JPH1069494A - Image retrieval method and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image retrieval method and the device capable of highly accurately retrieving a character string at a low cost and at a high speed. SOLUTION: A retrieval character string is developed to a corresponding character string image and prescribed filtering is performed to the developed character string image (S24). Then, the filtered character string image is segmented, segmented respective parts are character-recognized and plural recognition character candidates including the retrieval character string are acquired (S3). Then, the acquired plural recognition character candidates are combined and possible different recognition character string candidates are generated (S44). By the OR conditions of the recognition character string candidates, character string recognition is performed (S5).

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image search method and apparatus, and more particularly to an image search method and apparatus for searching a target image for a predetermined character string.

2. Description of the Related Art Conventionally, when performing OCR, in order to prevent erroneous recognition, recognition candidates with a low recognition rate are specified, and a person manually selects or corrects the candidates.

However, the necessity of this manual work makes it difficult to realize full automatic recognition processing of characters. In order to solve this, research has been conducted to perform this work mechanically, and there are examples of misrecognition and correction using word dictionaries and knowledge bases. If not used,
There are many barriers, such as not being practical. Also, in order to narrow down the number of recognition candidates to one, there is a problem that once erroneous correction is performed, a procedure for restoring the recognition candidate becomes troublesome. The present invention has been made in view of the above conventional example, and has as its object to provide an image search method and an image search method capable of performing high-accuracy character string search at low cost.

