JP3715473B2 - Address reader - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an address reading device used in a mail address reading / sorting machine that reads an address written on a mail by a character recognition technology and classifies the mail according to a destination.
[0002]
[Prior art]
A mail address reading and sorting machine has been put to practical use, which reads an address written on a mail by a character recognition technique and classifies the mail according to destination.
[0003]
An outline of the processing of the address reading device used in this mail address reading sorting machine is shown below.
[0004]
First, an image of a mail is input in the image input unit. Next, the area detection unit detects the position of the area where the address is described from the mail image. Next, each line constituting the address is detected from the address area in the line detection unit. Next, each character is detected from the line in the character detection unit. Next, in the character recognition unit, character recognition is performed by comparing each character image with a character recognition dictionary. Finally, in the address recognition unit, the address of the destination is determined by comparing the character recognition result with the address dictionary.
[0005]
For details of these processes, see, for example, Toshiba Review, vo 1.48, No. 7, pp 536-539 (1993).
[0006]
In the above processing, the area detection unit 2 is required to detect the address area with high accuracy from the postal matter. The reason for this is that if an erroneous region is detected by the region detection unit, there is no method for recovering the error in a later process.
[0007]
In particular, mails with cellophane windows open on envelopes and documents with addresses can be seen through the cellophane windows (hereinafter referred to as "mail with windows") P are distributed in large quantities. Therefore, it is particularly important to detect the correct area for such mail pieces.
[0008]
For this reason, the image input unit 1 has a plurality of optical systems, and not only an image used for character recognition (hereinafter referred to as “main image”) but also an image used for detecting a cellophane window. (Hereinafter referred to as “sub-image”) is also entered.
[0009]
The sub-image is obtained by imaging light that is specularly reflected (specularly reflected) from the postal matter. Since strong regular reflection occurs in the cellophane window portion of the mail with a window, the position of the cellophane window can be easily obtained by using the sub-image.
[0010]
The area detection unit obtains a sub-image at the same time as the main image from the image input unit, and uses the sub-image to determine the position of the cellophane window of the mail with cellophane window. And
[0011]
With the method as described above, the address reading device used in the conventional mail address reading sorter device can detect the position of the cellophane window with high accuracy and improve the accuracy of detecting the address description position. It was.
[0012]
[Problems to be solved by the invention]
In the conventional address reader as described above, not only the optical system for obtaining the main image but also the optical system for obtaining the sub-image for detecting the cellophane window is required, which increases the cost of the entire apparatus. It was.
[0013]
Further, in order to use a plurality of different optical systems, it is necessary to align the optical systems, and there is a problem that it takes time and labor to maintain the machine.
[0014]
The present invention eliminates the above-mentioned drawbacks, and enables the cellophane window, that is, the address description position to be detected with high accuracy using only the address reading optical system, and greatly reduces the performance of the apparatus. Another object of the present invention is to provide an address reader that can reduce costs.
[0015]
The present invention also detects the four corners of the cellophane window and the horizontal and vertical edges of the cellophane window, generates a plurality of cellophane window area candidates from the detected information of the four corners and edges, and the inside and outside of the cellophane window area candidates. It is an object of the present invention to provide an address reader capable of selecting a candidate having the highest possibility of a cellophane window by evaluating the difference in the density distribution of the pixels and the edge strength of the edge of the cellophane window.
[0016]
[Means for Solving the Problems]
According to the present invention, in an address reader for reading an address of a medium to be read, capturing means for capturing an image from the medium, information on four corners of a cellophane window candidate from the image captured by the capturing means, and horizontal of the cellophane window candidate First detection means for detecting vertical edge information by filtering, information on the four corners of the cellophane window candidate detected by the first detection means, and information on horizontal and vertical edges of the cellophane window candidate When the density of the inside and outside of the cellophane window candidate area detected by the second detecting means in the image captured by the capturing means is different from the second detecting means for detecting the cellophane window candidate area by Determining means for determining the cellophane window area on the assumption that the paper quality of the medium is different from the paper quality of the contents of the medium; And a reading means for reading the address by the image of the determined cellophane window region by said determining means in the elaborate images.
[0017]
According to the present invention, in an address reader for reading an address of a medium to be read, capturing means for capturing an image from the medium, information on four corners of a cellophane window candidate from the image captured by the capturing means, and horizontal of the cellophane window candidate First detection means for detecting vertical edge information by filtering, information on the four corners of the cellophane window candidate detected by the first detection means, and information on horizontal and vertical edges of the cellophane window candidate If the texture on the inside and outside of the cellophane window candidate area detected by the second detection means in the image captured by the capture means is different from the second detection means for detecting the cellophane window candidate area by white Determined by the determining means for determining the cellophane window area based on the number of changes to black and the determining means in the image captured by the capturing means. And a reading means for reading the address by the image of the cellophane window area.
