JPH0228897A - Address area detector - Google Patents

Abstract

PURPOSE:To detect an address-mentioned area efficiently and correctly by deleting a noise area from the feature information of the binary picture of a unit block area and after that, detecting the address-mentioned area from the projective feature of input picture information in the left area. CONSTITUTION:The image of a mail article 1 is picked up by a picture input part 2 and its output is inputted to a feature preparing part 3 and a noise area deciding part 4. The feature preparing part 3 converts this input picture into a block binary picture with the prescribed unit block area as a basis, obtains the feature information of the block binary picture in which information quantity is compressed and obtains the projective feature of the input picture information. An address area deciding part 5 deletes the noise area decided from the feature information of the block binary picture by a noise area deciding part 4 and after that, detects the address-mentioned area from the projective feature of the input picture information in the left area.

Description

[Detailed Description of the Invention] [Purpose of the Invention (Industrial Application Field) The present invention provides a method for quickly and efficiently detecting address writing areas in letters and mail written in various formats and reading addresses. The present invention relates to an address area detection device that can be provided.

(Prior Art) The format of a form that is subjected to character reading by an optical character reader (OCR) is generally determined, and the character reading device reads character information written in a predetermined area of a form. It is configured to read only the following. For this reason, in the past, forms have been prepared exclusively in accordance with a predetermined format, with character entry areas designated by a frame or the like.

In addition, for character reading devices that read documents in various formats, a document is prepared with preprinted ID numbers, ID codes, etc. that specify the format in a predetermined position, and the character reading device reads this ID information. By reading the file, the format is known and the area to be read is set. For example, the area to be read is known as coordinate value data from a reference position previously set in accordance with ID information, and the area is set.

However, for example, in the case of a character reading device that reads the address information of letters and mails, since the format (format) in which the addresses are written in letters and letters itself is not determined, it is necessary to detect the area to be read in advance. This is necessary. Moreover, it recognizes the correspondence, advertisement, sender's address, name and stamp, postmark, postal code, etc. on the surface where address information is written on letters and mail, and detects the address writing area after removing these information. Is required. However, there is a problem in that recognizing each of the above-mentioned types of information and detecting the address writing area while distinguishing it from the written information requires a large processing load, and the processing efficiency is extremely low.

In particular, reading and inputting address information on this type of letter/mail requires a processing capacity of, for example, about 10 pieces per second, so how can the address writing area of letters/postal items be detected quickly and with high precision? , the major issue is how to read and input the address information.

Therefore, the present inventors first divided the input image of letters and mail into predetermined block areas, binarized each block area according to its black pixel appearance frequency to obtain a block binary image, and obtained the block binary image. In addition to the projection information and the knowledge about letters and mail items, we have learned the amount stamp and postal code area at the top of the letter and the sender and postal code at the bottom of the letter, as shown in Figure 11.
We proposed a method that detects the advertisement text area and the addressee area from the input image area excluding these areas. According to such a method, it is possible to identify and detect the information written on letters and mail items while detecting the information in a global manner, so the processing speed (detection processing of the address writing area) can be significantly increased. is played.

However, as shown in FIG. 12, in letters and mail, address information is often written vertically and horizontally, and advertisements are often written on the right side of the address information. In such a case, based on the projection component of the block binary image mentioned above and the knowledge that the sender/advertisement writing area is often written in the lower area of letters and mail, it is possible to accurately determine the advertising writing area. As a result, there have been problems such as the inability to accurately detect the address writing area.

(Problem to be Solved by the Invention) As described above, conventional character reading devices are capable of reading address information on letters and postal items whose format is not determined without referring to ID information that specifies the format. There is a problem in that it is difficult to target the area to be read, and it requires a large processing load to detect the address writing area in letters and mail separately from other information and specify the area to be read. Moreover, there was a problem in executing the processing at high speed.

The present invention has been developed with these circumstances in mind, and its purpose is to fill the address writing area of letters and postal items with no specified format without using ID information or the like to specify the format. It is an object of the present invention to provide an address area detecting device which can detect address area simply, quickly, and with high precision and can read and input address information.

[Structure of the Invention] (Means for Solving the Problems) The address area detection device according to the present invention first inputs information on the side of a letter/postal item on which address information is written, and then inputs an input image into a predetermined image. The image is divided into unit block areas, and characteristic information of the image in each unit block area is obtained, for example, as the projection of the image in the unit block area of the binarized input image and the frequency of appearance of black pixels. Then, each unit block area is treated as one unit and binarized and expressed according to the appearance frequency of black pixels in the image determined for each unit block area, the input image is converted into a block binary image, and the block binary image is projected. Look for characteristics.

Thereafter, character writing areas other than the address writing area in the input image are detected based on the projection characteristics of this block binary image and the knowledge information regarding the letter/postal item, and these detected areas are extracted from the input image area. After deletion, the address writing area in the input image is detected from the projection feature in the remaining area of the input image.