In order to achieve the above object, an image search method and apparatus according to the present invention have the following arrangement. That is, an expansion step of expanding the search character string into a corresponding character string image, a filtering step of performing predetermined filtering on the character string image expanded in the expansion step, and a character string image filtered in the filtering step are performed independently. Segmentation into parts, performing character recognition of each of the segmented parts, and a recognition step of obtaining a recognized character string candidate, and another recognition that can be combined based on a difference between the recognized character string candidates obtained in the recognition step. A generating step of generating a character string candidate, and searching for a predetermined character string based on a logical sum condition of the another recognizable character string candidate generated in the generating step and the recognized character string candidate obtained in the recognition step Performing a character string search step. Further, another invention is a development means for developing a search character string into a corresponding character string image, a filtering means for performing a predetermined filtering on the character string image developed by the development means, and a filtering operation performed by the filtering means. Segmenting the character string image into independent parts, performing character recognition on each of the segmented parts, and a recognition unit that obtains a recognized character string candidate, based on the difference between the recognized character string candidates obtained by the recognition unit, combining Generating means for generating another possible recognition character string candidate; and a predetermined logical sum condition of the another combination character string candidate generated by the generation means and the recognition character string candidate obtained by the recognition means. And character string search means for searching for the character string. Another invention is a computer program product, comprising a computer usable medium having computer readable program code means, wherein the computer program product expands a search character string into a corresponding character string image,
Computer-readable first program code means, computer-readable second program code means for performing predetermined filtering on the character string image developed by the first program code means, and the second program code means;
Computer-readable third program code means for segmenting the character string image filtered by the program code means into independent parts, performing character recognition on each of the segmented parts, and obtaining recognized character string candidates; Computer-readable fourth program code means for generating another recognizable character string candidate that can be combined based on the difference between the recognized character string candidates obtained by the program code means, and a fourth program code means generated by the fourth program code means A fifth program code unit that searches for a predetermined character string based on a logical sum condition of the another combination of recognized character strings that can be combined and the recognized character string candidate obtained by the third program code unit.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A detailed description will be given of a character recognition / character string search processing configuration according to an embodiment of the present invention with reference to FIG. Referring to FIG. 1, the character recognition / character string search processing configuration according to the embodiment of the present invention includes an information processing device 8, an external storage device 4, a keyboard / mouse 5, a display device 6, and an image scanner 1. The CPU 7 controls the entire information processing device 8. The CPU 7 loads various kinds of processing software stored in the external storage device 4 into the RAM in advance, sequentially reads, interprets, and executes the software. The various kinds of processing software include OCR processing software 2 and document search software 3. The image scanner 1 obtains image information by scanning the document 9 and transfers the image information to the information processing device 8. The display device 1 displays various results processed by the information processing device 8 and commands / data input from the keyboard / mouse 5. The keyboard / mouse 5 inputs various commands and data such as search words and search conditions. The external storage device 4 includes programs (1) of OCR processing software and document search software which are loaded into the RAM and executed by the CPU 7.
0), a recognition dictionary 1 for character recognition / document search processing
5 is stored. Further, the external storage device 4 stores the input image data (1
1) The text information (13) of the recognition result by the OCR software 2 and a search file (14) by executing the document search software 3 are stored. This search file (14) is created based on text information. The external storage device 4 stores dictionary image data (12), which is a reference image pattern corresponding to each character of the search instruction character string input from the keyboard 5. FIG. 2A is a flowchart showing a processing procedure for creating index information for document search in information processing apparatus 8 of the present embodiment. FIG. 2B is a flowchart illustrating a character search processing procedure for performing a character search using the created index information. First, a processing procedure for creating index information for document search will be described with reference to FIG. 2A. In step S11, a search target image is input from the image scanner 1, and the external storage device 4
Is loaded into the RAM of the information processing device 8. In step S12, character recognition is performed on the search target image loaded in the RAM. Then, the most likely recognized character string is obtained. In step S13, the input image input in step S11 and the step S
Then, the character string recognized in step 12 and the index data describing the relationship between the input image and the recognized character string are stored in the search file (FIGS. 1, 14). This table of contents data is, for example, data for searching in which registered document image group a character string "image" was entered, and a document image including the character string using this can be quickly known. . This is the end of the description of the processing at the time of registration. Next, a character search process for performing a character search using the created index information will be described with reference to FIG. 2B. The processing from step 1 to step S4 shows a processing procedure for obtaining a search logical sum condition for the character string search processing in step 5 based on the input search character string. In steps S5 and S6, a character string closest to the input search character string is determined based on the obtained search logical OR condition, and the index information generated in the flowchart of FIG. Display an image.
The OCR processing software 2 describes the processing procedure from step 1 to step S4 and from step 11 to step 12. Further, the document search software 1 describes the processing procedure from step 5 to step S6.
Hereinafter, the processing procedure of FIG. 2 will be described for each step. In step S1, a search instruction character string is input from the keyboard 5. In step S2, reference image pattern data corresponding to each character of the input search instruction character string is extracted from the dictionary image data 12, and a character string image pattern corresponding to the input search instruction character string is generated. still,
Here, image processing parameters such as each character image interval and the amount of blurring of the image blurring process are determined by using a predetermined value or by inputting from the keyboard 5. In step S3, the character string image generated in step S2 is divided into individual characters. Then, a feature amount of a predetermined feature is extracted for each divided character, and a recognition dictionary is searched based on the feature amount to extract a recognition candidate. In this step, the input search instruction character string is included as at least one of the recognition candidates. Here, the type of the feature may be, for example, a directional feature or a shading feature, but the present invention is not limited to this feature. In step S4, based on the recognition candidates generated in step S3, possible characters are combined to generate a plurality of character strings as recognition candidates. Then, step S5
Generates an OR condition of a character string for the search processing in.
The above processing is processing in the OCR software 2.
Next, steps S5 and S6 which are processes in the document search software 3 will be described. In step S5, step S5
In the OR condition of the character strings generated in step 4, a character string close to those character strings is searched for from the character string in the index information including the document image and the character string included in the document image generated in the processing up to step S13. . Then, the search results (document images, character strings, etc.) are displayed on the display device 6. For example, the searched position is displayed by underlining the position of the input search target image and the character string corresponding to the search result. Next, referring to FIG.
The points in the search OR condition generation processing from 1 to S4 will be specifically described. First, for example, "multi" is input as a search instruction character string from the keyboard 5 (reference numeral 1000). Next, an image pattern corresponding to each character of the search instruction character string is fetched from the dictionary image data, and an image pattern of the character string is generated. A plurality of parameters for generating a character string image pattern to be generated here are prepared. These parameters are applied to an image pattern taken from dictionary image data to generate a character string image pattern. The parameters include image processing parameters such as a character image interval and a blur amount of the image blur processing. These parameters are determined by using predetermined values stored in the external storage device 4 in advance or by inputting from the keyboard 5. Next, in step S3, a series of character recognition processing (1002) of the document image stored in the RAM is started. Then, it is assumed that a character area is cut out with respect to the position of “R”, and then character recognition is performed in the cut out area, and the character is recognized as two characters “NO” and “RE”. With respect to a character such as "ru", a character cutout error is likely to occur, and for example, "ru" is likely to be erroneously recognized as "no" or "re". In this recognition process, it is assumed that a recognition candidate (1003) of a first recognition candidate: Manorechi and a second recognition candidate: Anoruchi are obtained. Next, in step S4, the obtained first and second recognition characters are compared, a different character portion is detected, and another character candidate character string is generated by replacing the character portion.
