JPH0264882A - Address reading device - Google Patents

Abstract

PURPOSE:To efficiently and correctly read and recognize address information by providing a character detecting and segmenting part to segment a character picture into one character unit according to the information of marginal distribution and the character detecting and segmenting part to segment the character picture into one character unit by using the method of labeling. CONSTITUTION:The first character detecting and segmenting part 3 using the method of marginal distribution and the second character detecting and segmenting part 4 using the method of labeling are provided, and the characteristic processing functions of them are used effectively and respectively. Namely, a character part in the address information up to the name of a town in which there is no possibility of the appearance of the Arabic numerals is processed by the first character detecting and segmenting part 3 so as to segment a character, and on the contrary, as for the character part after the name of the town in which there is the possibility of the appearance of the Arabic numerals, the character is segmented by using the second character detecting and segmenting part 4. Thus, a KANJI (Chinese character) part and an Arabic numeral part can be precisely segmented into every one character so that it can be submitted for character recognizing processing, and even the address information in which the KANJI and the Arabic numerals are mixed and written vertically can be correctly and efficiently read and recognized.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to an address reading device that can efficiently and accurately read and recognize address information written vertically on a form.

(Prior Art) In recent years, there has been remarkable progress in character recognition processing technology using optical character reading, and various devices for reading character information written on forms have been developed. This type of character reading device generally inputs character information written on a character-by-character basis into an area predetermined by a character frame etc. on a form as an image, and extracts characteristics of the character from the character image. However, recently, individual character images have been cut out from images of character strings written freely (free format) on a form, and the characters have been recognized.

A typical example of such a character reading device is an address reading device that reads address information written on letters, parcels, business cards, and the like. In this type of address reading device, accurately extracting each character that makes up a character string representing an address from an input image from a form is an important factor that affects recognition performance in character recognition processing. It becomes a factor.

Generally, characters are extracted from a form image in which character strings are written in a free format using image processing techniques such as marginal distribution and labeling. However, there are various formats for writing addresses on forms, and in particular, in the case of addresses written vertically, only the lower number and street address information are often written horizontally using arithmetic numerals. In other words, in an address string written vertically in kanji, multiple digits of arithmetic numerals may be written in the same direction to indicate the lower address.

When reading and recognizing addresses that contain a mixture of vertically written kanji characters and horizontally written arithmetic numeric strings with the same character orientation, the above-mentioned marginal distribution method is used to individually separate character images. If cut out, a multi-digit arithmetic numeric string written horizontally will be detected as a single character, causing recognition rejection or erroneous recognition in character recognition processing. In this regard, if a labeling method is used, it becomes possible to cut out each character of a horizontally written multi-digit arithmetic numeric string and use it for recognition processing.

However, if a labeling method is used, there is a risk that the kanji hen and ro will be separated as independent characters, which may cause misrecognition.

(Problem to be Solved by the Invention) When reading and recognizing vertically written address information, a method of marginal distribution in image processing is used to cut out and extract individual character images from the input character image. In this case, it is not possible to individually extract the horizontally written multi-digit arithmetic numerals that appear in the character string, and when labeling methods in image processing are used, it is difficult to separate There was a problem in that the characters were separated and cut out as separate characters. For this reason, a major problem remains: how to accurately recognize address information written vertically on shipping labels and the like.

The present invention was made in consideration of these circumstances, and its purpose is to provide an address reading device that can efficiently and accurately read and recognize address information written vertically in a free format. Our goal is to provide the following.

[Structure of the Invention] (Means for Solving the Problems) An address reading device according to the present invention cuts a character image into one character size from a character string image indicating vertically written address information according to information on its peripheral distribution. a first character detection cutting section that outputs a character;
and a second character detection and extraction unit that cuts out a character image into one character size from the character string image using a labeling method, and includes, for example, a character portion up to a town name in address image information according to knowledge information written in an address. , the first character detection and extraction section performs character extraction processing, and the character portions that may be written in arithmetic numerals such as blocks and house numbers after the town name in the address image information are processed as described above. The second character detection and extraction unit performs character extraction processing, and the first and second character detection and extraction units perform character recognition processing on the obtained character images, thereby converting the vertically written address into It is characterized by being able to read and recognize efficiently and accurately.

