JP2000057315A - Document filing apparatus and document filing method - Google Patents

Abstract

(57) [Summary] [PROBLEMS] In a document search in a filing system using character recognition, search omission and erroneous extraction are reduced.
The purpose is to perform a highly accurate search. SOLUTION: A character code created by a character recognizing unit at the time of document registration and a character feature created by a feature creating unit are stored. At the time of search, the part where the character codes do not match is compared with the feature. It becomes possible to search even a part where the number of characters of the keyword and the character recognition result are different due to the cut-out error.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for electronically filing images such as documents and drawings, and more particularly to a system for recognizing characters described in documents and drawings and storing the documents and drawings stored together with the images. The present invention relates to a document filing apparatus and a document filing method for performing full-text search using an input keyword.

[0002]

2. Description of the Related Art Conventionally, in order to electronically save, search and display a document image, a method of manually adding keyword information to a document image and saving the document image has been used.
Further, in order to save the trouble of manually inputting a keyword, a method of recognizing a character in a document image by a system having a character recognition function and storing a related keyword or a whole text together with the document image is used. In the latter case, erroneous recognition occurs due to imperfect character recognition performance. Therefore, for the keyword input for the search, “search noise” occurs in which a character string different from the input keyword is displayed as a search result. Further, even though the characters in the document image are the same as the input keyword, "search omission" may not be displayed as a search result due to erroneous character recognition.

In order to improve search accuracy, it is necessary to minimize search noise and search omission. Methods for reducing search noise and search omission at the time of search include a "method of retaining a plurality of character recognition result candidates" and a "method of making the search method ambiguous to assist incomplete character recognition performance". . As a “method of holding a plurality of candidates for character recognition results”, there is a method of holding character recognition results in a lattice structure, searching for character recognition results in a lattice structure, and searching for a character code to search for a correct character. ("Development of a full-text searchable document image database system" (8th Digital Library Workshop (Library and Information University, Oct. 23, 1996)).

The prior art 1 will be described with reference to FIGS. In FIG. 27, reference numeral 5 denotes input means for inputting a document image;
2 is character recognition means, 55 is search means, 56 is a recognition dictionary,
Reference numeral 51 denotes control means for controlling execution of character recognition, input of a document image, and control of a search using an input keyword.
Are display means for displaying search results for the input keyword;
Reference numeral 7 denotes a search data storage unit. FIG. 28 shows an example of a document image, and FIG. 29A shows the character image of FIG.
Reference numeral 5 indicates a character segmentation candidate point. FIG. 29B shows FIG.
The result of performing character recognition from the combination of the character extraction candidate points in (a) is shown.

In the prior art 1, when the character recognizing means 52 performs character recognition, a position sandwiched by black pixel connected components is set as a character extraction candidate point (candidate points 60 to 65 in FIG. 29A). Character recognition is performed on the image between the points, and it is determined whether or not to store the character based on the similarity of the recognition result candidate character, and the character candidate is retained or rejected. FIG. 29B shows an example of search data created by the character recognition means 52. For example, the character recognition result for the image sandwiched between the character extraction candidate points 61 and 62 in FIG. 29A is “SU” in FIG. 29B. The search means 55 searches the character string while following the search data of the lattice structure shown in FIG. 29B, and determines whether or not the character string matches the input keyword. Conventional technology 1
Aims to improve search performance by making the search data into a lattice structure and including a larger number of correct characters in response to a character segmentation error to improve the substantial recognition rate.

As a "method of assisting incomplete character recognition performance by making the retrieval method ambiguous", for example, Japanese Patent Laid-Open No.
As described in 272813, the calculation method of the matching degree between the input keyword and the recognition result is calculated as follows: m = the number of matched characters / the number of characters of the input keyword * 100 (%) ・ ・ ・ Calculated by formula (1) There are some which output as a search result even if all the search characters are not included in the candidate characters.

[0007] The prior art 2 will be described in JP-A-8-272813. FIG. 30 is a diagram showing the configuration of JP-A-8-272813. In FIG. 30, reference numeral 5 denotes an input means for inputting a document image;
Is a character recognition means for recognizing characters, 76 is a recognition dictionary storing standard patterns of characters used for character recognition, 7
5 is a search means, 71 is a control means for controlling the execution of character recognition or controlling the input of a document image and a search using an input keyword, 77 is a search data storage unit for storing images and recognition result data, 4 is a search data Display means for displaying the result.

First, a method of storing data will be described. The character recognizing means 72 cuts out and recognizes characters in the document image input from the input means 5, and recognizes up to four recognition result candidate characters for one character image.
Output to The control means 71 stores a character image and recognition result candidate characters, for example, four characters for each character in the search data storage 77.

Next, a search method will be described. FIG. 31 shows a part of the search data storage unit 77. Now, an arrow indicates a collation portion between the character recognition result and the input keyword when the input keyword for the search is “internal processing integrated type”.
The search means 75 collates with all the candidate characters up to the fourth place.
Now, when m is equal to or greater than a certain threshold value, for example, 60 (%) in Expression (1), this is set as a search result candidate. In this case, in FIG. 31, since the number of characters of the input keyword is collated with 6 characters for 7 characters, m = 6/7 * 100 = 85.7 (%), which is a search result candidate.

