and the original character pattern.

(2) Candidate word creation / certainty factor calculation (step 2) The candidate word creation means 17 creates candidate words by combining the candidate characters stored in the recognition result storage means 16. Further, the certainty factor calculating means 18 calculates the certainty factor based on the distance of the candidate character. Here, the certainty factor is the degree of certainty that the candidate word is correct, and in the present embodiment, “the certainty factor of the candidate word” = “the reciprocal of the distance of the candidate word / each candidate word for one input word” The sum of the reciprocals of the distances ".

FIG. 3 shows an example of the calculation result for the English sentence "I am a boy.", In which the candidate word and the certainty factor for each character pattern are given. In the figure, the certainty factor of the candidate word “I” with respect to the word “I” is 100%, the certainty factor of the candidate word “an” with respect to the word “am” is 60%, and the certainty factor with respect to the candidate word “am” is 40
%. Similarly, candidate words and certainty factors are given to the words "a", "boy", and ".".

(3) Post-processing correction (step 3) When the certainty factor of the candidate word is 100%, the output word determining means 20 selects the candidate word as it is. In FIG. 3, "I" and "boy" correspond to this. next,
For a word whose maximum confidence value for each word is lower than the threshold value, the candidate word is selected by using the grammatical features of the words whose maximum confidence value is higher than the threshold value for the word. I do. At this time, the dictionary search means 19 searches the part-of-speech dictionary 12 and the grammar dictionary 13 and refers to the contents thereof.

FIG. 4 shows an embodiment for the calculation result of FIG. 3, and if the threshold value of the certainty factor is 90%, the word "a"
m ”,“ a ”, and“. The maximum confidence value of the candidate word for "is lower than the threshold value. Since it is appropriate to have a grammatical period for "," the candidate word ". “,”, Among “,”. Is selected and output as a correction result. Further, for the word “a”, the next word is “boy” with a certainty factor of 100%, and its part of speech is a noun. Candidates for the word “a” The word "a" is an article, and no part of speech is given to other candidate words.
And it is appropriate that the article comes before the noun,
The candidate word "a" is selected as a result. Further, with respect to the word "am" before that, the verb "am" is selected because it is grammatically not allowed that two consecutive articles appear. In this way, by using the grammatical relationship with the preceding and following words, it is possible to correct the word to a word that is not grammatically incorrect even if the certainty factor is low.

Although the embodiment of the present invention has been described above for the case of English words, the present method can be applied to other languages by changing the grammatical knowledge to be used into the grammar of another language or programming language. It is possible. Further, the present invention is not limited to the above embodiments, and various modifications can be made based on the gist of the present invention, and they are not excluded from the scope of the present invention.

[0014]

As described above in detail, according to the present invention, the grammatical feature of a word is used. Therefore, even if the result of character recognition is incorrect, the word can be corrected to a grammatically correct word. You can

[Brief description of drawings]

FIG. 1 is a flow chart showing a post-processing method of a character recognition result according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a character recognition device that implements a post-processing method for character recognition results according to an embodiment of the present invention.

FIG. 3 is a diagram showing an example of a candidate word and a certainty factor in the embodiment of the present invention.

FIG. 4 is an explanatory diagram of a candidate word selection process according to the embodiment of this invention.

[Explanation of symbols]

 1 Character recognition 2 Candidate word creation / certainty factor calculation 3 Post-processing modification 4 Grammar dictionary 5 Part-of-speech dictionary

Claims (1)

[Claims] 1. A method of: (a) creating a candidate word for each word of a character recognition result and calculating a certainty factor; and (b) for a word whose maximum certainty factor is smaller than a predetermined threshold value. Character recognition characterized by correcting a word to a grammatically correct word by using a grammatical relationship with a word whose maximum confidence value of a candidate word close to the word is larger than a predetermined threshold value Post-processing method of results.