In the above example, since the first and third characters are different from each other, there are four possible combinations of character strings, and the combination of the above two candidates plus two character strings is possible. That is, a. Manoreti b. Ano Lchi c. Anoreti d. Ano Lchi. These four types of character strings (1004) are erroneously recognized character string candidates because the expected character string is “multi”. However, in the process of character recognition, it cannot be avoided that these recognition candidates are selected, and correct recognition becomes difficult. Therefore, in the embodiment of the present invention, with the reverse idea, a misrecognition candidate character string that may be selected in the character recognition process is also recognized as a recognition candidate for the expected recognition candidate “multi,” and these misrecognition candidates are recognized. By taking the logical OR condition of the recognition candidate character string and the expected recognition candidate (1005), a method of performing comprehensive recognition is provided. Therefore, in the ordinary full-text search, for a search designated character string such as “multi” which may possibly separate characters, an erroneous search is likely to occur in the search for a document image. By performing comprehensive recognition including a character string, a “multi” character string can be selected as a recognition result. In the processing up to step S4, a search OR condition corresponding to the input character string has been generated. A character string that satisfies the search OR condition is set
It is possible to search from the character string of the index information including the document image and the character string included in the document image generated by the processing up to 13 and display the search result (document image, character string, etc.) on the display device 6. it can. Next, a detailed processing procedure of step S2 will be described with reference to FIG. In step S20, the degree of ambiguity is set as one parameter for generating a character string image pattern corresponding to the input search instruction character string. This ambiguity may be a default value of the ambiguity stored in the external storage device in advance, or may be input from the keyboard 5. Here, this degree of ambiguity is introduced to help lower the accuracy of the character recognition process that is expected when the resolution of the document image to be searched is low or contains a lot of noise. That is, when the quality of the document image to be searched is poor, the degree of ambiguity is determined to be high, and when the quality is good, the degree of ambiguity is determined to be low. The user roughly determines the quality of the document image to be searched and sets the corresponding ambiguity. The size of the character image pattern (font) and the character pitch size corresponding to each character of the input search instruction character string are set according to the set ambiguity. This setting reduces the character font and the character pitch when the degree of ambiguity increases. When the degree of ambiguity is reduced, the character font is increased and the character pitch is increased. Next, in step S21, reference image pattern data corresponding to each character of the input search instruction character string is extracted from the dictionary image data 12, and a character corresponding to the character font size and character pitch set in step S20 is extracted. Generate a column pattern. In this case, as the reference image pattern data (font), a character font that is not good at recognition is used. In step S22, noise corresponding to the degree of ambiguity is superimposed on the character string image pattern generated in step S21. That is, if the degree of ambiguity is large, the noise is determined to be strong, and if the degree of ambiguity is small, the noise is determined to be weak. The intensity of this noise is, for example, superimposed randomly on a character string image pattern using many noise dots as strong noise. Conversely, as weak noise, the noise level for the character string image pattern can be adjusted by randomly superimposing the character string image pattern using a small number of noise dots. In step S23, a blurring process is performed on the character string image pattern processed in step S22. This blur amount is also determined in step S2.
Similar to 2, when the degree of ambiguity is large, the blur amount is large,
Conversely, when the degree of ambiguity is small, the blurring amount may be reduced. With the above procedure, the process of generating the character string image pattern corresponding to each character of the input search instruction character string in step S2 can be realized. As explained above,
In the embodiment of the present invention, a search character string, for example, “multi-me” is bitmap-developed so that the OCR is likely to cause erroneous recognition.
A character string “Manorechi” is obtained as one of the misrecognition tendency character strings, and the input document image is searched by ORing the search designated character string and these misrecognition tendency character strings in an OR relationship. Search performance can be improved. In the embodiment of the present invention, in addition to the document image search described above, the present invention can be applied to a search using a database. Further, application to a case where the likelihood of recognition of some information of the optical reading device by the OMR or the barcode is low can be performed by the same method. In the embodiment according to the present invention, the peculiarity of the OCR is obtained by bitmap-expanding the search designated character string and performing the OCR. It goes without saying that the characteristic peculiar to the OCR may be used to reflect the character string that tends to be misrecognized. In addition,
The present invention can be applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.) but also to an apparatus (for example, a copying machine, a facsimile device, etc.) including one device. May be. Further, an object of the present invention is to provide a storage medium storing a program code of software for realizing the functions of the above-described embodiments to a system or an apparatus, and a computer (or CPU or MPU) of the system or apparatus to store the storage medium. It goes without saying that the present invention is also achieved by reading and executing the program code stored in the medium. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention. Examples of a storage medium for supplying the program code include a floppy disk, hard disk, optical disk, magneto-optical disk, and C
A D-ROM, a CD-R, a magnetic tape, a nonvolatile memory card, a ROM, and the like can be used. When the computer executes the readout program code, not only the functions of the above-described embodiments are realized, but also an OS (Operating System) running on the computer based on the instruction of the program code.
It goes without saying that a case where the functions of the above-described embodiments are implemented by performing some or all of the actual processing, and the processing performs the functions of the above-described embodiments. Further, after the program code read from the storage medium is written into a memory provided on a function expansion board inserted into the computer or a function expansion unit connected to the computer, the function expansion is performed based on the instruction of the program code. It goes without saying that a CPU or the like provided in the board or the function expansion unit performs part or all of the actual processing, and the processing realizes the functions of the above-described embodiments. When the present invention is applied to the storage medium, the storage medium stores program codes corresponding to the above-described flowcharts. However, in brief, each module shown in the memory map example of FIG. Is stored in the storage medium. That is, at least a “search character string input module” for inputting a search character string and a “search O for generating a search OR condition based on the input search character string”
"R condition generation module", "search target image input module" for inputting a search target document image, "character recognition module for performing character recognition of a character string image pattern generated based on a search character string and input search target document image""A character string search module for acquiring a character string that satisfies the generated search OR condition from the character string of the search target document image,
What is necessary is just to store the program code of each module of the “display module” for displaying the search result in the storage medium. As described above, according to the embodiment of the present invention, the OCR
If the misrecognition of is systematic, obtain the characters of the search specified character string and a group of character strings indicating the tendency of misrecognition in accordance with this,
By performing the search in the relationship between the original search designation character string and the OR, even when the OCR recognition rate is low, a document image search with few omissions in search can be performed. In addition, there is no need to create a correct answer-misrecognition table for characters and character strings that are likely to be misrecognized in advance, and a tendency of misrecognition that matches the characters of the search designated character string is obtained. As a result, a more accurate misrecognition tendency can be obtained than when the correct-misrecognition table is applied. Since the present embodiment is a technical extension related to a search designation word, a conventionally existing OC
For a document image search system using R, simply applying the embodiment according to the present invention before the search condition input stage,
A search that is resistant to OCR misrecognition becomes possible. These advantages eliminate the need for manual OCR misrecognition and correction work, enable the construction of an unmanned system from input of document images to storage of image data and text data via OCR, and without imposing a load on the computer and processing speed. A relatively inexpensive system can be realized without fear of a decrease. [Second Embodiment] An example of searching for a likely character string as a whole based on the above-described search OR condition has been described. However, the present invention is not limited to this, and any one of the search OR conditions may be used.
A character string that satisfies the above conditions may be searched, and a document image including the character string may be used as a search result.

