JPH10328623A - Postal address recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a postal address recognition device in which detection precision of an address region is raised and recognition fate of an addressee, an address and a character string is enhanced. SOLUTION: The postal address recognition device is equipped with an image input part 12, an image memory 13 storing image data fetched in the image input part 12 and an image processing part (a recognition processor 14, a character recognition dictionary 15, a work memory 16 and an address dictionary 17) performing address region cut-out, character cut-out, character recognition and adress collating for the image data stored in the image memory 13. Furthermore, when a tack seal, on which the address character string of an addressee is printed, is stuck on a postal item, the postal address recognition device is provided with both an edge detection means for detecting the edge of the tack seal and an address region judgement means for judging the address region by the position of the edge. Both a method for utilizing variable-density difference of ground color of the postal item and the tack seal and a method for utilizing thickness of the tack seal are used as the edge detection means.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a postal address recognition device for recognizing an address character string of an address described in a postal matter, and more particularly, to a method for detecting an address character string region accurately in a short time. Mail address recognition device capable of.

[0002]

2. Description of the Related Art In a postal address recognizing apparatus, it is necessary to detect an area in which a postal address is described in order to recognize the address character string of the postal address described in a mail.
Conventionally, when this address area is detected, there is a method in which a black-and-white changing point is obtained using differentiation in the horizontal and vertical directions, and the black-dot changing point is calculated by projecting black dots. With the method, it is difficult to distinguish between an address character string of an address, an advertisement and a sender address.

[0003] Also, as a related patent relating to address area detection of a postal address recognition apparatus, for example, Japanese Patent Application Laid-Open No. Hei 7-93474.
There is one using a labeling process as described in Japanese Unexamined Patent Publication (Kokai) No. H10-26095. Here, labeling is performed on the connection of the black dots in the binary digital image read from the postal matter, the obtained labels are integrated, and it is determined whether the address of the integrated labels is a handwritten stamp. Detects the address area by extracting only the label with the size of the handwriting or stamp candidate size, detecting the area where the extracted label concentrates, and determining whether the area conforms to the address description format. I have to. However, even in this case, it is difficult to accurately distinguish the destination address character string from the advertisement, the sender address character string, and the like.

[0004]

In the prior art described above, the addressing area of a postal matter is determined by a method of projecting black dots in the former method and a method of labeling a set of black dots in the latter method. However, as described above, there is a problem that it is difficult to distinguish the address and address string from the advertisement and the sender address string and the like as described above. Also, for tilted images,
When black dot projection processing is performed, there is a problem that a correct area cannot be obtained because the total in the vertical and horizontal directions is shifted due to the inclination. In addition, both the former method of performing the black dot projection process and the latter method of performing the labeling process require processing for all the black dots of the postal image, so that the processing time is long. there were. SUMMARY OF THE INVENTION The object of the present invention is to solve the above-mentioned problems, and it is possible to distinguish an address character string from an advertisement, a sender address character string, etc., which is effective even for a tilted image, and is accurate in a short time. Another object of the present invention is to provide a postal address recognition device capable of recognizing postal addresses.

[0005]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention utilizes a fact that a character string printed on a sticker is a normal address and address character string, and detects the sticker. This recognizes the mail address of the mail.

More specifically, an image input unit (12) for receiving image data of a mail, and the image input unit (1)
An image memory (13) for storing the image data captured in 2), and an image processing unit (105) for extracting an address area, extracting characters, recognizing characters, and comparing addresses with respect to the image data stored in the image memory (13). Recognition processor 1
4, character recognition dictionary 15, work memory 16, address dictionary 1
In the case where the postal address recognition device having 7) is further attached with a tack seal (including cellophane) on which the address character string of the address is printed, the edge detecting means (for detecting the edge of the tack seal) 3 to 5) and a destination area determining means (by the recognition processor 14) for determining a destination area based on the position of an edge. As the edge detecting means, the background color of the postal matter and the density difference of the tack seal are used to scan the postal matter to which the tack seal is affixed to detect the position of the density change point, and based on the position of the density change point. And a method of detecting the position of a shadow caused by light projected on a postal matter to which a tack seal is attached, using the thickness of the tack seal, and detecting the edge based on the position of the shadow. There is.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the present invention, when an address character string of an address is printed on a tack seal (including cellophane), an edge detecting means for detecting the edge of the tack seal, and an edge Addressing area determining means for determining an addressing area according to the position of the addressing area is provided. As the edge detecting means, the background color of the postal matter and the density difference of the tack seal are used to scan the postal matter to which the tack seal is affixed to detect the position of the density change point, and based on the position of the density change point. Or a method of detecting the position of a shadow caused by light projected on a postal matter with a tack seal using the thickness of the tack seal, and detecting the edge based on the position of the shadow. I use. This configuration utilizes the fact that the character string printed on the tack seal is a normal address / address character string. Mail,
In particular, the address / address character string of direct mail is very effective in many cases using a tack seal. Also,
The inclination of the tack seal is obtained by the inclination of the edge. Since the inclination of the sticker matches the inclination of the character string, by performing the inclination adjustment of the character string, a highly accurate recognition process can be performed.