(Function) According to the present invention, an image obtained by capturing and inputting a letter and postal item is divided into predetermined unit block areas, and each unit block area is treated as one unit and binarized into a block binary image. , find the projection features in an arbitrary rectangular area of this block binary image. Then, according to the projection characteristics in an arbitrary rectangular area of the block binary image and the knowledge of letters and mail items, the postal stamp face, postal code section, and advertisement/correspondence areas other than the areas to be detected are detected, and these detection areas are is removed from the input image. Then, the projection information of the remaining area is obtained by adding the projection information obtained for each unit block area, and the address writing area is detected according to this projection information.

Therefore, from the projection information of an arbitrary rectangular area of a block binary image that only indicates whether or not each unit block area plays an important role as an information writing area, various information writing areas of letters and mail can be efficiently calculated from a global perspective. This makes it possible to effectively determine the type of the information writing area according to the knowledge information regarding letters and mail. Then, the information writing area other than the address writing area targeted for detection is removed, and the address writing area is detected from the projection information of the input image in the remaining area, so the format is not determined for letters and mail. You can easily specify the address writing area for reading input inside.
Moreover, it becomes possible to detect at high speed.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic diagram of an embodiment of the apparatus, and ``■'' indicates letters and mails to be read by this apparatus. The image input unit 2, which is an image scanner or a TV left camera, inputs information on the side of the letter/postal item L on which address information, etc. is written, and for example, binarizes the captured input image using a predetermined threshold value. , and is stored in an image memory as a binary digital image and subjected to character reading processing (destination area detection processing) described below.

The feature creation unit 3 includes an image division unit 3a, a projection feature creation unit 3b,
It consists of a black pixel frequency feature creation section 3c, a block binary image creation section 3d, a block projection feature creation section 30, etc., and divides the input image (binary digital image) into predetermined unit block areas and creates the entire image. It is configured to detect and create characteristic information of.

That is, the image dividing section 3a in the feature creating section 3 divides the input image (binary digital image) into unit block areas of a predetermined size. The input image is divided into unit block areas of (tex 1B), for example, by the image dividing section 3a. The projection feature creation section 3b thus obtains the projection components of the black pixels in each unit block area divided by the image division section 3a in the vertical and horizontal directions, and uses these as projection features in each unit block area. There is.

On the other hand, the black pixel frequency feature creation section 3c measures the frequency of appearance of black pixels within the block region for each unit block region, and obtains the measured value as feature information of the unit block region. The black pixel frequency in each unit block area is compared with a predetermined threshold value in the block binary image creation section 3d, and it is determined whether the unit block area is represented entirely by black pixel components or by white pixel components. This will be used to determine whether the This block binary image creation section 3d obtains a block binary image in which each unit block area is expressed in binary terms as one unit. The block projection feature creation unit 3e obtains the projection of the block binary image obtained as described above in the X direction and the Y direction, and uses this as global feature information of the input image.

In other words, the feature creation unit 3 divides each unit block of the input image divided into predetermined unit block regions into "1" or "O" depending on the frequency of appearance of black pixels in that block region.
'', and the projected components of the block binary image in the horizontal direction (X direction) and vertical direction (Y direction) are obtained.At the same time, the input in each unit block area is obtained. The projected components of the image are determined respectively.The projected components of each unit block area are used in the area determination process described later.

Therefore, the noise area determination section 4 uses the feature information obtained by the feature creation section 3, especially the projection information in an arbitrary rectangular area of the block binary image, and the knowledge information obtained in advance regarding the letter and postal item 1. Accordingly, a character writing area other than the addressee writing area to be detected is detected, and this detected area is removed from the input image as a noise area.

Knowledge information regarding the letter postal item 1 includes, for example, that the postal code column in the letter postal item 1 is normally set at the upper right position of the mail piece, and that the area on the stamp face where the stamp is attached is located at the top right of the postal item. It consists of information such as the upper left corner of the mail piece, and the lower area of the mail piece where correspondence or advertisements may be written.

The address writing area determination unit 5 determines the remaining area in the input image from which the noise area has been removed as described above, according to the feature information obtained by the feature extraction unit 3, especially the projection information in each unit block area. , the projection information there is obtained at high speed. Then, the line pitch, character pitch, etc. in the character writing area are determined from the projection and information, and the addressee writing area 6 in the input image is detected according to this information.

FIG. 2 is a diagram showing a schematic flow of the procedure for detecting the address area of the letter/postal item 1 in this apparatus.

This process is performed by continuously reading and inputting images of letters and mail items 1, which are continuously conveyed at a speed of, for example, about 10 per second, using the image input section 2 (step a). The feature creation unit 3 divides the input image read and input in this way, calculates the projection feature in the divided unit block areas, and measures the frequency of appearance of black pixels (Stella, Print, etc.). b). Then, the above-described block binary image is created according to the black pixel frequency information, and projection information of the block binary image is calculated (step C).