[0018]
An address reading device according to the present invention includes an image reading means for outputting a digital image pattern based on density data by optically reading and scanning the surface of a medium on which the address is written, and a digital image pattern from the image reading means. And a region detection means for detecting a description area of the address and outputting a binarized pattern based on the digital image pattern of the address description area, and a bevel based on the binarization pattern from the area detection means. Thus, a line detection means for detecting each line constituting the address in the address description area, and a character pattern for each character by projection or the like based on the binarized data for each line detected by the line detection means. The character detection means to be cut out and the character pattern for each character supplied from the character detection means are used as a reference pattern for English letters, numbers, and Chinese characters in the character recognition dictionary. Character recognition means for recognizing a character candidate by a matching method with an address, and address recognition means for recognizing an address by collating a character recognition result supplied from the character recognition means with an address dictionary. The area detection unit uses four corner detection filters, a horizontal edge detection filter, and a vertical edge detection filter for the digital image pattern based on the density data for one screen from the image reading unit, respectively. Cellophane window area candidate detection means for detecting position candidates of four corners, horizontal edges, and vertical edges, and integrating the detected position candidates to detect a cellophane window area candidate; and this cellophane window area candidate detection means For the cellophane window area candidate detected by the above, the cellophane window is determined by the above filter output, density difference, position, size, etc. Each was evaluated candidate location, and a selection means for selecting only likely a cellophane window candidates.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an address reading device used in a mail address reading sorting machine according to an embodiment of the present invention will be described with reference to the drawings.
[0022]
FIG. 1 shows a schematic configuration and schematic processing of an address reader according to the present invention.
[0023]
This address reading device reads an address written on a postal matter P, and as shown in FIG. 1, an image input unit 1, an area detection unit 2, a line detection unit 3, a character detection unit 4, and a character recognition unit. It comprises a part 5, a character recognition dictionary 6, an address recognition part 7, and an address dictionary 8.
[0024]
First, as shown in FIG. 2, the photoelectric conversion unit as the image input unit 1 is a surface for reading characters such as an address (address) written on an envelope as a postal matter P conveyed on the conveyance path 9. Is read and scanned in accordance with the light source that irradiates the postal matter P with light, the lens that guides the reflected light from the postal matter P, and the captured image (original figure) by the light guided through this lens. The light receiving device outputs a digital image pattern. The digital image pattern as the read image is supplied to the area detection unit 2. The digital image pattern is matrix data (multi-value pattern) based on density data.
[0025]
The envelope as the postal matter P is provided with a cellophane window Pa so that the address written on the contents is visible.
[0026]
The area detection unit 2 detects the position of the area where the address is described based on the matrix data as the image of the postal matter P from the image input unit 1, and bitmap data (2 based on the matrix data of the address area) Converted into a value pattern) and supplied to the row detector 3.
[0027]
The row detection unit 3 detects each row constituting the address in the address region by projection or the like based on the bitmap data from the region detection unit 2.
[0028]
In the character detection unit 4, the character pattern for each character is cut out by projection or the like based on the bitmap data for each row detected by the row detection unit 3. The character pattern (bitmap data (binarized pattern)) for each character thus cut out is supplied to the character recognition unit 5.
[0029]
The character recognizing unit 5 recognizes character candidates by matching a character pattern for each character supplied from the character detecting unit 4 with a reference pattern for alphabets, numbers, and kanji in the character recognition dictionary 6. This recognition result is supplied to the address recognition unit 7.
[0030]
The address recognition unit 7 recognizes the address of the destination by collating the character recognition result supplied from the character recognition unit 14 with the address dictionary 8, and the answer as the final result, that is, the address recognition result is Is output.
[0031]
Next, detection of the cellophane window Pa area of the postal matter P in the area detection unit 2 will be described.
[0032]
The cellophane window Pa area is detected by detecting the following features.
[0033]
1. In the matrix data of the postal matter P, the edge of the cellophane window Pa is often visible. This is a shadow of the level difference between the envelope and the cellophane window Pa on the image (see FIG. 3A).
[0034]
2. In the matrix data of the postal matter P, the inside and outside of the cellophane window Pa often have different densities. This occurs when the paper quality of the envelope and the paper quality of the contents are different (see FIG. 3B).
[0035]
3. In the matrix data of the postal matter P, the inside and outside of the cellophane window Pa often have different textures. This occurs when there is a pattern on the envelope or the contents (see (c) of FIG. 3). For example, the texture can be determined based on the number of changes from white to black per 1 cm of the binarized pattern.
