JP2977848B2 - Dictionary construction method for character recognition - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a recognition dictionary configuration method in a character recognition device.

[Conventional technology]

In the character recognition device, an input character is identified by using a recognition dictionary in which the characteristics of the character are registered in advance for each character code, and determining the distance between the characteristic amount of the input character and the characteristic amount of the dictionary. Conventionally, in order to enhance the maintainability of this type of recognition dictionary, for example, as described in JP-A-63-198180, the feature amount itself registered in the dictionary is feature information in a form capable of reproducing the original image, When adding or deleting registered characters, there is a method of reproducing an original image based on feature information.

[Problem to be Solved by the Invention] In the above-mentioned conventional method, since the feature amount registered in the dictionary is the feature information in a form capable of reproducing the original image, the types of feature amounts that can be used as the dictionary are extremely limited. (For example, a direction code cannot be used), and a dictionary configured as an average feature amount of a plurality of character images as a component,
The original image cannot be faithfully reproduced. Further, means for generating the original image based on the dictionary feature information is required.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems of the conventional method and to facilitate maintenance of a recognition dictionary without affecting recognition processing.

[Means for solving the problem]

In order to achieve the above object, the character recognition dictionary construction method of the present invention comprises, for each character, a record that stores at least a character code and an average feature amount of one or more character images of the character. A dictionary file for recognition
In addition to the dictionary file, the dictionary file includes an original image of the character and an original image storage file that holds the feature amount of the image corresponding to each record.

Further, the character recognition dictionary construction method of the present invention includes a recognition dictionary unit and an original image storage unit. The recognition dictionary unit includes at least a character code and one or more character images of the character in order for each character. , And the original image storage unit stores the original image of the character and the characteristic amount of the image in order for each character.

(Operation)

In order to facilitate the maintenance of the recognition dictionary, it is only necessary to store the image based on the feature amount in the dictionary. However, in an optical character recognition device (OCR), the data of this original image is quite large. If this data is stored in the recognition dictionary as a set of feature values and original images for each feature value record, a large memory area is required for recognition processing, or complicated processing for saving memory is required. Is required, and in that case, there is a possibility that the processing time will increase.

For example, in the case of kanji OCR, if JIS first and second level characters are to be recognized, a dictionary of 6353 character types must be provided. Storing a 64 × 64 normalized image for each requires about 3 megabytes of space. Since the original image data is not required at the time of the recognition processing, this area is a useless area when all the dictionaries are read at the time of the recognition processing. Also, in order to save this area, if the original image is skipped and loaded onto the memory, 6353 file seeks are required, leading to an increase in processing time.

According to claim (1), the original image of the feature amount in the dictionary and the feature amount of the image are stored in separate files, so that only the recognition dictionary is loaded into the memory for use in the recognition process. Thus, the recognition processing can be performed in exactly the same manner as when the original image is not stored. In claim (2),
Only the part of the dictionary in which the feature amounts used for recognition are registered is loaded into the memory, and in the recognition processing, processing such as skipping the original image part is not required.

〔Example〕

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 shows a dictionary configuration according to a first embodiment of the present invention. In the present embodiment, the dictionary configuration has three levels of a dictionary file 11, an original image storage file management file 12, and an original image storage file 13.

The dictionary file 11 is a file for registering a feature amount of a character image, and corresponds to a conventional recognition dictionary. The dictionary file 11 is composed of a plurality of records arranged in the order of record numbers,
One record has a header part and a data part. The header section stores a record number (No. 4 in this example), a character code, and the like, and the data section stores an average feature amount (for example, a histogram of a direction code) of one or a plurality of the character images. You.

The original image storage file management file 12 is a dictionary file
This is a file for managing an original image storage file 13 that holds a character image that is the basis of the feature amount registered in 11. The records in the management file 12 correspond one-to-one with the records in the dictionary file 11, and are arranged in the order of the record numbers similarly to the dictionary file 11. One record consists of a header part and a data part, and the header part has a record number (No. 4 in this example),
The number of original images (the number of samples used to take the average, 8 in this example) and others are stored, and the original image file name (ocr. 04s in this example) is stored in the data portion.