After that, using the knowledge information regarding the letter/postal item 1]2, the noise area determining unit 4 determines the advertisement writing area, etc., and the detected area is deleted from the input image area (step d). . This process calculates the projected component in an arbitrary rectangular area of the block binary image, and determines where in the input image the rectangular area detected as the character writing area is located and its size. This is done by checking the above information and comparing it with the above-mentioned knowledge information.

Through this detection process, the charge stamp face/postal code field determination section located at the top of the input image is removed as a noise area, and the advertisement/correspondence/sender name determination section located at the bottom of the input image is removed as a noise area. removed. The detection of the noise region and the removal of the detected region from the input image may be performed step by step, or may be performed in parallel.

Therefore, the address area determination unit 5 calculates the projections determined for each unit block area in the remaining areas of the input image obtained by removing the noise areas (character writing areas not subject to detection) as described above. The acid content is added vertically and horizontally, and the projected components of the input image in the remaining regions are determined (step e). Then, from this projection information, the line spacing information etc. for the character writing area are obtained, and according to the knowledge information regarding the letter/postal item 1 mentioned above, only the character writing area written in the center part of the input image is detected as the address writing area. (step f).

In other words, according to the projection characteristics of the block binary image mentioned above, it is possible to determine whether the unit blower area is an information writing area or not, and how the unit block areas of data "1" are clustered. After removing the area other than the address writing area from a global perspective, the projection information in the remaining area is calculated in a unit process. Detection purpose from information□
The address writing area can be detected easily and efficiently.

Next, to specifically explain the above-mentioned address writing area detection processing, the information (input image) of the address writing surface of a letter/postal item l to be processed is given as shown in FIG. 3, for example, and the image input It is assumed that the unit 2 captures and inputs the image as a binary digital image.

For such an input image, the image division section 3a in the feature creation section 3 vertically and horizontally divides the input image into predetermined unit block areas as shown in FIG. This division is, for example, 400 pixels in the horizontal (horizontal) direction and 8 pixels in the vertical (vertical) direction.
This is done by dividing an input image of 00 pixels into 16 pieces each in the vertical and horizontal directions. By this division, a (L6×1B) unit block is set having a (25×50) pixel area as a unit block area.

The black pixel frequency feature creation unit 3c measures the frequency of appearance of black pixels in each of these unit block areas, and obtains the measurement results as shown in FIG. The block binary image creation unit 3d discriminates the appearance frequency of black pixels in each unit block area shown in FIG. The information of "1" and "0" indicating the presence or absence of the data is converted into binary information.As a result, a blower binary image as shown in Fig. 6 is created, and the input image consists of (L'6XIB) pixels. expressed as data.

For such a block binary image, the projection feature creation unit 3
e is horizontal in the arbitrary rectangular area.

The projected component in the vertical direction is obtained as feature information of the block binary image.

At this time, the projection feature creation section 3b calculates the horizontal and vertical projections of the image information of each unit block area, as shown in FIGS. 7(a) and 7(b). These projected components are temporarily stored in correspondence with the unit block area, and are used in address area determination processing, which will be described later.

FIG. 8 is a diagram schematically showing the concept of noise area determination processing performed using the projection features of such block binary images and knowledge information regarding letters and mail items.

That is, the noise area determination unit 4 detects the unit block of data "1" in the upper part of the block binary image based on the a priori knowledge that the postage stamp/postal code field of the letter/postal item 1 exists in that information part. A group of areas is detected, and the entire area above the detected area is detected as the postage stamp face/postal code field area α as shown in FIG. Then, the area detected as the postage stamp face/postal code field area α is deleted from the blower binary image, and the data of the unit block area where the area was detected is set to "O".

In addition, the noise area determination unit 4 uses the block 2 in which the area detected as the postal stamp/postal code column area α is deleted as described above.
From the value image, based on the a priori knowledge that advertising text, correspondence, sender's name, etc. will be written in the lower area of the letter/postal item, the horizontal projection feature is added to the lower edge of the letter as shown in Figure 8. Check out advertisements and correspondence! Detect the sender name area β. This 4. Advertising text, correspondence, etc. detected in this way
The sender name area β is also deleted from the block binary image.

After determining and detecting the information recording areas α and β other than the address area from a global perspective as described above and deleting them from the block binary image, the address writing area 6 is determined by the address area determination unit 4.
Detection processing is performed.

This process of detecting the address writing area/area 6 is performed by first detecting its vertical projection feature as shown in FIG. seek. The calculation of this vertical projection feature is
This is performed by determining the sum of the vertical projection components determined for each unit block area of the remaining input image area.