[0036]
4). In the matrix data of the postal matter P, a part of the character line inside the cellophane window Pa may be hidden by the position of the envelope contents being displaced. This case is characterized in that the end of the character line is aligned with the edge of the cellophane window Pa (see FIG. 3D). For example, it is possible to detect whether or not the end of the character line is aligned at the start point and end point of each line in the line detection unit 3.
[0037]
The area detection unit 2 includes a memory that stores matrix data for one screen of the postal matter P, a detection filter storage unit that stores various detection filters (9 × 9 pixels), and a matrix data of the memory. By moving and multiplying one pixel at a time by various detection filters in the detection filter storage unit, the filter processing is performed on the entire surface, and a large output is obtained at a position where there is a high possibility that a corner or edge exists. The corners (candidates of the four corner positions of the cellophane window Pa), horizontal edges (positions of the horizontal edges of the cellophane window Pa), vertical edges (positions of the horizontal edges of the cellophane window Pa) based on this peak value ), A detection unit for detecting a cellophane window Pa region candidate by the corner, horizontal edge, and vertical edge detected by the detection unit, and the detected cellophane window P It is constituted by a selection unit for selecting a cellophane window Pa region by evaluating the area candidate.
[0038]
The detection filter storage unit stores a corner detection filter, a horizontal edge detection filter, and a vertical edge detection filter.
[0039]
As the corner detection filter, as shown in FIG. 4A, the upper right corner detection filter, as shown in FIG. 4B, the lower right corner detection filter, as shown in FIG. The upper left corner detection filter, as shown in FIG. 4D, stores the lower left corner detection filter.
[0040]
As the horizontal edge detection filter, a detection filter as shown in FIG. 5 is stored. As the vertical edge detection filter, a detection filter as shown in FIG. 6 is stored.
[0041]
Next, detection of the cellophane window Pa area of the postal matter P in the area detection unit 2 will be described with reference to the flowchart shown in FIG.
[0042]
That is, the matrix data for one screen of the postal matter P is moved by one pixel by the upper right corner detection filter and multiplied, so that the filtering process is performed on the entire surface, and the peak value (the presence of a corner may be present). A large output is obtained at a position having high characteristics), and the upper right corner is detected from this peak value (ST1). The detection result of the upper right corner is output as a coordinate value (corner position information).
[0043]
Next, the matrix data for one screen of the postal matter P is moved by one pixel at a time by the lower right corner detection filter and multiplied, so that the filter processing is performed on the entire surface, and the peak value (the presence of a corner exists). A large output is obtained at a position where there is a high possibility of being present), and the lower right corner is detected from this peak value (ST2). The detection result of the lower right corner is output as a coordinate value (corner position information).
[0044]
Next, the matrix data for one screen of the postal matter P is moved by one pixel at a time by the upper left corner detection filter and multiplied, so that the filter processing is performed on the entire surface, and the peak value (the presence of a corner exists). A large output is obtained at a highly likely position), and the upper left corner is detected from this peak value (ST3). The detection result of the upper left corner is output as a coordinate value (corner position information).
[0045]
Next, the matrix data for one screen of the postal matter P is moved by one pixel by the lower left corner detection filter and multiplied, so that the filter processing is performed on the entire surface, and the peak value (the presence of a corner exists). A large output is obtained at a highly likely position), and the lower left corner is detected from this peak value (ST4). The detection result of the lower left corner is output as a coordinate value (corner position information).
[0046]
Thereby, the candidate of the position of the four corners of the cellophane window Pa is detected by the corner detection filter.
[0047]
Next, the matrix data for one screen of the postal matter P is moved by one pixel by the horizontal direction edge detection filter and multiplied to perform the filtering process on the entire surface, and the peak value (existence of the horizontal direction edge) A large output is obtained at a position where there is a high possibility that the horizontal edge is detected), and a horizontal edge is detected from this peak value (ST5). The detection result of the horizontal edge is output as coordinate values (edge position information) of the two end portions of each edge.
[0048]
Accordingly, the position of the horizontal edge of the cellophane window Pa is detected by the horizontal edge detection filter.
[0049]
Next, the matrix data for one screen of the postal matter P is moved by one pixel by a vertical edge detection filter and multiplied, so that the filter processing is performed on the entire surface, and the peak value (existence of the vertical edge exists). A large output can be obtained at a position where there is a high possibility of being detected), and a vertical edge is detected from this peak value (ST6). The detection result of the vertical edge is output as the coordinate values (edge position information) of the two ends of each edge.
[0050]
Thereby, the position of the edge of the cellophane window Pa in the vertical direction is detected by the vertical edge detection filter.