There are as many original image storage files 13 as there are record numbers in the dictionary file 11 and the management file 12. One original image storage file 13 includes a header part and a data part, and the header part stores the file name, the number of original images (8 in this example), and others, and the data part stores the original image and its characteristic amount. And other data (such as creation date) are stored for the number of original images in the header (8 in this example). The average of the feature amounts of each image in the original image storage file 13 is registered in the dictionary file 11.

The dictionary file 11, the original image storage file management file 12, and the original image storage file 13 are stored in an external storage device such as a floppy disk. Only the dictionary file 11 is loaded into the memory and used for the recognition process. Since the recognition process using the dictionary file 11 is the same as the conventional one, further description will be omitted. Hereinafter, the dictionary correction processing will be described.

When the dictionary is modified, the original image storage file is managed by the dictionary file 11 and the corresponding original image storage file management file 12.
Confirm the original image by referring to 13. For example, when confirming the original image of the fourth feature amount (record number 4) in the dictionary file 11, first, a corresponding record is found from the original image storage file management file 12 using the record number 4 as a key, and The original image storage file 13 is found by using the original image storage file name (ocr.04s in this example) stored in the record as a key. By taking out the original image from the original image storage file 13 and displaying it, the original image can be checked at a glance.

When deleting a feature quantity in the dictionary file 11, the corresponding record (for example, record number 4) in the dictionary file 11 is deleted.
), The original image storage file name of the corresponding record in the original image storage file management file 12, and the original image storage file 13 are deleted. In addition, when deleting a part of the feature amount that is the basis of the averaged feature amount in the dictionary file 11, the corresponding original image in the original image storage file 13 is displayed to confirm the user. After that, the original image and its characteristic amount are deleted, the average is calculated again with the characteristic amount of the remaining original image, and the dictionary file 11 is deleted.
Correct the feature value of the corresponding record in the new average value.
In this case, it is necessary to change the number of original images in the original image storage file management file 12 and the original image storage file 13.

When a new feature is added to the dictionary file 11, an original image storage file 13 is created, the original image and its features are stored for the number of samples, and the original image storage file 13 is stored in the original image storage file management file 12. Add a record and save the original image save file name.

When a new feature amount is added to a certain feature amount in the dictionary file 11 and the average is taken, an original image and a feature amount to be newly added to the corresponding original image storage file 13 are added. The average of all the characteristic amounts in the original image storage file 13 including the added characteristic amount is obtained, and the characteristic amount of the corresponding record in the recognition dictionary file 11 is corrected to the newly obtained average value. Also in this case, the number of original images in the original image storage file management file 12 and the original image storage file 13 is changed at the same time.

FIG. 2 shows a dictionary configuration according to a second embodiment of the present invention. In this embodiment, as shown in FIG. 2 (a), a dictionary is composed only of the dictionary file 21, and the first half of the dictionary file 21 is used as the recognition dictionary unit 22, and the second half is used as the original image storage unit 23. Part 22
, All feature amounts used for recognition are registered, and the original image storage unit 23 stores the original image and its feature amounts.

As shown in FIG. 2B, the recognition dictionary unit 22 includes a header portion of the entire recognition dictionary unit and a plurality of data records. In the header part, the number of features used for recognition (3 in this example)
3) Other information is stored. One data record stores a record number, a character code, the number of original images, a feature amount, and the like, and the data record has a number of feature amounts (in this example, a number of record numbers 1 to 33) used for recognition. .

The original image storage unit 23 includes an offset storage unit and an original image / feature amount storage unit as shown in FIG. 2 (c). The offset storage unit stores an offset amount to the original image storage position corresponding to each feature amount (data record) of the recognition dictionary unit 22. For example, “65” indicates that the offset amount to the original image corresponding to the first (record number 1) feature amount of the recognition dictionary unit 22 is 65 bytes. Similarly, "413
"4" indicates that the offset amount to the original image corresponding to the 33rd (record number 33) feature amount is 4134 bytes. The original image / feature storage unit stores a plurality of sets of the original image corresponding to each feature of the recognition dictionary unit 22 and the feature thereof.