[0010]

In the case of the prior art 1, since the character cutout candidate point is determined at the break of the connected component of the black pixel, adjacent characters come into contact with each other, for example, "/ W" in FIG. If the characters are consecutive, character extraction as one character cannot be performed, and as a result, the character is erroneously recognized, resulting in a problem that search is omitted. Also,
Even if character extraction is successful, if a correct result is not output in character recognition, search omission occurs as described above.

[0011] Further, in the prior art 2, no matter what part of the character to be compared with the character to be matched with the input keyword is the same degree of matching if the matching part is common. was there. Thus, for example, if the input keyword is "Japanese", the character strings "Japanese", "Japanese", "Japan", "Japan", "Japan" are all m = 2/3 * 100 = 67% (from equation (1)) has the same degree of match, and is output and displayed as a search result.

[0012] Here, in the case of "Japanese entry", "in" is erroneously recognized, and when it is actually "Japanese", the above "Japanese", "Japan", "Japan", "Japan is" Since the degree of coincidence is the same as that of the others, when displayed in descending order of the degree of coincidence,
Is buried in these and displayed. The user needs to search for more desired results from such search noises displayed by the display means 4, and the smaller the threshold for permitting the inconsistency, the more search noises are output, so the user really wants to search. There is a problem that the document is buried in the search noise, and as a result, it is difficult for the user to use the document. In addition, there is a problem that an increase in the threshold increases search omission. The present invention has been made to solve these problems.

[0013]

According to a first aspect of the present invention, there is provided a document filing apparatus comprising: an input unit for inputting a document image; a recognition dictionary in which standard patterns of respective characters are stored in advance; and a document input by the input unit. A character recognizing unit that extracts characters from an image, recognizes the extracted characters with reference to the recognition dictionary, and creates a character code, a feature creating unit that creates a feature for each character recognized by the character recognizing unit, A search data storage unit for storing the character codes created by the character recognition unit and the features created by the feature creation unit; a feature dictionary for holding the features of the standard pattern in advance; Search feature creation means for acquiring a feature of a character from the feature dictionary; and collating and searching the input keyword against data in the search data storage unit. When the predetermined condition comprises search means for outputting reference to processing results of the search feature creation means, and display means for displaying the search result of the search means.

[0014] In the document filing apparatus of the second aspect, the feature creating means may include: a vertical, horizontal, and right-upward portion of a character outline in each character rectangle from which a character is cut out when the character recognizing means performs character recognition; The configuration is such that a four-way component feature falling to the right is created.

According to a third aspect of the present invention, in the document filing apparatus, the search unit matches the character code of the input keyword with the character code of the search data storage unit in matching the input keyword with data in the search data storage unit. For the part, the distance between the character codes is calculated, and for the part where the character codes do not match, the characteristic of the character in the search data storage unit and the characteristic in the characteristic dictionary are compared to calculate the distance. The search result is determined based on the distance of the character code and the distance of the feature.

In the document filing apparatus according to a fourth aspect of the present invention, the search means, when comparing the input keyword with the search data storage unit, if the ratio of the number of characters matching the input keyword is equal to or more than a predetermined value, The feature of the character in the search data storage unit and the feature in the feature dictionary at a portion where codes do not match are compared, and the degree of matching is calculated.

In the document filing apparatus according to a fifth aspect, the feature creation unit tests the character code recognized by the character recognition unit using a predetermined standard, and determines that each character recognition result is a correct character. For a character that has been created, only the character code output by the character recognition unit is stored without creating a feature, and if it cannot be determined as a correct character, the character code output by the character recognition unit and the feature created by the feature creation unit Is saved.

In the document filing apparatus according to a sixth aspect of the present invention, the search means determines that a portion of the search data storage section in which the search data holds only the character code determines a match from a distance based on only the character code, and determines the character code and the characteristic. The retained part is configured to calculate the degree of coincidence from the character code and the feature.

[0019] In the document filing apparatus according to claim 7, the character recognizing means determines whether the document is written vertically or horizontally, and stores the result in the search data storage.
When the number of characters of the character string of the search data storage unit is different from the number of characters of the input keyword for matching the feature, whether the data in the search data storage unit is in vertical writing,
The feature is configured to be re-created in accordance with a predetermined criterion based on the information indicating whether the writing is horizontal.

In the document filing apparatus according to the present invention, when the number of characters of the character string in the search data storage unit differs from that of the input keyword, the search unit may use the dynamic programming method to search for the input keyword and the corresponding character in the search data storage unit. The configuration is such that features are compared.

According to a ninth aspect of the present invention, the character recognizing means determines whether the document is written vertically or horizontally, and stores the result in the search data storage.
The feature creator has a feature creation method corresponding to vertical writing and horizontal writing, and uses the corresponding feature creation method based on a result of determining whether the character recognizing means is vertical writing or horizontal writing. It is configured to create features.

According to a tenth aspect of the present invention, there is provided a document filing apparatus.
The feature creation means includes a plurality of different feature creation methods, and is configured to select a corresponding feature creation method according to the type of the character of the input keyword.

[0023] In the document filing apparatus according to the eleventh aspect,
When the input keyword is an alphabetic character or a symbol, the feature creating means creates an integrated feature by partially overlapping features of adjacent characters constituting the input keyword.

According to a twelfth aspect of the present invention, there is provided a document filing apparatus.
The character recognizing unit performs character segmentation at the time of character recognition, stores rectangular information for each character in the search data storage unit, and forms a rectangular shape of each character of the input keyword output by the search feature creating unit. Based on the information of the character rectangle obtained from the search data storage unit and the number of characters to be compared among the characters of the input keyword, it is determined whether or not the feature is to be compared. And a feature collation judging means for judging that the character string does not match the input keyword.