As described above, according to the present invention, a high-precision character string search can be performed at high speed and at low cost.

[Brief description of the drawings]

FIG. 1 is a hardware configuration diagram of an embodiment of the present invention.

FIG. 2A is a flowchart illustrating a processing procedure for acquiring a document image and character string information included in the document image according to the embodiment of this invention.

FIG. 2B is a flowchart illustrating a character string search processing procedure according to the embodiment of this invention.

FIG. 3 is a diagram for explaining a search OR condition generation step according to the embodiment of this invention.

FIG. 4 is a flowchart showing a detailed processing procedure of step S2.

FIG. 5 is a diagram showing a layout of each program module stored in a predetermined computer-readable recording medium.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image scanner 2 RAM (OCR processing software) 3 RAM (document search software) 4 External storage device 5 Keyboard 6 Display device 7 CPU

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] September 9, 1996

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Detailed description of the invention

[Correction method] Change

[Correction contents]

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image search method and apparatus, and more particularly to an image search method and apparatus for searching a target image for a predetermined character string.

[0002]

2. Description of the Related Art Conventionally, when performing OCR, in order to prevent erroneous recognition, recognition candidates with a low recognition rate are specified, and a person manually selects or corrects the candidates.

[0003]

However, the necessity of this manual work makes it difficult to realize full automatic recognition processing of characters. In order to solve this, research has been conducted to perform this work mechanically, and there are examples of misrecognition and correction using word dictionaries and knowledge bases. If not used,
There are many barriers, such as not being practical.

[0004] Further, in order to narrow down the number of recognition candidates to one, there is a problem that once erroneous correction is performed, a procedure for restoring the erroneous correction becomes troublesome. The present invention has been made in view of the above conventional example, and has as its object to provide an image search method and an image search method capable of performing high-accuracy character string search at low cost.

[0005]

In order to achieve the above object, an image search method and apparatus according to the present invention have the following arrangement. That is, an expansion step of expanding the search character string into a corresponding character string image, a filtering step of performing predetermined filtering on the character string image expanded in the expansion step, and a character string image filtered in the filtering step are performed independently. Segmentation into parts, performing character recognition of each of the segmented parts, and a recognition step of obtaining a recognized character string candidate, and another recognition that can be combined based on a difference between the recognized character string candidates obtained in the recognition step. A generating step of generating a character string candidate, and searching for a predetermined character string based on a logical sum condition of the another recognizable character string candidate generated in the generating step and the recognized character string candidate obtained in the recognition step Performing a character string search step.

