JP2586372B2 - Information retrieval apparatus and information retrieval method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention searches character data.
Information retrieval apparatus and information retrieval method, especially document image
Information retrieval device and information retrieval method for character data obtained from
About the law.

[0002]

2. Description of the Related Art In recent years, with the increase in document information, the demand for an apparatus that digitizes and stores (files) a document image and retrieves and retrieves the document image has been increasing. In such an electronic file device, an information search device that enables a search based on the contents of a document of a document image using character recognition is desired.

Hereinafter, the above-described conventional information retrieval apparatus will be described with reference to the drawings. FIG. 5 is a flowchart of a conventional information retrieval apparatus using character recognition.

When registering a document, the document is input as document image data using an image input unit 1 such as a scanner and stored in a document image storage unit 2. Further, the character pattern portion included in the document image data is recognized using the character recognition means 3. The registrant corrects a portion where there is a plurality of candidate characters whose location is uncertain in character recognition using a keyboard or the like. Recognition result data storage means 6 associates the recognition result data with the document image data.
To accumulate.

When searching for a target document from the document registered as described above, a search condition such as a keyword is input from the keyboard 7 or the like, and recognition result data satisfying the condition is searched by the character search means 8 or the like. The document image data corresponding to the recognition result data is output.

However, in the above-described method, the above-described correction work is required after the character recognition at the time of document registration, which is troublesome.

If the above correction is not performed, an error may be included in the recognition result data. When a character string search for a search target keyword is performed, if the recognition result data including the keyword includes an error, Because they do not match, search omission occurs. In order to prevent search omission, a matching method that allows a mismatch of up to several characters when searching for a keyword, and a method of searching for a character string having a pattern shape similar to the search character string together with the search character string have been considered. As such a conventional example,
This is described in IEICE Technical Report CA87-25 (May 29, 1987), JP-A-4-158478, and the like.

[0008]

However, if a collation method that allows a mismatch of up to several characters to prevent omission of search, which is performed in the prior art, an inappropriate search result is generated (over-search). There is. For example, if you search for a search string (such as "free") using a search method that allows an error of up to one character, it is determined that it matches a completely different string (such as "reason" or "natural"). I will.

In the method of searching for a search character string and a character string having a similar shape together, the tendency of misrecognition differs from that expected in advance when, for example, a document image contains a character having a different typeface. Search omission occurs when characters are included.

In the prior art, when document recognition data to be searched does not include information such as a part where recognition is uncertain or a candidate at the time of recognition, a correction operation is performed using only this data. It is difficult. Information that is not included in the above-described document recognition data must be stored separately, and is difficult to manage.

[0011]

FIG. 1 is a block diagram showing the configuration of the present invention. As shown in FIG. 1, a first information retrieval apparatus that solves the above problem retrieves document recognition data including an arbitrary retrieval keyword from a document recognition data group obtained by character recognition of document image data. In the information retrieval apparatus, a document image data storage unit 10 for storing input document image data, and a character code of a character portion included in the document image data recognized as a candidate character code
Select a sentence <br/> character recognition means 11 for determining an estimate of the probability of the character codes, the likelihood of the character code of the estimated value
Sum is in the character code list that stores multiple character codes
Exceeds a certain probability set to contain accurate characters
Until the estimated value is higher,
The character code list added to the character code list
If there is only one candidate character code in
Select a character code that is a candidate for
If the number of codes is within the specified threshold,
With a recognition code indicating that there are multiple character codes
Multiple candidates including the character code of the first candidate
Select the character code to be used, and select the number of character codes in the list.
If the value exceeds the specified threshold,
Means for evaluating the likelihood of selecting an identification code indicating that there are a large number of codes, and document recognition data storage means 1 for storing these selected codes as document recognition data
3 and an arbitrary keyword input from the input means in the sentence
Characterized in that it comprises a retrieval unit 14 for retrieving the writing recognition data accumulation hand stage. In addition to the first information retrieval device, the second information retrieval device includes selection of a correct character code from a plurality of candidate character codes in the document recognition data of the document recognition data storage means, and a large number of the character codes. Correction means 15 for inputting a correct character code as an identification code indicating the above.

Here, the certainty evaluation means 12 enumerates and outputs a plurality of candidate character codes for characters whose recognition is uncertain. However, when the number of candidates is large, enumerating the candidates increases the document recognition data, and instead outputs only an identification code indicating that all characters are candidates (many candidates).