(Operation) According to the present invention, the characteristic processing functions of the first character detection and extraction section using the marginal distribution method and the second character detection and extraction section using the labeling method are made effective. The first character detection and extraction unit is used to extract the characters in the address information up to the town name where there is no possibility of an arithmetic numeral appearing, and conversely, the character part up to the town name where there is no possibility of an arithmetic numeral appearing is extracted. The second character detection and extraction unit is used to extract the characters after the town name in the address information, so the kanji characters and arithmetic numeric characters are extracted character by character with high accuracy and then processed for character recognition. can be provided. As a result, even if the address information is written vertically with a mixture of Chinese characters and arithmetic numerals, it is possible to read and recognize it accurately and efficiently.

(Embodiment) Hereinafter, an address reading device according to an embodiment of the present invention will be described with reference to the drawings.

Fig. 1 is a schematic diagram of the main parts of the device according to the embodiment, and ■ is composed of an image sensor and a TV right camera, and photoelectrically converts written information from forms such as letters and parcels with address information written on them. It is a photoelectric conversion unit that is read manually. The form image information imaged and input from this photoelectric conversion unit l is stored in the image memory 2.
The address information is stored in the address information and subjected to the address information reading recognition process described below.

The feature of this device is that it processes address information from the input image stored in the image memory 2 in order to recognize and process individual characters as a preliminary step to recognizing the address information written on the form. As a character detection and extraction unit for detecting and cutting out individual character images that make up the character string that represents the character string, the first one uses the marginal distribution method in image processing and cuts out the character image into individual characters according to the information on the peripheral distribution. The present invention includes a character detection and cutout section 3 and a second character detection and cutout section 4 that cuts out a character image character by character from an input image using a labeling method in image processing.

This first character detection and extraction section 3 extracts a character image using the marginal distribution, basically by obtaining the vertical and horizontal projected components of the input image, and extracting character strings according to these projective characteristics. Find the block area where is written in each string.

Then, a projected component perpendicular to the character string direction of the character string area is determined, and based on this projected component, the boundaries of each character area are detected from the character string area to determine each character writing area. . In this case, the character writing pitch may be estimated from the projected component, and the character string writing area may be divided according to this character pitch to detect individual character areas.

Furthermore, the extraction of character images using labeling in the second character detection and extraction section 4 basically involves dividing the input character image into two parts.
This method detects a cluster area of connected black pixels as one character writing area based on the connection relationship of black pixels indicating a character writing part in the binarized image. In this case, the inside of a rectangular frame circumscribing the character writing area indicated by the connection of black pixels may be detected as one character area.

The character recognition section 5 then performs character recognition processing on the character images detected by the first or second character recognition sections 3 and 4, and outputs the recognition result to the address recognition section 6. Then, the address recognition section 6 recognizes the address meant by the string of characters from the string of character recognition results given from the character recognition section 5.

This address recognition involves, for example, checking whether the character strings of the recognition results constitute a place name such as a prefecture name or a city, town, village name, etc. for each predetermined group. In addition, in the case of independent text information that does not have any meaning as a place name, it can be recognized by checking whether it has a meaning as a lower number or a street address. It recognizes information such as addresses and street addresses.

In particular, when the recognized characters are arithmetic numerals,
It is recognized as indicating address information.

To explain in more detail the character recognition processing performed by the first and second character detection and extraction sections 3 and 4 and the character recognition section 5, the character recognition section 5 first performs the The character detection and cutting section 3 of No. 1 is activated to cut out a character image one character at a time starting from the first character of the character string image. Specifically, for example, as shown in FIG. 2, a given character string image A of a vertically written address is cut out based on the vertical projected component B of the input image stored in the image memory 2, and then the character string image Lateral projected component C of A
Detect individual character breaks from .

Then, character images are cut out one character at a time starting from the first character part. In this case, "Higashi", "Kyoto", "Rto"...
Character images are cut out in this order. And this first
Cutting out of the character image by the character detection and cutting section 3 is repeated according to the aforesaid knowledge until the characters up to the town name of the address information are cut out.

However, in the character recognition section 5, the first
The character detection and extraction unit 3 performs character recognition processing on the character image extracted using the marginal distribution method, and the address recognition unit 6 extracts address information (prefecture name and municipality name) according to the recognition result (recognized character candidates). ) is recognized.

After that, when the characters up to the town name have been extracted and recognized, the second character detection and extraction unit 4 is activated, and instead of the character detection and extraction performed by the first character detection and extraction unit 3, Character detection and extraction is performed by the second character detection and extraction section 4 using a labeling method. By detecting and cutting out characters using this labeling method,
For example, horizontally written arithmetic numbers that appear in a vertically written character string are individually cut out. Specifically, in FIG. 2, "24" is not cut out as a single character, but r2
They are each cut out as two independent characters: J r4J. Then, the character recognition unit 5 performs character recognition processing on the character image detected and extracted by the second character detection and extraction unit 4, and the address recognition unit 6 uses the recognition result to indicate the lower number and street number after the town name in the address information. Recognize address information.