Further, another invention is a developing means for developing a search character string into a corresponding character string image, a filtering means for performing a predetermined filtering on the character string image developed by the developing means, and a filtering means. Recognition means for segmenting the filtered character string image into independent parts, performing character recognition on each of the segmented parts, and obtaining recognized character string candidates;
Generating means for generating another recognizable character string candidate that can be combined based on the difference between the recognized character string candidates obtained by the recognizing means, and another recognizable character string candidate that can be generated by the generating means And a character string search means for searching for a predetermined character string based on a logical OR condition of the recognized character string candidates obtained by the recognition means.

Another aspect of the invention is a computer program product comprising a computer usable medium having computer readable program code means, wherein the computer program product converts a search character string into a corresponding character string image. Computer-readable first program code means to be expanded, computer-readable second program code means to perform predetermined filtering on the character string image expanded by the first program code means, and the second program code Computer-readable third program code means for segmenting the character string image filtered by the means into independent parts, performing character recognition on each of the segmented parts, and obtaining recognized character string candidates, and the third program. Computer-readable fourth program code means for generating another recognizable character string candidate that can be combined based on the difference between the recognized character string candidates obtained by the program code means, and the fourth program code means generated by the fourth program code means. A fifth program code unit for searching for a predetermined character string based on a logical sum condition of another combination of recognized character strings that can be combined with the recognized character string candidate obtained by the third program code unit.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A detailed description will be given of a character recognition / character string search processing configuration according to an embodiment of the present invention with reference to FIG. Referring to FIG. 1, the character recognition / character string search processing configuration according to the embodiment of the present invention includes an information processing device 8, an external storage device 4, a keyboard / mouse 5, a display device 6, and an image scanner 1.

The CPU 7 controls the entire information processing apparatus 8. The CPU 7 loads various processing software stored in the external storage device 4 into the RAM in advance,
It is read, interpreted and executed sequentially. The various kinds of processing software include OCR processing software 2 and document search software 3. The image scanner 1 obtains image information by scanning the document 9 and transfers the image information to the information processing device 8.

The display device 1 displays various results processed by the information processing device 8 and commands / data input from the keyboard / mouse 5. Keyboard / Mouse 5
Inputs various commands and data such as search terms and search conditions. The external storage device 4 is loaded into the RAM and
Each program (10) of OCR processing software and document search software executed by the PU 7 and a recognition dictionary 15 for character recognition / document search processing are stored.

The external storage device 4 stores input image data (11), O
Text information of recognition result by CR software 2 (1
3) In addition, a search file (14) by executing the document search software 3 is stored. This search file (14) is created based on text information. The external storage device 4 stores dictionary image data (12), which is a reference image pattern corresponding to each character of the search instruction character string input from the keyboard 5.

FIG. 2A shows an information processing apparatus 8 according to this embodiment.
9 is a flowchart showing a processing procedure for creating index information for document search in the first embodiment. Also, FIG.
9 is a flowchart illustrating a character search processing procedure for performing a character search using created index information. First, a processing procedure for creating index information for document search will be described with reference to FIG. 2A.

In step S11, a search target image is input from the image scanner 1, stored in the external storage device 4, and then loaded into the RAM of the information processing device 8. In step S12, character recognition is performed on the search target image loaded in the RAM. Then, the most likely recognized character string is obtained. In step S13, the input file input in step S11, the character string included in the image recognized in step S12, and table of contents data describing the relationship between the input image and the recognized character string are stored in a search file (FIG. 1). , 1
4). This table of contents data is, for example, data for searching in which registered document image group a character string "image" was entered, and a document image including the character string using this can be quickly known. .

The description of the registration process has been completed.
Next, a character search process for performing a character search using the created index information will be described with reference to FIG. 2B. The processing from step 1 to step S4 shows a processing procedure for obtaining a search logical sum condition for the character string search processing in step 5 based on the input search character string. In steps S5 and S6, a character string closest to the input search character string is determined based on the obtained search logical OR condition, and the index information generated in the flowchart of FIG. Display an image.

Note that the OCR processing software 2 executes steps 1 to S4 and steps 11 to 1
2 is described. Further, the document search software 1 describes the processing procedure from step 5 to step S6. Hereinafter, the processing procedure of FIG. 2 will be described for each step. In step S1, a search instruction character string is input from the keyboard 5.

In step S2, reference image pattern data corresponding to each character of the input search instruction character string is extracted from the dictionary image data 12, and a character string image pattern corresponding to the input search instruction character string is generated. .
Here, the image processing parameters such as the character image interval and the blur amount of the image blur processing are determined by using a predetermined value or input from the keyboard 5.