The information retrieval method according to the present invention provides a document image data
From the document recognition data group obtained by character recognition of
Information search to search for document recognition data including search keywords
Accumulates input document image data in search method
Document image data storage step , included in document image data
Recognize the character pattern of the character part to be
Select a code and obtain an estimate of the likelihood of the character code.
Character recognition step and character code certainty
The sum of fixed values is a character code library that stores multiple character codes.
Predetermined probability that exact characters are included in the strike
Character codes that are candidates in descending order of estimated value until
Is added to the character code list, and the character code list is added.
If there is only one candidate character code in the
Select a character code that is a candidate for the head, and select a sentence in the list
If the number of character codes is within the specified threshold,
And a recognition code indicating that there are multiple character codes
Multiple candidates including the character code of the first candidate
Select the character code of the character code in the list
If the number exceeds the specified threshold,
Select an identification code that indicates that there are many character codes.
The evaluation step of the chilliness and the selected code
Document recognition data storage step to store as document recognition data
And-up, any of the keywords you enter, the stored document
It has a search step for searching from recognition data.
Sign.

[0014]

The probability evaluation means 12 is a character recognition means 1
The certainty of the character code obtained at the time of character recognition by 1
The sum of the estimated values of
Specified to ensure that the correct characters are included in the
Until the probability is exceeded, the character
Code is added to the character code list, and the character code
When the number of candidate character codes in the list is one
Selects the character code to be the first candidate,
When the number of character codes inside is within the specified threshold
Is a recognition code indicating that there are multiple candidate character codes.
Multiple characters including the character code of the first candidate with
Select a character code that is a candidate for
If the number of codes exceeds the specified threshold,
Select an identification code indicating that there are many character codes
I do.

For this reason, when matching the document recognition data obtained as described above with the search target keyword for each character, the matching condition is determined by determining whether each character of the keyword is one of the candidates for the document recognition data. By assuming that they match, search omissions can be reduced. In addition, when recognition is possible with high certainty, such as when image conditions are good, candidates are not enumerated, so that the occurrence of unnecessary matches is reduced.

[0016] In addition, to evaluate the uncertainty of the character recognition, because you have selected to choose the candidate with the likely location of the error, extra candidate is removed, the amount of data of document recognition becomes excessive Is prevented.

Further, since the document recognition data includes information of a part and a candidate whose recognition is uncertain, it is possible to perform a correction work later using the information.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below with reference to the drawings, but the present invention is not limited to the following embodiments.

FIG. 2 is a block diagram showing the configuration of the embodiment of the present invention. In FIG. 2, a document is read as image data by image reading means such as a scanner 16,
The document image data is stored in the document image data storage unit 17. The character recognizing means 18 reads out the image data from the document image data storing means 17, identifies the area in which the character is written, recognizes the character pattern, estimates the candidate character code and its certainty, and sets the probability. Output by value.

In order to estimate the certainty, the coordinates of various characters in the space are determined in advance by assuming a vector space having several statistic axes obtained by processing patterns as axes. Then, the character pattern to be recognized at the time of character recognition is processed in the same manner to obtain the coordinates in the vector space, and the coordinates that were nearby during the learning are found and estimated from the distance between them.

The reliability evaluation means 19 follows an algorithm as shown in FIG. For each character, the most influential candidate among the candidates output by the character recognition means is placed at the top of the list. Then, based on the likelihood of recognition, a probability P that a correct character is included in the list is obtained. If the probability P is smaller than a certain threshold value Pth, the next most probable candidate is added to the list from the candidates, the probability P is obtained again, and the process is repeated. If the number of candidates (the length of the list) in the selected list is one, only this character string code is output.
If the number of candidates in the list is smaller than a certain threshold number Nth, a character code of an output candidate is output together with an identification code indicating that there are a plurality of candidates. If the number of output candidates exceeds Nth, an identification code indicating that all characters are candidates, that is, an identification code indicating a large number of candidates, is output instead of the character codes of the output candidates.

As a specific example of the identification code, a description based on a regular expression can be considered. For example, an expression listing {"B", "E", "3"} as candidates is "
[BE3] ”, and the unrecognizable expression is“. As a result of recognizing the pattern of NEC, N is clearly recognized, E candidates are {"B", "E", "3"}, and C
Is a large number of candidates, and if it is described by a regular expression, "N
[BE3] * ”.

The document recognition data accumulating means 20 accumulates, as character recognition data, a code string composed of a character code and an identification code output by the evaluation means 19.

The character string search means 21 compares and matches the search target keyword input from the input means such as the keyboard 24 with the document recognition data read out from the document recognition data storage means 20, and converts the document recognition data including the search target keyword. Search for.

FIG. 4 shows an example of this character string collation in which the text of the document recognition data and the keyword are compared one character at a time, and the character string of the keyword and the text match in all characters. , It is determined that a keyword match has been established. However, when there is an identification code indicating a plurality of candidates, if the character of the corresponding keyword is included in the candidates listed here, the character is regarded as matching. Also, if there is an identification code indicating a large number of candidates, this character is considered to match regardless of the character of the corresponding keyword.