In the document filing method of claim 13,
An input step of inputting a document image, and characters are cut out from the document image input in the input step, and the cut-out characters are recognized by referring to a recognition dictionary in which a standard pattern of each character is stored in advance to create a character code. A character recognition step, a feature creation step of creating a feature for each character recognized by the character recognition step, and a character code created by the character recognition step and a feature created by the feature creation step stored in a search data storage unit. A search data step, a search feature creating step of acquiring a feature of each character of a search input keyword input at the time of search from a feature dictionary in which features of a standard pattern are held in advance, and The result of the search feature creation step is compared with the result of the search feature creation step under a predetermined condition. Comprising a search step of referring to processes output, and a display step of displaying the search result of the search step.

[0026]

Embodiment 1 Embodiment 1 of the present invention will be described below. First, a document registration method will be described with reference to FIGS. FIG. 1 is a block diagram according to Embodiment 1 of the present invention. In FIG. 1, reference numeral 5 denotes input means for digitizing an image of a paper document by photoelectric conversion using a scanner at the time of document registration, or inputting a photoelectrically converted image via a network or the like, and reference numeral 9 denotes recognition used for character recognition. The dictionary 2 extracts characters from the image given by the input means 5, determines vertical writing or horizontal writing, further recognizes the extracted characters with reference to the recognition dictionary 9, and outputs a character code. Means 3 are feature creating means for creating a feature for each character rectangle for which the character recognizing means 2 has performed character recognition.

Reference numeral 10 denotes a search data storage unit for storing the character codes, characteristics, and types of vertical writing and horizontal writing created by the character recognizing means 2 and the feature creating means 3, 4 a display means for displaying search results and document images, 6 Is a retrieval means for retrieving a character portion in an image corresponding to an input keyword input by a user at the time of retrieval from the retrieval data storage unit 10; Feature collation judging means for judging whether or not to perform, 1
Reference numeral 1 denotes a feature dictionary that holds the features of the standard pattern in advance, 8 denotes a feature dictionary 11 that stores the features of each character of the input keyword that has been input.
A search feature creating means for processing the features of each read character; a document image storing means for storing the digitized document image from the input means; This is control means for managing registration and search processing.

FIG. 2A shows an example of a document image, and FIG. 2B shows the result of character recognition of the document image of FIG. 2A. FIG. 3 is an explanatory diagram of an area dividing method in a case where a feature is created from a result extracted for each character in the character recognition processing. FIG. 4 is an example of a mask used to create a directional component feature. 6 is a flowchart of a document registration process, and FIG. 7 is a flowchart of search data creation (the process of step S102 shown in FIG. 6). First, a document registration method will be described with reference to the flowcharts of FIGS.

In step S101 of FIG. 6, a document image is inputted by the input means 5. In order to realize the input unit 5, a paper document may be digitized by photoelectric conversion using a scanner, or an image that has been photoelectrically converted in advance may be input via a network or the like. FIG. 2A shows an example of an input document image. The image input by the input unit 5 is a binary image in which each pixel value takes a value of 1 (black) or 0 (white). Next, in step S102, data to be stored in the search data storage unit 10 is created. Here, the control means 1 passes the input image to the character recognition means 2 and starts character recognition.
Then, the process proceeds to step S103, and the character code, feature, vertical writing, and horizontal writing type created in step S102 are stored in the search data storage unit 10.

Next, details of the search data creation in step S102 by the character recognition means 2 and the feature creation means 3 will be described in accordance with the flow of processing shown in FIG. First, FIG.
In step S201, the character recognition unit 2 extracts a character area from the input image. The method of extracting a character region is, for example, to connect regions where black pixels are continuous in a document image, determine whether or not the character is a character string from the width and height values of the connected components of the black pixels, These are combined into one region.

Next, at step S202 in FIG.
Determines vertical writing and horizontal writing for each area. The determination method uses a known method, for example, from the arrangement of character strings in the area,
The width and height of each character string are determined, and an area where many vertically long character strings exist is determined as vertical writing, and an area where many horizontally long character strings exist is determined as horizontal writing. Next, in step S203, the character recognition means 2 executes character recognition. Here, characters are cut out using a known technique and character recognition is performed, and one character or a plurality of candidate characters are created for one character image.

In the character extracting method, for example, each character string image in the area determined in S201 is scanned in the vertical direction and the horizontal direction, and the peripheral distribution of the number of black pixels is obtained. Divide into images for each character. The character recognition process counts the number of black pixels in each small area of, for example, 8 × 8 dimensions for an image divided into single characters by character segmentation, and calculates the sum of absolute values of differences for each dimension with respect to a standard pattern. Then, a character having a standard pattern with a small sum of the absolute values of the differences is extracted and output as a recognition result.

Next, the process proceeds to step S204 in FIG. 7, and the feature creating means 3 creates a feature from each character. Here, FIG.
As shown in FIG.
The image is divided, and four-direction component features (horizontal, vertical, upward-sloping, downward-sloping features) of the edge of the character image are extracted for each region.
In the method of creating the four-direction component feature, the mask shown in FIG. 4 is scanned in each of the eight divided areas as shown in FIG. 3, and a bit AND of the image and the mask is obtained. When the result is the same as that of the mask, the direction component of the mask is increased by one. FIG. 5 shows an example in which features are created in this manner. FIG. 5 shows character codes created by judging and recognizing vertical writing and horizontal writing, and the characteristics of each character. In FIG. 5, “horizontal” is a horizontal component feature, “vertical” is a vertical component feature, “upper right” is a right-up component component, and “lower right” is a right-down component feature.