In step S3, the character string image generated in step S2 is divided into characters. Then, a feature amount of a predetermined feature is extracted for each divided character, and a recognition dictionary is searched based on the feature amount to extract a recognition candidate. In this step, the input search instruction character string is included as at least one of the recognition candidates.
Here, the type of the feature may be, for example, a directional feature or a shading feature, but the present invention is not limited to this feature.

In step S4, based on the recognition candidates generated in step S3, possible characters are combined to generate a plurality of character strings as recognition candidates. Then, OR the character strings for the search processing in step S5.
Generate a condition. The above processing is the OCR software 2
It is processing in. Next, steps S5 and S6 which are processes in the document search software 3 will be described.

In step S5, a character string close to the character string is indexed by the OR condition of the character string generated in step S4, including the document image generated by the processing up to step S13 and the character string included therein. Search from the character string of Then, the search results (document images, character strings, etc.) are displayed on the display device 6. For example, the searched position is displayed by underlining the position of the input search target image and the character string corresponding to the search result.

Next, with reference to FIG. 3, the points in the search OR condition generation processing in steps S1 to S4 will be specifically described. First, for example, "multi" is input as a search instruction character string from the keyboard 5 (reference numeral 1000).

Next, an image pattern corresponding to each character of the search instruction character string is fetched from the dictionary image data, and an image pattern of the character string is generated. A plurality of parameters for generating a character string image pattern to be generated here are prepared. These parameters are applied to an image pattern taken from dictionary image data to generate a character string image pattern. The parameters include image processing parameters such as a character image interval and a blur amount of the image blur processing. These parameters include
Using a predetermined value stored in the external storage device 4 in advance,
It is determined by inputting from the keyboard 5.

Next, in step S3, a series of character recognition processing (1002) of the document image stored in the RAM is started. Then, it is assumed that a character area is cut out with respect to the position of “R”, and then character recognition is performed in the cut out area, and the character is recognized as two characters “NO” and “RE”. With respect to a character such as "ru", a character cutout error is likely to occur, and for example, "ru" is likely to be erroneously recognized as "no" or "re".

In this recognition processing, it is assumed that a recognition candidate (1003) of a first recognition candidate: Manorechi and a second recognition candidate: Anoruchi are obtained.

Next, in step S4, the obtained first and second recognition characters are compared, a different character portion is detected, and the character portion is replaced, so that another recognition candidate character string is obtained. Generate. In the above example, since the first and third characters are different from each other, there are four possible combinations of character strings, and the combination of the above two candidates plus two character strings is possible. That is, a. Manoreti b. Ano Lchi c. Anoreti d. Ano Lchi.

These four character strings (1004) are erroneously recognized character string candidates because the expected character string is "multi". However, in the process of character recognition, it cannot be avoided that these recognition candidates are selected, and correct recognition becomes difficult. Therefore, in the embodiment of the present invention,
In the reverse idea, an erroneous recognition candidate character string that may be selected during the character recognition process is also recognized as a recognition candidate for the expected recognition candidate “multi,” and the logic of the erroneous recognition candidate character string and the expected recognition candidate is recognized. By taking the sum condition (1005), a method for comprehensive recognition is provided.

Therefore, in a normal full-text search, "multi"
It is easy to cause erroneous search in the search of document image for the search specified character string that has the possibility of character separation like, but perform comprehensive recognition including the erroneous recognition candidate character string described above Thus, the character string “multi” can be selected as the recognition result. In the processing up to step S4, a search OR condition corresponding to the input character string has been generated.

A character string that satisfies the search OR condition is searched from the character string in the index information including the document image generated by the processing up to step S13 and the character string included therein, and the search result (document image , Strings, etc.)
Can be displayed on the display device 6. Next, a detailed processing procedure of step S2 will be described with reference to FIG.

In step S20, the degree of ambiguity is set as one parameter for generating a character string image pattern corresponding to the input search instruction character string. This ambiguity may be a default value of the ambiguity stored in the external storage device in advance, or may be input from the keyboard 5. Here, this degree of ambiguity is introduced to help lower the accuracy of the character recognition process that is expected when the resolution of the document image to be searched is low or contains a lot of noise. That is, when the quality of the document image to be searched is poor, the degree of ambiguity is determined to be high, and when the quality is good, the degree of ambiguity is determined to be low.

The user roughly determines the quality of the document image to be searched and sets the corresponding ambiguity. The size of the character image pattern (font) and the character pitch size corresponding to each character of the input search instruction character string are set according to the set ambiguity. This setting reduces the character font and the character pitch when the degree of ambiguity increases. When the degree of ambiguity is reduced, the character font is increased and the character pitch is increased.