In order to notify the search result, the document recognition data including the above-mentioned search character string or the document image data corresponding to the document recognition data is displayed on the display 22.
Display from

The document recognition data correcting means 23 displays the document recognition data on the display 22 and, for a place where a plurality of candidates are listed, gives a keyboard 24 to the user.
Is the right one. If the candidate is not specified, enter the correct character code on the keyboard 2.
Input from 4.

[0028]

According to the above embodiment, first, when a character pattern is recognized, the probability cannot be guaranteed with only one candidate, and a plurality of characters are selected so that the estimated value of the probability is sufficient. Since the characters are set as candidates or all characters are specified as candidates, it is possible to prevent the size of the text data as the recognition result from being excessively large.

Second, since candidates are listed in accordance with the probability of recognition for each recognized character, both search omission and over-searching can be reduced.

Third, even when the tendency of misrecognition is different, such as when a document image contains partial noise or a different font, retrieval can be performed without preparing data or the like indicating the tendency of misrecognition individually. .

Fourth, since the modification work can be performed later, the labor and time for the modification work before the search can be saved.

Fifth, when these corrections are performed by performing a morphological analysis or the like to select a candidate character, a process for selecting a candidate depending on whether or not this word is included in the dictionary is also performed. These processes can be postponed until there is room in the processing device.

Sixth, as long as the specifications for describing a plurality of candidates in the recognition result are unified, the character recognition means may be different. Recognition results created by different corresponding character recognition means can be searched by the same search means.

Seventh, even in the case of switching to higher-performance document recognition means in the future, the document recognition data accumulated so far can be continuously used.

[Brief description of the drawings]

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

FIG. 2 is a block diagram showing an embodiment of the present invention.

FIG. 3 is a flowchart showing an operation of an embodiment of the likelihood evaluation means.

FIG. 4 is a diagram illustrating an operation example of a character string search.

FIG. 5 is a block diagram showing a conventional technique.

[Explanation of symbols]

 16 Scanner 17 Document image data storage means 18 Character recognition means 19 Probability evaluation means 20 Document recognition data storage means 21 Character string search means 23 Correction means

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁶ Identification code Agency reference number FI Technical display location G06F 15/403 350C

Claims (3)

(57) [Claims] An information retrieval apparatus for retrieving document recognition data including an arbitrary search keyword from a document recognition data group obtained by character recognition of document image data, comprising: a document image storing input document image data; Data storage means, character recognition means for recognizing a character pattern of a character portion included in the document image data, selecting a candidate character code, and obtaining an estimated value of certainty of the character code, and character recognition means for determining the certainty of the character code. If the sum of the estimates is
The correct character is included in the character code list that stores the code.
The estimated value until a predetermined probability is determined to be exceeded.
Character codes that are candidates in the descending order of
Sentence that is a candidate in the character code list
If the number of character codes is one, the character code
Mode, and select the number of character codes in the list.
If it is within the specified threshold, the candidate character code
The first candidate with a recognition code indicating that there are
Select multiple candidate character codes including character codes
A threshold for determining the number of character codes in the list.
If the value exceeds the value, there are many candidate character codes.
And certainty evaluating means for selecting an identification code indicating the bets, the document recognition data storage means for storing these selected code as document recognition data, the document recognizes any keyword entered from the input means
An information search device comprising a search means for searching from a data storage means . 2. The document recognition data storage means according to claim 1, wherein :
Characters from multiple candidate character codes in data
Indicates the selection of a code or that there are many character codes
Correction means for inputting the correct character code for the identification code is provided.
2. The information retrieval apparatus according to claim 1, wherein the information is retrieved. 3. A sentence obtained by character recognition of document image data.
Documents containing arbitrary search keywords from the book recognition data group
In an information search method for searching for recognition data, a document image data storage for storing input document image data is provided.
Product step and the character pattern of the character part contained in the document image data.
And select a candidate character code, and confirm the character code.
The sum of the character recognition step for obtaining the estimated value of the kinshipness and the estimated value of the certainty of the character code is determined by a plurality of character codes.
The correct character is included in the character code list that stores the code.
The estimated value until a predetermined probability
Character codes in the character code list
Add a character code that is a candidate in the character code list.
If the number of characters is one, the character code of the first candidate
Is selected, and the number of character codes in the list is determined.
If it is within the threshold, there are multiple candidate character codes.
First sentence with a recognition code indicating that
Select multiple candidate character codes, including character codes,
A threshold value for the number of character codes in the list is determined.
If it exceeds, check that there are many candidate character codes.
Evaluation step of the certainty of selecting the identification code shown
And store these selected codes as document recognition data.
Storing the sentence recognition data and inputting any keyword into the stored document recognition data.
A search step for searching from the data
Information retrieval method.