In the recognition processing, the record length of the recognition dictionary unit is calculated from the number of recognition feature amounts (33 in this example) held in the dictionary header of the recognition dictionary unit 22, and the data record corresponding to that length is stored in the memory. To load. Therefore, processing such as skipping of the original image portion is unnecessary, and processing that is the same as the conventional recognition processing can be performed.

The dictionary correction processing is performed as follows. For example, in order to refer to the original image corresponding to the first feature amount of the recognition dictionary unit 22, the original image corresponding to the first offset (65 in this example) stored in the offset storage unit of the original image storage unit 23. Extract and display images. Since the number of original images (the number of samples) corresponding to the feature amount is stored in each data code of the recognition dictionary 22, it is sufficient to extract the number of original images from the offset byte position.

When deleting a feature in the recognition dictionary, the corresponding image in the recognition dictionary unit 22 is deleted, and at the same time, the original image and the corresponding feature of the corresponding part in the original image storage unit 23 are deleted. Correct the offset value of the whole part. Also, when deleting a part of the feature amount that is the basis of the averaged feature amount, the corresponding original image is displayed, and after confirming the user, the corresponding image in the original image storage unit 23 is deleted. The original image and its characteristic amount are deleted, the average is calculated again with the remaining characteristic amount, and the corresponding characteristic amount in the recognition dictionary unit 22 is corrected to a new average value. Also in this case, the offset value is corrected.

To add a feature in the recognition dictionary, the feature is added to the recognition dictionary unit 22 (addition of a data record), and at the same time, the original image and its feature are added to the original image storage unit 23. Correct the offset value. The recognition dictionary unit 22
When a new feature amount is further added to a certain feature amount and the average is taken, an original image to be newly added to the corresponding portion of the original image storage unit 23, the feature amount is added, After correcting the offset, the average of all the target feature amounts in the original image storage unit 23, including the added original image feature amounts, is obtained, and the corresponding feature amounts of the recognition dictionary unit 22 are newly obtained. To the average value.

〔The invention's effect〕

According to the dictionary construction method of the present invention, the following effects can be obtained.

(1) In the claims (1) and (2), the maintenance of the recognition dictionary is facilitated by storing the image from which the feature amount in the recognition dictionary is based and the feature amount of the image. No means for generating the original image based on the characteristic information is required, and there is no restriction on the use of the characteristic amount.

(2) In claim (1), by storing the original image of the feature amount in the dictionary and the feature amount of the image in a separate file,
In the recognition process, only the recognition dictionary can be loaded into the memory and used, and the recognition process can be performed in exactly the same manner as when the original image is not stored.

(3) In claim (2), each feature used for recognition on the recognition dictionary, its original image, and the feature of the image are separately stored, so that processing such as skipping of the original image portion is skipped. Is unnecessary, and a file for managing the original image storage file is also unnecessary.

[Brief description of the drawings]

FIG. 1 and FIG. 2 are diagrams showing an embodiment of a dictionary configuration according to the present invention. 11: Dictionary file, 12: Original image storage file management file, 13: Original image storage file, 21: Dictionary file, 22: Recognition dictionary unit, 23: Original image storage unit.

Claims (2)

(57) [Claims] 1. A recognition dictionary file including at least a character code and a record holding an average feature amount of one or more character images of the character for each character; And an original image storage file for storing an original image of the character and a feature amount of the image corresponding to each record. 2. A recognition dictionary unit and an original image storage unit. The recognition dictionary unit stores, in order for each character, at least a character code and an average feature amount of one or a plurality of character images of the character. The original image storage unit stores, in order for each character, an original image of the character and a feature amount of the image, a dictionary construction method for character recognition.