Next, as described above, the process proceeds to step S103 in FIG. 6, and the control means 1 searches the character codes, characteristics, and types of vertical writing and horizontal writing created by the character recognizing means 2 and the feature creating means 3 shown in FIG. The data is stored in the data storage unit 10.

Next, the operation at the time of retrieval will be described with reference to FIGS. FIG. 8 is a flowchart of the search,
9 and 10 are diagrams for explaining the search operation. First,
The search operation will be described based on the flowchart of FIG. First, in step S301 in FIG. 8, the search unit 6 checks the input keyword input by the user with the character code data in the search data storage unit 10. The search unit 6 searches the search data storage unit 10 and, if a character that matches the input keyword exists, holds a number indicating the storage position of the character in a buffer.

In FIG. 9, reference numeral 21 denotes a number indicating the position of a character in the search data. First, data number 1 in FIG.
To match. In FIG. 9, reference numeral 20 denotes the character number of the character whose input keyword matches the search data. Now, data number 1
When the search data is compared with the input keyword, the character number 20 whose character code matches each character of the input keyword is stored in a buffer (not shown). The character number 1 is associated with the character “sentence” of the input keyword, 3 is associated with “recognition”, and 4 is associated with “sense”.
FIG. 10 shows a comparison with data number 2 in FIG. 5, and 24 in FIG. 10 shows the character number of the character whose input keyword matches the search data of data number 2.

Next, in step S302, the search means 6 calculates a search candidate area. Here, the character number of the character that matches the input keyword is tested to determine whether or not to be a candidate area. The condition for determining is that the characters that match the search data account for 30% or more of all the characters that make up the input keyword, the character numbers are arranged in the order in which the input keywords appear, and the matching character numbers Is considered as a candidate area when they are close to each other. 9 is the candidate area, and in the example of FIG. 10, 24 is the candidate area.

Next, proceeding to step S303, the search means 6 determines whether or not to match features between the search data and the input keyword. The result of 20 in FIG. 9 satisfies the above condition, and the character of the unmatched input keyword character “character” is compared with the feature of the search data “u”. Input keyword "letter"
The search feature creating means 8 reads the feature corresponding to the character from the feature dictionary 11 and loads the feature of the character "character" into a buffer (not shown). In addition, the search means 6 similarly loads a search feature “U” into a buffer (not shown). Examples of loaded buffers (not shown) are shown at 22 and 23 in FIG.

If the result of the determination in step S303 is that the feature between the search data and the input keyword is to be compared, the process proceeds to step S304, where the search means 6 calculates the distance between the feature between the search data and the input keyword. . How to calculate the distance between features

[0040]

(Equation 1)

It is assumed that However, Fdic is the dictionary feature value, Fi
mg is the feature value of the search data, I is the number of direction components, and J is the number of features for each direction component, where I = 4 and J = 8. Also, 1 ≦ i ≦ I and 1 ≦ j ≦ J. Now, when D is calculated in the example of FIG. 9, D1 [dic, img] = 12. In the example shown in FIG. 10, D2 [dic, img] = 49.

Next, proceeding to step S305, the search means 6 determines whether or not to be a keyword candidate based on the entire distance between the input keyword and the search data. Now, the distance between the character in the input keyword and the character in which the character code data of the search data is matched is set to 0, and the total distance is calculated by Dist = ΣD / the number of characters in the input keyword. In the example of FIG. 9, Dist1 = 12/4 = 3, and output is performed as a candidate when is smaller than a predetermined value A.
In the example, Dist2 = 49/4 = 12. For example, when rejecting with A = 10, the example in FIG. 10 is rejected from the candidates. Also, without rejecting with a threshold, if the “sentence recognition” having a different shape from the input keyword has a larger distance than “sentence recognition” closer to the input keyword, and the candidates are sorted and displayed in ascending order of distance, “Sentence recognition” close to the input keyword is displayed closer to the correct candidate than “sentence recognition”, so that the user can reduce the trouble of finding the correct answer from the candidates and improve the usability. If the result of the determination in step S303 indicates that the matching between the search data and the input keyword is not to be performed, the process proceeds to step S304 and step S3.
Skip 05 to end.

In the first embodiment, the feature to be created is a four-direction component feature. However, the feature is not limited to this feature, and may be another feature, for example, a mesh feature or a histogram feature. Also,
As the feature created by the feature creating unit 3, the feature used by the character recognizing unit 2 for character recognition may be used. Further, the method of calculating the distance of the feature and the method of calculating the distance to the entire input keyword are not limited to those described above. Further, the value of A in the first embodiment is not limited to this.

In the first embodiment, a feature is created and stored for all recognition results. However, the present invention is not limited to this. For example, the evaluation value of character recognition is extremely high, and it is determined that the recognition result is definitely correct. If possible, the storage for the search data can be reduced by omitting the storage of the characteristic value of the character code. For example, document registration is executed using the flowchart shown in FIG. 11, and in step S205, the feature creating unit 3 creates a feature for a character whose similarity in character recognition is equal to or less than a certain value. FIG. 12 shows an example of the result. In FIG. 12, features are created for the characters “U”, “N”, and “Ken”. At the time of the search, the search means 6 checks only the character code having no feature in the search data storage unit 10 using the mathematical formula (4).
The calculation is performed using Expression (4) and Expression (2).