Hereinafter, an embodiment of the mail address recognition apparatus of the present invention will be described in detail with reference to the drawings. FIG. 1 is an overall block diagram for explaining a postal address recognition apparatus of the present invention. As shown in FIG. 1, a postal address recognition device 1 of the present invention
Reference numeral 0 denotes an image input unit 12 for reading and inputting image data from a CCD scanner or the like, an image memory 13 for storing the read image data, a recognition processor 14 for recognizing characters from the image data, and storing a standard character pattern for character recognition. It comprises a character recognition dictionary 15, a work memory 16 for storing intermediate results and the like, and an address dictionary 17 for storing addresses. In addition, a character string of an address and address is described on the mail 11.

Next, the processing of the postal address recognition device 10 will be described. First, a mail item 1 with an address / address string
1 is input from the image input unit 12 and stored in the image memory 13. The stored image data is subjected to address recognition, a character string to be recognized, a character direction, a character-by-character cutout, etc., using a recognition processor 14, and then a character recognition dictionary 15 is used to perform character recognition. I do. The character recognition result is compared with an address string by using the address dictionary 17 and the recognition result is stored in the work memory 16. In this way, the mail address is recognized. The address string stored in the work memory 16 is converted into a zip code or the like, and is used for sorting mails.

FIG. 2 shows an image input unit 12 useful for the present invention.
1 is an example of image data of a mail item input from the Internet. In the figure, reference numeral 20 denotes an image of a postal matter (hereinafter, simply referred to as a postal matter), reference numeral 21 denotes an image of a tack seal (hereinafter, simply referred to as a tack seal), and reference numeral 22 denotes an image of an address / address character string printed on the sticker (hereinafter, simply referred to as a sticker). Address mail address string),
Reference numerals 23 and 24 denote images of the sender address character string and the postal code which are printed on the mail 20 in advance.
Conventionally, the area of the destination address character string 22 and the area of the sender address character string 23 have the same rank, and when performing character recognition, it is difficult to determine which character string is the address character string to be recognized. . In the present invention, the recognition process of the image 22 of the correct address / address character string can be performed by detecting the frame of the image 21 of the tack seal using the shading change point or the like.

FIGS. 3 to 5 show an example of a destination area detecting method according to the present invention. FIG. 3 is a view for explaining a method of detecting the tack seal 21 from the mail 20 read by the optical scanning by the image input unit 12 and stored in the image memory 13, and FIG. It is a shading change point table obtained by reading. In order to detect the tack seal 21, it is necessary to detect the edge of the tack seal 21. For this reason, for example, a difference in density is detected by focusing on a difference in color. That is, in FIG. 3, the image data stored in the image memory 13 is not simply data expressed in white or black (binary data), but data having shades (multi-valued data). In FIG. 3, it is assumed that the background is at the gray level 0, the mail 20 is at the gray level 7, and the tack seal 21 is at the gray level 0. Reading from the image memory 13 is performed in the scan horizontal direction 31 (x coordinate axis direction) and the scan vertical direction 32 (y coordinate axis direction).
The gray level change point 33 is detected from the read data, stored in the gray level change point table, and then the edge is detected based on the gray level change point table.

An example of a method for detecting a gray-scale change point from multi-value data stored in the image memory 13 will be described. On the multivalued data pattern in the image memory, the x-coordinate axis is set in the horizontal direction and the y-coordinate axis is set in the vertical direction. The density change points are detected along the x-coordinate axis direction and the y-coordinate direction, and registered in the density change point table.

That is, first, multi-value data is read in the x-coordinate direction, and when multi-value data, that is, a point at which the density has changed is detected, the coordinate value at that time, the density after the change, and 1 is registered in the horizontal flag indicating that the detection is performed at the time of reading in the x coordinate axis direction. This process is repeated by changing the y coordinate value from 0 to the storage width Y of the image memory corresponding to the width of the mail.