Next, at step S21, reference image pattern data corresponding to each character of the input search instruction character string is extracted from the dictionary image data 12, and is converted to the character font size and character pitch set at step S20. Generate the corresponding string pattern. In this case, as the reference image pattern data (font), a character font that is not good at recognition is used.

In step S22, noise corresponding to the degree of ambiguity is superimposed on the character string image pattern generated in step S21. That is, if the degree of ambiguity is large, the noise is determined to be strong, and if the degree of ambiguity is small, the noise is determined to be weak. The intensity of this noise is, for example, superimposed randomly on a character string image pattern using many noise dots as strong noise. Conversely, as weak noise, the noise level for the character string image pattern can be adjusted by randomly superimposing the character string image pattern using a small number of noise dots.

In step S23, a blurring process is performed on the character string image pattern processed in step S22. As in the case of step S22, the blur amount may be increased when the degree of ambiguity is large, and may be decreased when the degree of ambiguity is small. With the above procedure, the process of generating the character string image pattern corresponding to each character of the input search instruction character string in step S2 can be realized.

As described above, in the embodiment of the present invention, a search character string, for example, "multi-me" is bitmap-developed so that OCR is likely to cause erroneous recognition, and this is subjected to OCR. , Is obtained as one of the misrecognition tendency character strings, and the input document image is searched for the search designated character string and the misrecognition tendency character strings in an OR relationship. , Search performance can be improved.

In the embodiment of the present invention, in addition to the document image search described above, the present invention can be applied to a search using a database. Further, application to a case where the likelihood of recognition of some information of the optical reading device by the OMR or the barcode is low can be performed by the same method. In the embodiment according to the present invention, O
A CR peculiar habit was obtained by developing a search designated character string into a bitmap and performing OCR on the character string.
It goes without saying that the CR peculiarity may be reflected in the misrecognition tendency character string.

The present invention can be applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), but can be applied to a single device (for example, a copier, a facsimile). Device). Further, an object of the present invention is to provide a storage medium storing a program code of software for realizing the functions of the above-described embodiments to a system or an apparatus, and a computer (or CPU or MPU) of the system or apparatus to store the storage medium. It goes without saying that the present invention is also achieved by reading and executing the program code stored in the medium.

In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention. Examples of a storage medium for supplying the program code include a floppy disk, hard disk, optical disk, magneto-optical disk, and C
A D-ROM, a CD-R, a magnetic tape, a nonvolatile memory card, a ROM, and the like can be used.

When the computer executes the readout program code, not only the functions of the above-described embodiment are realized, but also the OS (Operating System) running on the computer based on the instruction of the program code. ) May perform some or all of the actual processing, and the processing may realize the functions of the above-described embodiments.

Further, after the program code read from the storage medium is written into a memory provided on a function expansion board inserted into the computer or a function expansion unit connected to the computer, based on the instructions of the program code, It goes without saying that the CPU included in the function expansion board or the function expansion unit performs part or all of the actual processing, and the processing realizes the functions of the above-described embodiments.

When the present invention is applied to the storage medium, the storage medium stores program codes corresponding to the above-described flowcharts. Each module shown will be stored in a storage medium. That is, at least a “search character string input module” for inputting a search character string, a “search OR condition generation module” for generating a search OR condition based on the input search character string, and a “search for inputting a search target document image” "Target image input module", "Character recognition module" that performs character recognition of a character string image pattern generated based on a search character string and an input search target document image, and searches for a character string that satisfies the generated search OR condition The program codes of the "character string search module" obtained from the character string of the target document image and the "display module" for displaying the search result may be stored in the storage medium.

As described above, according to the embodiment of the present invention, when the OCR erroneous recognition is systematic, the characters of the search designated character string and the character string group indicating the tendency of the erroneous recognition are obtained. Then, by performing this search in the relationship between the original search designation character string and OR, even when the OCR recognition rate is low, a document image search with few omissions in search can be performed. In addition, correct characters and character strings that are easily misrecognized in advance
Since there is no need to create a misrecognition table and a tendency of misrecognition that matches the sequence of characters in the search designated character string is obtained, the accuracy of each character is higher than when the correct-misrecognition table is applied. High misrecognition tendency is obtained.

Since the present embodiment is a technical extension related to a search designation word, the embodiment according to the present invention is applied to a conventional document image search system using OCR before a search condition input stage. A search that is resistant to OCR misrecognition becomes possible only by adapting. These advantages eliminate the need for manual OCR misrecognition and correction work, and enable the construction of a system that enables unattended operation from document image input to storage of image data and text data via OCR. A relatively inexpensive system can be realized without fear of a decrease. [Second Embodiment] An example of searching for a likely character string as a whole based on the above-mentioned search OR condition has been described. However, the present invention is not limited to this.
A character string that satisfies the above conditions may be searched, and a document image including the character string may be used as a search result.