[0045]

(Equation 2)

The distance in the entire candidate area is calculated by Dist = (ΣD + ΣC) / the number of characters of the input keyword (Equation (3)), and it is determined whether the distance matches the input keyword.

As described above, in the first embodiment, even if the character recognition results do not match due to erroneous recognition,
By comparing the characteristics of the characters that do not match and performing a search using the same, it is possible to select a correct candidate from an incorrect candidate based on the similarity.

Embodiment 2 Next, a search method in the case where the number of characters of the input keyword differs from that of the search data due to a character extraction error or the like will be described with reference to FIGS.
This will be described with reference to FIG. Now, as shown in FIG. 13A, "J" and "E" are erroneously cut out as one character due to a character cut-out error of the character recognizing means 2, and the search data shown in FIG. 13B is output. If the input keyword "REJEC
An example of searching using "T" will be described.

First, in step S301 in FIG. 8, collation between the input keyword and the search data by the character code is performed.
Here, input keywords and search data "R""E"
"C" and "T" match. In step S302, a candidate area is calculated. Here, since the character arrangement and the order of “R”, “E”, “C”, and “T” are correct, the area is set as a search candidate area. Next step
In S303, it is determined whether or not matching between features is performed. here,
Since the number of matched characters is 4/6 = 66.7% of the input keyword, which is 30% or more, feature matching is performed.

An example will be described in which the search data "work" is collated with the input keyword "JE". The search feature creating means 8 reads the features of “J” and “E” from the feature dictionary 11 to a buffer (not shown). FIG. 14 shows the features of “J” and “E”. In FIG. 14, portions surrounded by [] are the same direction component features, and are arranged in the order of the region numbers shown in FIG. Next, the feature of “work” and the feature of “JE” are collated, but the number of characters is different, so the number of features is also different. For this reason, the method for obtaining the difference as in the first embodiment cannot be used. In this case, matching is performed by a generally well-known dynamic programming method (DP matching). Here, since the search data is written horizontally, the upper and lower two areas are integrated as shown at 15 and 16 in FIG. 18 and collation directed to the arrow is performed by dynamic programming. In FIG. 18, reference numeral 15 denotes the feature of the input keyword or the feature of the search data (the feature of the input keyword "JE" in this example), and 16 denotes the feature of the search data "work" in this example. is there.

The search feature creation means 8 creates the features "J" and "E" by connecting them in the horizontal direction for each component. Here, the horizontal, vertical, upper right, and lower right direction components are defined as a region “1”.
It is divided into “2”, “3” and “4” and areas “5”, “6”, “7” and “8”, and each direction component of “J” and “E” is connected. In FIG. 15, the horizontal (upper) is obtained by connecting the horizontal components of the areas “1”, “2”, “3”, and “4” in FIG. 17 by connecting the feature dictionaries in the order of “J” and “E”. Horizontal (bottom) refers to areas “5” and “6” in FIG.
The horizontal components of “7” and “8” are arranged by connecting the feature dictionaries in the order of “J” and “E”. FIG. 16 shows the search data “work” of FIG. 13 rearranged by the search feature creating means 8 in the same manner as in FIG. Now, the distance between certain component features

[0052]

(Equation 3)

Assume that Here, FD is a character feature in the input keyword, FI is a character feature in the search data, and n = 1 is FIG.
5, the horizontal (upper) component is shown in FIG. 16, where n = 2 is horizontal (lower), n = 3 is vertical (upper), n = 4 is vertical (lower), n = 5 is upper right (upper), n = 6 indicates upper right (lower), n = 7 indicates lower right (upper), and n = 8 indicates lower right (lower). Here, 1 ≦ i ≦ I, I = 8, and 1 ≦ j ≦ J, and J = 4. For example, i = 2, j
When = 1, FDni indicates the character feature included in the portion 13 indicated by the dotted line in FIG. 15, that is, the area of the portion indicated by "2" and "6" in FIG. 17, and Finj indicates the portion 14 indicated by the dotted line in FIG. The distance between the character feature of the area indicated by “1” and “5” in FIG. 17 is calculated using Expression (5). At this time

[0054]

(Equation 4)

Dist [dic, img] = Ddp (I, J) / I The distance dist between each feature is calculated as Expression (7). Here, FIG.
5. When actually calculated in the example of FIG. 16, dist [dic, img] = 25
/ 8 = 3. Next, in step S305, the distance between the input keyword and the entire candidate area in the search data is calculated. Formula (3)
Therefore, Dist = 3/6 = 0.5. As described above, when the number of features is different, the features can be re-created and searched by using DP matching.

In this example, since the search data is written horizontally, two vertical regions are integrated and the characteristics of a plurality of characters are connected horizontally. However, when the search data is written vertically, 17 in FIG.
Then, the four horizontal areas are integrated as shown in (18) and (18), and collation is performed toward the arrow (vertical) by dynamic programming. Search feature creation means 8
Links features vertically. How to calculate the distance between features

[0057]

(Equation 5)

Equations (6), (7) and (3) are used. In Expression (5) ■, n is 16 in order to compare four directional components for four regions.