From the vertical projection features obtained in this way, for example, information on the line pitch of character strings can be obtained, and areas other than the address writing area can be determined based on information such as the position of the area where the projected component appears in the input image. delete. Specifically, if a character writing area with an uncertain or unstable line pitch is detected on the right edge of a letter/postal item, this is determined to be an advertising text writing area and is deleted from the address area detection target area. . If a character writing area is detected at the left end of the letter/postal item 1, it is determined that it is a sender writing area and is deleted from the address area detection target area. The noise area deletion process according to the projection information is performed on the basic premise that the address writing area is the central area of the letter/postal item 1 and that the address writing area is written in an orderly manner at a substantially constant line pitch.

Thereafter, the horizontal and vertical projection features of the remaining input image area are obtained as shown in Figure 10, and the rectangular area specified by the projection pattern is detected as the address writing area 6.

As a result, the address writing area 6 of the letter/postal item can be accurately determined as a rectangular area shown by the projection pattern in the central area left after the noise area has been deleted.

In this way, according to this device, the input image is treated as a block binary image whose information content has been significantly compressed, the characteristic information of the input image is determined, noise areas are eliminated from a global perspective, and the input image in the remaining area is Since the address writing area 6 is detected from the projection feature of , it becomes possible to detect the address area of a letter or postal item easily and quickly. Furthermore, it is possible to accurately detect only the address writing area from an input image of a letter/postal item in which one of various pieces of information other than address information is written.

As a result, great practical effects can be achieved, such as the ability to improve the reliability of the address information reading and input process.

Furthermore, since the projection information is obtained for each unit block area when an image is input, the series of processes described above can be efficiently pipelined, and the processing speed can be significantly improved. It has great practical effects, such as the ability to

Note that the present invention is not limited to the embodiments described above. Here, the input image was converted to a (LAX 1B) block binary image and address detection processing was performed, but the number of block divisions and the threshold for obtaining the block binary image are It may be determined according to the specifications. Further, here, the feature information was obtained from the appearance frequency of black pixels in the unit block area, but it is of course possible to obtain the appearance frequency of white pixels as the feature information. Alternatively, it is also possible to block the multi-valued input image as it is, obtain its density histogram, and detect features. In addition, the knowledge information used for various area detections is not limited to the above-mentioned examples. It is also possible to apply the area detection processing according to the present invention to IC inspection, etc. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Effects of the Invention] As explained above, according to the present invention, a captured input image of a letter and mail is converted into a block binary image based on a predetermined unit block area, and this block 2 whose information amount is compressed is
After removing the noise area from the feature information of the value image, the address writing area is detected from the projection characteristics of the input image information in the remaining area, so it is possible to write the address in letters and mail easily, efficiently, and at high speed. It becomes possible to accurately detect the area, and great practical effects such as improving the reliability of address information reading and input processing can be achieved.

[Brief explanation of the drawing]

1 to 10 show an embodiment of the address area detection device according to the present invention, FIG. 1 is a schematic configuration diagram of the embodiment device, and FIG. 2 is a schematic diagram of the processing procedure in the embodiment device. Figure 3 is a diagram showing an example of the format of letters and mail; Figure 4 is a diagram showing the format of letters and mail.
Figure 5 shows the concept of blocking input images.
The figure shows an example of information on the appearance frequency of black pixels in a unit block area, Figure 6 shows an example of a block binary image, and Figure 7 schematically shows the detection of projection features in a unit block area. , Fig. 8 is a diagram showing the concept of detecting a noise area from a block binary image and deleting the area, and Figs. 9 and 10 respectively show the process of detecting an address area from the projection characteristics of the image information of the remaining area. It shows. FIG. 11 and FIG. 12 are diagrams each showing an example of the format of letters and mail items to be processed. ■...Letter/postal item (reading target) 2...Image input section, 3...Characteristic creation section, 3a...Image division section, 3
b... Projection feature creation unit, 3C... Pixel frequency feature creation unit, 3d... Block binary image creation unit, 3e... Block projection feature creation unit, 4... Noise area determination unit, 5・
. . . Address area determination unit, 6 . . . Address area. Figure 9 Figure 11 Figure 10 Figure 12 Figure

Claims (1)

[Claims] An image input unit that inputs information on the side where address information of letters and mail is written, and an image input unit that divides the input image into predetermined unit block areas and calculates projected components and black pixels of the image in each unit block area. a means for determining the frequency, and
means for obtaining a block binary image that is binarized and expressed with each unit block area as one unit according to the black pixel frequency obtained for each unit block area;
means for determining a projection feature of a value image, and detecting character writing areas other than address writing areas in the input image based on the projection characteristics of the block binary image and knowledge information regarding the letter/postal item; An address area detection device comprising: means for deleting an area from the input image area; and means for detecting an address writing area in the input image from the projection feature in the remaining area of the input image. .