FIG. 3 shows the flow of the above-mentioned processing procedure. As shown in this processing procedure, in this device, the first character detection and extraction unit 3 extracts character images using the marginal distribution method in order from the first character (step al), and then extracts the characters in the extracted character image. Recognition processing is performed (step a2).

Then, recognition processing of address information up to the town name is performed according to the character recognition result (step a3), and when address information up to the town name' is obtained (step a4), the second character detection and cutting section 4 is activated. Ru.

Regarding the address information after the town name, the second character detection and extraction section 4 extracts the character image using a labeling method (step a5), and character recognition processing is performed on the extracted character image. (Step a8).

Thus, according to such address information reading recognition processing, for the address part up to the town name in the vertically written address information where there is no possibility of horizontally written arithmetic numerals appearing,
Since character recognition processing is performed on the character image obtained by character detection extraction using marginal distribution, it is no longer necessary to separate and extract the kanji hen and ro and recognize them as separate characters. It becomes possible to accurately recognize individual character information.

In addition, for the characters after the town name in address information, the characters are detected and extracted using a labeling method, so even if a string of arithmetic numbers written horizontally appears in a string of vertically written characters, Since each of these characters can be cut out independently, it is possible to cut out each character independently, so it is possible to
It becomes possible to read and recognize address information accurately.

As a result, it becomes possible to accurately read and recognize the address information by effectively utilizing the characteristics of character detection and extraction based on the peripheral distribution and labeling techniques.

It is also possible to control the character detection and extraction processing by the first and second character detection and extraction sections 3-4 as shown in FIG. That is, first, the first character detection and extraction unit 3 extracts a character using the marginal distribution method (step bl), and then performs character recognition on the extracted character image (step b2). Then, it is determined whether the character recognition result is valid or not (Step b3)
If the validity is lacking, the second character detection and extraction unit 4 uses a labeling method, assuming that the character image extracted as one character by the marginal distribution method may consist of multiple characters written horizontally. A character image is cut out (step b4), and character recognition processing is performed on the cut out character image (step b5).

The above processing control is repeated in order from the first character part of the character string, and the labeling method is used to detect and extract characters only for character parts that may be multiple horizontally written characters in a vertically written character string. Make it.

As explained above, according to the present device, there is a first character detection and cutting section that cuts out a character image in units of characters according to peripheral distribution information, and a first character detection and cutting section that cuts out a character image in units of characters using a labeling method. Character detection and extraction sections 2 are provided, and the characteristics of each of these character detection and extraction sections are effectively used to extract character images from the input image and perform character recognition processing. It becomes possible to accurately recognize address information consisting of characters, and address information including the street name and street number after the town name where horizontally written arithmetic numbers may appear.

Note that the present invention is not limited to the embodiments described above. For example, it is suitable for recognizing horizontally written alphanumeric characters appearing in a vertically written kanji character string. In addition, various modifications can be made without departing from the gist of the invention.

[Effects of the Invention] As explained above, according to the present invention, even if horizontally written characters may appear in a character string indicating vertically written address information, all of the characters are Each address can be read and recognized with high accuracy, and the address information can be read and input accurately, resulting in great practical effects.

[Brief explanation of the drawing]

The figures show an address reading device according to an embodiment of the present invention, in which FIG. 1 is a schematic diagram of the main parts of the embodiment device, FIG. 2 is a diagram showing an example of a character string indicating address information, and FIG. The figure shows the flow of the processing procedure, and FIG. 4 is a diagram showing a modification of the processing control mode. 1... Photoelectric conversion unit, 2... Image memory, 3... First character detection cutting unit using the marginal distribution method, 4...
・Second character detection and extraction section using a labeling method,
5...Character recognition unit, 6...Address recognition unit. Applicant's agent Patent attorney Takehiko Suzue

Claims (1)

[Claims] The first step is to cut out a character image into individual characters from a character string image indicating vertically written address information according to information on its peripheral distribution.
and a second character detection and extraction section that extracts the character image into individual characters from the character string image using a labeling method, and for the character part up to the town name in the address information. Character recognition processing is performed on the character image obtained in the first character detection and extraction section, and character parts after the town name that may be written in arithmetic numerals such as blocks and house numbers are processed using the second An address reading device characterized in that a character recognition process is executed on a character image obtained by a character detection and cutting section.