In the second embodiment, when the number of features of the input keyword differs from that of the search data, the matching is performed by the dynamic programming. However, the present invention is not limited to this. It is also possible to match the numbers and perform the matching as in the first embodiment.

Third Embodiment Next, an example in which even if a character in the search data storage unit 10 is divided in the middle, the character can be searched by recreating the character feature is shown in FIGS. 8, 20 to 24. This will be described with reference to FIG. For the character image of FIG.
, And a feature is created by the feature creating means 3 as shown in FIG. Here, a case of searching for the input keyword “S / W” from the search data shown in FIG. 21 will be described.

In step S301 of FIG. 8, the input keyword is collated with the character code of the search data. Now, only the character code "S" matches the character code of FIG. 21 with the character code of the input keyword "S / W". Next, step S302 in FIG.
To calculate a candidate area, and in step S303, the search means 6 determines whether or not to collate the feature. Now, since the ratio of the matching character to the entire input keyword is 1/3 = 33.3%, it is set as the matching candidate. Next, in step S304, the feature of the character code that does not match is collated. The feature of the search data “NO N” in FIG. 21 and the feature of “/ W” taken into the buffer from the feature dictionary 11 are collated in the same manner as in the first embodiment, and D
(dic, img) = 23. Here, if the input keyword is an alphabetic character, the character often contacts, and the feature is recreated correspondingly.

The method of re-creation will be described with reference to FIGS. If the input keyword is an alphabetic character, the search feature creating means 8 recreates the feature in response to the contact of the character. At this time, if the character next to the character shown in FIG. 22 is an alphabetic character, the feature corresponding to the contact is recreated. Here, since the input keyword includes “/”, the features of “/” and “W” are recreated. The standard pattern is held in the feature dictionary 11, and is created by virtually dividing the standard pattern as follows. The area (領域 from arrow 30 to arrow 31) which is 1/4 from the left of the part shown by a solid line in “W” in FIG.
The region 36) sandwiched between the right and left of the feature indicated by “/” in FIG.
4 (37 in FIG. 23). After that, the remaining image of "W" (the area between the arrows 31 and 32) is divided into four equal parts by the dotted lines indicated by 33 to 35, and each direction component feature is recreated. The results of the recreated features are shown in FIG. Next, the re-created feature is compared with the feature of the search data. That is, D = 14 is obtained by comparing the feature of FIG. 21 with the feature of FIG.

The value is smaller than the distance before re-creation of the feature, and it can be seen that the shape is closer to the shape of the image. As a result, the distance from the search data for the English input keyword is reduced, and search omission is less likely to occur.

In this embodiment, an example has been described in which a standard pattern is held and then recalculated to create a feature. However, the present invention is not limited to this, and the feature of the standard pattern of each character is created when the feature dictionary 11 is created. It is also possible to recreate the feature by integrating the components adjacent to the feature when the search feature creating means 8 loads the feature of the character to be matched from the feature dictionary 11 into the buffer. It is possible. For example, the feature of the standard pattern in which the directional component feature in each area is created is stored in the feature dictionary 11 by dividing the horizontal component into sixteen instead of four in the horizontal direction. When the search feature creating means 8 creates a feature, the feature is recreated by integrating the four adjacent areas except for the alphabet and the symbol, and the alphabet and the symbol are integrated with the left quarter of the area and the character adjacent to the left. , The remaining 3/4 to 4
Since the division, that is, the area of 12/16 is divided into four equal parts, the features are integrated by 3/16 (three adjacent features) in the third embodiment.
Can be created.

Also, since English characters are hardly adopted in the vertical writing, the accuracy of the characteristics is changed between the vertical writing and the horizontal writing when the feature dictionary 11 is created, and the horizontal writing is divided more finely to create the features. It is also possible to create with coarse features. Similarly, the capacity of the feature dictionary 11 can be reduced.

Embodiment 4 A method of reducing processing time and suppressing search noise by determining whether a candidate is a correct candidate before performing feature matching will be described with reference to FIGS. 25 and 26. Now, FIG.
Search data of "REACT" shown in Fig. 5 and the input keyword "RES
PECT ”will be described. S described in FIG.
x, sy, w, and h are sx, sy, the x and y coordinates of the upper left point of each character rectangle in the search data, w is the width of the rectangle,
h is the height of the rectangle.

First, in S301 and S302 in FIG. 8, each character code is collated, and a matching part is created in a buffer (not shown). Here, they match "R", "E", "C", and "T". Next, in S303 of FIG. 8, the search means 6 determines whether or not to perform feature matching. 4/7 = 57% match with input keyword,
Since the order relation and the adjacency relation are satisfied, it is set as a candidate area. Since "A" of the search data does not match "SPE" of the input keyword, the feature matching determination means 7 determines whether or not to match features of each character. Here, the feature collation determination means 7 reads w and h shown in FIG. 25 and obtains the character rectangular shape of “A”. Here, h / w = 1.0.

The search feature creating means 8 estimates rectangle information from the table of FIG. 26 for each character of the input keywords "S", "P" and "E". Here, each of the characters “S”, “P” and “E” belongs to the others 43. Therefore, the feature matching determination means 7 calculates a rectangular shape when “S”, “P”, and “E” are connected. Now, since the character height of the search data is 60, "S""P"
The rectangular shape connecting "E" is 60 x 0.7 x 3 = 126 to 60
× 1.2 × 3 = 216. Character width of search data "A" =
60 and the possible value of the connection width of "SPE" is 126 to 216, so if the difference between the widths of the characters matching the input keyword and the features of the search data is more than twice as large as one, the feature matching is performed. If you add a condition to remove it from the list without doing it,
"A" and "SPE" can be excluded from the candidate without matching.