[0042]

As described above, according to the present invention, a high-precision character string search can be performed at high speed and at low cost.

Claims (19)

[Claims] An expansion step of expanding a search character string into a corresponding character string image; a filtering step of performing predetermined filtering on the character string image expanded in the expansion step; and a character string filtered in the filtering step Segmentation of the image into independent parts, performing character recognition of each segmented part, and a recognition step of obtaining a recognized character string candidate, based on the difference between the recognized character string candidates obtained in the recognition step, can be combined A generation step of generating another recognized character string candidate; and a predetermined character under a logical OR condition of the another combinable character string candidate generated in the generating step and the recognized character string candidate obtained in the recognition step. A character string search step of searching for a column. 2. The method according to claim 1, wherein the character string image corresponding to the search character string is a character string image having a relatively low recognition rate. 3. The method according to claim 2, wherein the predetermined character string is included in a search target image, and the predetermined filtering type is determined based on an ambiguity of the search target image. 2. The image search method according to 1. 4. The method according to claim 3, wherein the ambiguity of the search target image is high if the image quality of the search target image is poor, and low if the image quality is good. . 5. The image according to claim 4, wherein in the expanding step, the character string image corresponding to the search character string based on the ambiguity is expanded into a character string image having a character pitch and a character size. retrieval method. 6. The development step, wherein, if the degree of ambiguity is large, the character string image corresponding to the search character string is developed into a character string image having a small character pitch and a small character size. 5. The image search method according to 5. 7. The filtering method according to claim 1, wherein a filter that superimposes strong noise is used when the degree of ambiguity of the search target image is large.
Image search method described in. 8. The image search method according to claim 1, wherein, as the predetermined filtering type, a filter that increases blurring is used when the ambiguity of the search target image is large. 9. Expanding means for expanding a search character string into a corresponding character string image, filtering means for performing predetermined filtering on the character string image expanded by the expanding means, and a character string filtered by the filtering means Segmenting the image into independent parts, performing character recognition on each of the segmented parts, and a recognition unit that obtains a recognized character string candidate, based on the difference between the recognized character string candidates obtained by the recognition unit, can be combined. Generating means for generating another recognized character string candidate; and a predetermined character based on a logical sum condition of the another recognizable character string candidate generated by the generating means and the recognized character string candidate obtained by the recognizing means. An image search apparatus comprising: a character string search unit that searches for a column. 10. The image search apparatus according to claim 9, wherein the character string image corresponding to the search character string is a character string image having a relatively low recognition rate. 11. The method according to claim 11, wherein the predetermined character string is included in a search target image, and the predetermined filtering type is determined based on an ambiguity of the search target image. 10. The image search device according to 9. 12. The image search apparatus according to claim 11, wherein the ambiguity of the search target image is high when the image quality of the search target image is poor, and low when the image quality is good. . 13. The image according to claim 12, wherein the expansion unit expands the character string image corresponding to the search character string based on the ambiguity into a character string image having a character pitch and a character size. Search device. 14. The character string image corresponding to the search character string having a narrow character pitch and a small character size if the degree of ambiguity is large. 14. The image search device according to claim 13. 15. The type of the predetermined filtering is:
The image search apparatus according to claim 9, wherein a filter that superimposes strong noise is used if the degree of ambiguity of the search target image is large. 16. The predetermined filtering type is:
10. The image search device according to claim 9, wherein a filter for increasing blur is used when the degree of ambiguity of the search target image is large. 17. A computer program product comprising a computer usable medium having computer readable program code means, said computer program product developing a search string into a corresponding string image. Possible first program code means, computer-readable second program code means for performing predetermined filtering on the character string image developed by the first program code means, and filtered by the second program code means Computer-readable third program code means for segmenting the character string image into independent parts, performing character recognition on each of the segmented parts, and obtaining recognized character string candidates, and the third program code means Computer-readable fourth program code means for generating another recognizable character string candidate based on the difference between the recognized character string candidates obtained in the above step, and the combination possible generated by the fourth program code means A computer program product comprising: a fifth program code means for searching for a predetermined character string based on a logical sum condition of another recognized character string candidate and a recognized character string candidate obtained by the third program code means. . 18. The image search method according to claim 1, wherein the recognition step includes at least the search character string in the obtained recognized character string candidates. 19. The image search apparatus according to claim 9, wherein the recognition unit includes at least the search character string in the obtained recognized character string candidates.