For example, the scan horizontal direction 3 with the y coordinate value b
When reading 1 (x-coordinate axis direction), the x-coordinate value 0 changes the gray level from 0 to 7, and the x-coordinate value a changes the gray level to 7
From 0 to 0. Therefore, in the gradation change point table, 0 is stored in the x coordinate value, b is stored in the y coordinate value, 7 is stored in the gray level, 1 is stored in the horizontal flag, and a is stored in the x coordinate value, b is stored in the y coordinate value, and the gray level is stored. 0 and 1 are stored in the horizontal flag. Similarly, the scan horizontal direction 31 is calculated using the y coordinate value (b + i).
(X-coordinate axis direction), the x-coordinate value 0
Changes the gray level from 0 to 7 at x-coordinate value a. Accordingly, 0 is stored as the x coordinate value, (b + i) is stored as the y coordinate value, 7 is stored as the gray level, and 1 is stored as the horizontal flag, and the x coordinate value is a and the y coordinate value is (b + i). , 0 as the gray level, and 1 as the horizontal flag. At this time, when the y-coordinate value is changed and the repetition is performed, if the shading changes at the same x-coordinate value as the previous time, the registration in the shading change point table is omitted, so that the shading change point table is saved. Uses less memory space.

Next, the multivalued data is read in the y-coordinate axis direction, and when the multivalued data, that is, the point where the density is changed is detected, the coordinate value at that time, the density after the change, and y 1 is registered in the vertical flag indicating that the detection is performed at the time of reading in the coordinate axis direction. Next, this processing is repeated while changing the x coordinate value from 0 to the storage width X of the image memory corresponding to the length of the mail.

For example, in the x-coordinate value c, the scanning vertical direction 3
2 (in the y coordinate axis direction), when the y coordinate value is 0, the gray level is changed from 0 to 7, and when the y coordinate value is d, the gray level is 7
From 0 to 0. Therefore, in the gray level change point table, c is stored in the x coordinate value, 0 is stored in the y coordinate value, 7 is stored in the gray level, and 1 is stored in the vertical flag. 0 and 1 are stored in the vertical flag. Similarly, the scan vertical direction 3 is determined by the x coordinate value (c + j).
2 (y coordinate axis direction), the y coordinate value is 0
Changes the gray level from 0 to 7 and the gray level from 7 to 0 at the y-coordinate value d. Therefore, in the gray point change point table, (c + j) is stored in the x coordinate value, 0 is stored in the y coordinate value, 7 is stored in the gray level, and 1 is stored in the vertical flag, and (c + j) is stored in the x coordinate value and d is stored in the y coordinate value. , 0 in the gray level, and 1 in the vertical flag. At this time, when the x coordinate value is changed and repeated, if there is a change in the shade with the same y coordinate value as the previous time, the registration in the shade change point table is omitted, so that the shade change point table is saved. Uses less memory space. FIG. 4 shows an example of the shading change point table generated in this way.

The above-described processing, that is, the scanning horizontal direction (x-coordinate axis direction) and the scanning vertical direction (y-coordinate axis direction) are performed on the multivalued data (shading) of the image memory obtained by optically scanning the mail. By repeating the above-mentioned processing, a gray level change point table as shown in FIG. 4 is created. The coordinates of the edge of the gradation change point (the horizontal and vertical positions of the change point) are obtained by referring to the gradation change point table created in this manner. The shape and size of the part different from the ground color of the object can be recognized.
If the shape recognized in this way is a substantially quadrilateral and the size is equivalent to the address column (address, name) of ordinary mail (for example, about several cm × (5 to 15) cm) ,
It is considered that there is a high possibility that a portion where the shading is different from the ground color of the mail is a tack seal, and an area inside the portion is set as a destination area. In addition, if the shape is small or has a complicated or irregular shape, it is considered that there is a high possibility of an advertisement or a mark, and is excluded from candidates for the address area. According to this embodiment, it is easy to detect a mailing address area.

The above-described method of detecting the edge of the tack seal is effective when the ground color and the tack seal have a difference in shading. For example, when the ground color is white and the tack seal is also white, the shading difference between the two is used. If there is no, detection by concentration cannot be performed. Therefore, there is a demand for a tack seal edge detection method that is effective even when there is no difference in shading between the ground color and the tack seal. As one of the methods of detecting the edge of the tack seal, when the tack seal has a certain thickness or more, the following method of detecting the edge of the tack seal can be considered.

FIG. 5 is a diagram for explaining an embodiment in which the edge of the tack seal is detected using the thickness of the tack seal. As shown in FIG. 3A, the area without the tack seal has a flat surface, but has a thickness 51 only at the tack seal 22. For this reason, when inputting image data, light sources 521 to 524 are provided obliquely above and to the left, right, up and down of the mail 20 for area detection. As shown in FIG. 3B, when light from the light sources 521 to 524 is directed obliquely to the mail, a shadow 53 of the tack seal is generated. This shadow 53 is detected and input as image data for area detection. Since the area detection image data thus input basically appears as a line only in a thick portion, the detection of the tack seal is easy.
The address area can be detected in the same manner as described above depending on the detected shape and size of the tack seal. That is, the detected shape is substantially quadrilateral and the size is equivalent to the address column (address, name) of ordinary mail (for example, several cm ×
(Approximately 5 to 15 cm), the area inside the area is the destination area. In addition, if the shape is small or has a complicated or irregular shape, it is considered that there is a high possibility of an advertisement or a mark, and is excluded from candidates for the address area. According to the present embodiment as well, it is easy to detect a mailing address area.