In the case where there is a certain difference in the width of the feature to be compared as described above, it is possible to avoid the apparently unmatched matching by regarding the feature as not to be compared and assuming that the feature does not match. In this case, the search means 6 does not perform feature comparison in S303, similarly does not execute S304, and does not make it a candidate in S305.

In the fourth embodiment, whether the input keyword is compared with the search data is determined based on the rectangular width of the feature. However, the present invention is not limited to this. For example, the number of characters to be compared with the input keyword and the feature in the search data is determined. If the difference between them is 2 or more, the matching may not be performed.

In the fourth embodiment, when comparing features, it is possible to eliminate useless matching by determining whether or not to perform feature matching using an input keyword and a character rectangle of a character recognition result. As a result, processing time is shortened,
Search accuracy can be improved.

[0072]

As described above, according to the present invention,
In claims 1 to 3 and claim 13, character codes and features are stored, and a search is performed based on the character codes and features at the time of search. Distance can be given.

According to the fourth aspect of the present invention, the feature matching is performed in a region where the character code matching ratio is equal to or more than a certain value, so that an increase in search noise can be suppressed and the processing time can be shortened.

According to a fifth aspect of the present invention, when the character recognition result recognized by the character recognition means is a correct character, no character is created and only the character code is stored. Since the feature created by the means is stored, it is possible to reduce the storage capacity of the search data storage unit.

According to a sixth aspect of the present invention, the portion where the search data holds only the character code determines the match from the distance based on only the character code, and the portion holding the character code and the feature calculates the degree of matching from the character code and the feature. It has been
Search noise can be reduced.

According to the seventh and eighth aspects, when the number of characters of the character string in the search data storage unit differs from that of the input keyword, the feature is re-created according to a predetermined criterion. Recognition data can also be searched with an appropriate distance.

According to the ninth aspect, a feature is created by using a corresponding feature creation method based on a determination result of vertical writing or horizontal writing. Therefore, character extraction is performed in both vertical writing and horizontal writing. Searches that deal with errors are possible.

According to a tenth aspect, a feature creation method corresponding to each of the types of characters of the input keyword is selected. In an eleventh aspect, when the input keyword is an alphabetic character or a symbol, adjacent ones constituting the input keyword are arranged. Since the integrated feature is created by partially overlapping the features of the characters, it is possible to perform a search in response to erroneous recognition due to contact between adjacent characters that is likely to occur in English characters.

According to the twelfth aspect, since it is configured to avoid collation between character strings that are clearly different using the rectangular information and the number-of-characters information, it is possible to reduce search noise and shorten processing time.

[0080]

[Brief description of the drawings]

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

FIG. 2 is an explanatory diagram of a registration image according to the first embodiment.

FIG. 3 is an explanatory diagram of a character cutout result and a feature creation area.

FIG. 4 is an explanatory diagram of a mask for creating a four-direction component feature.

FIG. 5 is an explanatory diagram of contents of search data.

FIG. 6 is a flowchart of a registration process.

FIG. 7 is a flowchart of character recognition and feature creation.

FIG. 8 is a flowchart of a search.

FIG. 9 illustrates a search operation according to the first embodiment.

FIG. 10 illustrates a search operation in Embodiment 1.

FIG. 11 is a flowchart of a modification of the registration process.

FIG. 12 is an explanatory diagram of search data by a deformation registration process.

FIG. 13 is an explanatory diagram of search data used in the second embodiment.

FIG. 14 is an explanatory diagram of a feature dictionary of “J” and “E”.

FIG. 15 is an explanatory diagram showing an example in which features of “J” and “E” are recreated.

FIG. 16 is an explanatory diagram showing an example in which the feature of “work” is re-created.

FIG. 17 is an explanatory diagram showing area numbers.

FIG. 18 is an explanatory diagram showing a collation method in the case of horizontal writing in the second embodiment.

FIG. 19 is an explanatory diagram showing a matching method in the case of vertical writing in the second embodiment.

FIG. 20 is an explanatory diagram showing an example of a registered document according to the third embodiment.

FIG. 21 is an explanatory diagram showing contents of search data according to the third embodiment.

FIG. 22 is an explanatory diagram showing characters for recreating a feature.

FIG. 23 is an explanatory diagram showing a feature re-creation method.

FIG. 24 is an explanatory diagram of a recreated feature dictionary.

FIG. 25 is an explanatory diagram showing contents of search data according to the fourth embodiment.

FIG. 26 is an explanatory diagram showing a character code-shape determination table.

FIG. 27 is a block diagram of a conventional technique 1;

FIG. 28 is an explanatory diagram showing a document image used in the related art 1.

FIG. 29 is an explanatory diagram showing an example of a character extraction candidate point and character storage according to the related art 1.

FIG. 30 is a block diagram of a conventional technique 2.

FIG. 31 is an explanatory diagram showing an image / recognition result database according to the related art 2.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 control means, 2 character recognition means, 3 feature creation means, 4 display means 5 input means, 6 search means, 7 feature collation determination means, 8 search feature creation means, 9 recognition dictionary, 10 search data storage section, 11 feature dictionary 12. Document image storage means.

Continued on the front page F term (reference) 5B050 BA10 BA16 EA03 EA07 GA08 5B075 ND07 NK02 NK07 NK13 NK24 NK31 PP02 PP12 PP22 PR06 PR10 QM02 QM08 UU06

Claims (13)

[Claims] An input means for inputting a document image, a recognition dictionary in which standard patterns of respective characters are stored in advance, and characters are cut out from the document image input by the input means.
Character recognition means for recognizing the cut-out character with reference to the recognition dictionary and generating a character code; feature generation means for generating a feature for each character recognized by the character recognition means; A search data storage unit for storing a character code and a feature created by the feature creating unit; a feature dictionary for holding features of the standard pattern in advance; and a feature of each character of a search input keyword input during a search. A search feature creating unit that obtains from a dictionary, and performs a search by collating the input keyword with data in the search data storage unit and performing a search by referring to the result of the search feature creating unit when a predetermined condition is satisfied. A document filing apparatus comprising: a search unit that performs search; and a display unit that displays a search result of the search unit. 2. The method according to claim 2, wherein the character recognition unit performs four-directional component vertical, horizontal, upward-sloping, and downward-sloping component outlines of the character in each character rectangle from which the character is cut out when performing character recognition. 2. The document filing apparatus according to claim 1, wherein the document filing apparatus is configured to create a document. 3. The search means according to claim 1, wherein, in comparing the input keyword with data in the search data storage unit, a character code corresponding to a part where the input keyword matches the character code of the search data storage unit is used. Calculate the distance between each other, and in the portion where the character codes do not match, compare the characteristics of the character in the search data storage unit with the characteristics in the feature dictionary and calculate the distance to calculate the distance. 2. The document filing apparatus according to claim 1, wherein a search result is determined based on a feature distance. 4. The method according to claim 1, wherein the search unit is configured to determine, when comparing the input keyword with the search data storage unit, that a character code does not match when a ratio of the number of characters matching the input keyword is a predetermined value or more. 4. The document filing apparatus according to claim 3, wherein a feature of the character in the search data storage unit is compared with a feature in the feature dictionary, and a matching degree is calculated. 5. The feature creating means tests the character code recognized by the character recognizing means using a predetermined criterion, and applies a feature to a character whose individual character recognition result is determined to be a correct character. The character code output by the character recognizing unit is saved without creating the character code, and the character code output by the character recognizing unit and the feature created by the feature creating unit are saved when it cannot be determined that the character is the correct character. The document filing apparatus according to claim 1, wherein: 6. The search means determines that a portion of the search data storage unit in which search data holds only a character code determines a match based on a distance based on only a character code, and a portion that holds a character code and a feature uses a character code. 6. The document filing apparatus according to claim 5, wherein the degree of coincidence is calculated from the feature. 7. The character recognizing means determines whether the document is written vertically or horizontally, and saves the result in the search data storage section. The search feature creating means includes an input keyword for matching features. When the number of characters in the character string in the search data storage unit differs from that in the search data storage unit, the feature is re-created based on information on whether the data in the search data storage unit is vertical writing or horizontal writing according to a predetermined standard. 2. The document filing apparatus according to claim 1, wherein 8. When the number of characters of a character string in the search data storage unit differs from the number of characters in the input keyword, the search unit performs matching between the features of the input keyword and the characters in the search data storage unit by dynamic programming. The document filing apparatus according to any one of claims 1 to 3, wherein: 9. The character recognizing means determines whether the document is written vertically or horizontally and saves the result in the search data storage unit, and the feature creating means corresponds to the vertical writing and the horizontal writing. Claims: A feature is provided in which each of the feature creating methods is provided, and a feature is created using a corresponding feature creating method based on a result of determining whether the character recognition unit is vertical writing or horizontal writing. Item 2. The document filing device according to Item 1. 10. The apparatus according to claim 1, wherein the feature creating means includes a plurality of different feature creating methods, and selects a corresponding feature creating method according to the type of character of the input keyword. Document filing device 11. The method according to claim 1, wherein the feature creating means creates an integrated feature by partially overlapping features of adjacent characters constituting the input keyword when the input keyword is an alphabetic character or a symbol. The document filing apparatus according to claim 10, wherein 12. The character recognizing means performs character segmentation at the time of character recognition, saves rectangular information for each character in the search data storage unit, and stores each character of the input keyword output by the search feature creating means. From the rectangular shape of the character and the information of the character rectangle obtained from the search data storage unit and the number of characters to be compared among the characters of the input keyword, it is determined whether or not the feature is to be compared. 13. The document filing apparatus according to claim 1, further comprising a feature collation determining unit that determines that the character string in the search data storage unit does not match the input keyword. 13. An inputting step of inputting a document image,
A character recognition step of extracting characters from the document image input in the input step, referring to a recognition dictionary in which a standard pattern of each character is stored in advance, and recognizing the extracted characters to create a character code; A feature creation step of creating a feature for each character recognized by the step;
A search data step of storing the character code created by the character recognition step and the feature created by the feature creation step in a search data storage unit; A search feature creation step of acquiring from a feature dictionary in which features are retained; and searching by collating the input keyword with data in the search data storage unit. A document filing method, comprising: a search step of referring to and processing and outputting; and a display step of displaying a search result of the search step on a display unit.