FIG. 6 is a flowchart for explaining the flow of an address recognition process related to the present invention. First, a tack seal edge generation process is performed on the mail image data stored in the image memory 13 (step 61). This can be performed by using the method of detecting the gray level change point shown in FIGS. 3 and 4, or the method of detecting the shadow based on the thickness of the tack seal shown in FIG. The shape and size of the edge of the tack seal generated in this way are analyzed to determine the presence or absence of the tack seal (step 62). If it is determined that the area is a tack seal, the inside of the tack seal is set as a destination area (step 63). In this case, as described in the description of FIGS. 3 to 5, when the size of the area surrounded by the edge is small or the shape is complicated, the area is excluded without being the destination area, and another area is excluded. As a candidate. If there is no tack seal, another address area detection process (step 64), for example, a black dot projection process (step 6)
5) Perform detection and the like. The detected area is set as a destination area (step 66).

Next, a character cutout process is performed on the detected address area using, for example, the connection of black dots (step 67). Next, for example, four types of rotation processing of 0 °, 90 °, 180 °, and 270 ° are performed on one character of a character candidate, and the character recognition dictionary 15
By performing the matching process, the character direction determining process is performed by setting the direction having the highest similarity as the character direction of the character string (step 68). This character direction determination processing is performed on a plurality of cut-out characters, and the direction of the tack seal is determined by majority decision. In the above description, rotation processing of four types of angles is performed and matching processing with the character recognition dictionary 15 is performed. However, by making the rotation angle finer and performing more types of rotation processing, the tack seal is attached diagonally. Even if it is, it becomes possible to recognize the address / address character string with high accuracy. After the determination of the character direction, the character recognition process of the address character string is performed by the matching process with the character recognition dictionary 15 (step 69). The character recognition result for each character is subjected to an address string collation process using the address dictionary 17 (step 70), and the recognition result is output to the work memory 16.

If the address cannot be compared in the address string matching process in step 70, it is considered that the area does not include an address, and the same processing is performed for another area. That is, it is also possible to adopt, as the candidates for the destination area, those which are excluded as inappropriate because the size of the edge is too small or the shape is complicated or irregular in step 63.

[0023]

According to the postal address recognition apparatus of the present invention, it is possible to distinguish between an address character string and an advertisement or a sender address character string, which is effective for a tilted image.
It is possible to accurately recognize the mail address in a short time.

[Brief description of the drawings]

FIG. 1 is a block diagram of an entire postal address recognition device according to the present invention.

FIG. 2 is an image example of a postal matter effective for the present invention.

FIG. 3 is a diagram for explaining a destination area detection method according to the present invention.

FIG. 4 is an example of a shading change point table according to the present invention.

FIG. 5 is a diagram for explaining a destination area detection method using a tack seal shadow according to the present invention.

FIG. 6 is a flowchart showing a flow of processing in the present invention.

[Explanation of symbols]

10: Postal address recognition device, 11: Mail, 12: Image input unit, 13: Image memory, 14: Recognition processor, 1
5: character recognition dictionary, 16: work memory, 17: address dictionary, 20: postal matter (image), 21: tack seal (image), 22: address / address character string (image), 23: sender address character Column (image of), 24: Zip code (image of),
521 to 524: light source, 53: shadow

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Naoji Ikeda 1-280 Higashi Koikekubo, Kokubunji-shi, Tokyo Inside the Central Research Laboratory, Hitachi, Ltd. Inside the Central Research Laboratory

Claims (3)

[Claims] 1. An image input unit for receiving image data of a postal matter, an image memory for storing the image data captured by the image input unit, and an address area cut out from the image data stored in the image memory. Cut out,
In a postal address recognition device having an image processing unit for character recognition and address collation, an edge detection means for detecting an edge of a tack seal on which an address is printed, and a destination area is determined based on the edge detected by the edge detection means. A postal address recognition device, further comprising an address area determining means for performing the operation. 2. The method according to claim 1, wherein the edge detecting means scans the postal matter to which a sticker is attached to detect a position of a shading change point, and detects an edge based on the position of the shading change point. Postal address recognition device. 3. The mail according to claim 1, wherein the edge detecting means detects a position of a shadow generated by light projected on the postal matter to which the tack seal is attached, and detects an edge based on the position of the shadow. Address recognition device.