[0051]
Next, the position information of the corners and edges obtained from the results of the filter processing by the four types of filters is integrated to detect cellophane window position candidates (ST7).
[0052]
That is, in order to obtain the cellophane window position candidate, the following method is used.
[0053]
(A) A minimum combination (see FIGS. 8A, 8B, and 8C) in which a frame can be determined is extracted from the detected corner, horizontal edge, and vertical edge, and this is set as a cellophane window candidate.
[0054]
For example, two corners shown in FIG. 8 (a), one corner shown in FIG. 8 (b), one horizontal edge, one vertical edge, two horizontals shown in FIG. 8 (c) A cellophane window position candidate is obtained by the edge and the two vertical edges.
[0055]
(B) Of the obtained cellophane window candidates, those that are too small or too large are deleted.
[0056]
(C) The overlapping of the remaining cellophane window candidates is examined, and those having a large overlap are integrated (see FIGS. 9A and 9B).
[0057]
For example, the cellophane window candidates 1 and 2 shown in FIG. 9A are integrated into a cellophane window candidate shown in FIG. 9B.
[0058]
Next, each candidate for the cellophane window position is evaluated by each filter output, density difference, position, size, etc., and only those candidates that are likely to be cellophane windows Pa are selected (ST8).
[0059]
The following are used as evaluation criteria.
[0060]
(1) Output values of corner detection filter and edge detection filter The larger this value, the more likely the cellophane window is.
[0061]
(2) Difference between average values of inside and outside cellophane window candidates The larger the value, the more likely the cellophane window is.
[0062]
(3) Difference in texture between the inside and outside of the cellophane window candidate This can be measured by a known texture analysis method. The greater the difference, the more likely the cellophane window is.
[0063]
(4) Number of text lines included inside the cellophane window candidate that are interrupted at the edge of the cellophane window The higher this number, the more likely the cellophane window is.
[0064]
(5) Location of the cellophane window candidate The possibility of a cellophane window increases as it exists in the center of the postal matter P.
[0065]
(6) The closer the cellophane window candidate is to the standard size, the more likely the cellophane window is.
[0066]
The following methods are conceivable for comprehensive evaluation of the above-mentioned plurality of criteria.
[0067]
(1) Each score of each criterion is compared with each threshold value, and the number of criteria that are equal to or greater than the threshold value is regarded as the total score.
[0068]
(2) A total score is obtained by a known neurocomputer technique.
[0069]
(3) A total score is obtained by a known linear discriminant analysis method.
[0070]
In this way, if an overall score is obtained, it is compared with a certain threshold value, and if it is larger than the threshold value, it may be adopted as a cellophane window.
[0071]
By the way, the cellophane window candidate detected by the processing as described above may include a frame area preprinted on the postal matter P. Unlike the cellophane window, there is a high possibility that the character string written in the box is a sender address (see FIG. 10). Therefore, it is important to distinguish between the cellophane window and the surrounding frame. This distinction is made based on the following information.
[0072]
1. If the line is thick, there is a high probability of a frame.
[0073]
2. If the density difference between the inside and outside of the area is small, the possibility of a surrounding frame is high.
[0074]
3. When the position of the area is at the lower right of the postal matter P, the possibility of a surrounding frame is high.
[0075]
As described above, even if there is no special optical system for detecting the cellophane window, it is possible to detect the cellophane window with high accuracy using only the optical system for character recognition, thereby reducing the cost of the apparatus. Is possible.
[0076]
Furthermore, the maintenance work of the optical system is halved and the alignment of a plurality of optical systems is not required, so that the cost of maintenance can be reduced.
[0077]
【The invention's effect】
As described above in detail, according to the present invention, it is possible to detect the cellophane window, that is, the address description position with high accuracy by using only the address reading optical system, and without greatly degrading the performance of the apparatus. In addition, it is possible to provide an address reader that can reduce the cost.
[0078]
The present invention also detects the four corners of the cellophane window and the horizontal and vertical edges of the cellophane window, generates a plurality of cellophane window area candidates from the detected information of the four corners and edges, and the inside and outside of the cellophane window area candidates. It is possible to provide an address reader that can select a candidate having the highest possibility of the cellophane window by evaluating the difference in the density distribution of the pixels and the edge strength of the edge of the cellophane window.
[Brief description of the drawings]
FIG. 1 is a diagram showing a schematic configuration of an address reader according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining a mail piece and an image input unit.
FIG. 3 is a view for explaining the characteristics of a mail piece with a cellophane window;
FIG. 4 is a diagram illustrating an example of a corner detection filter.
FIG. 5 is a diagram showing an example of a horizontal direction edge detection filter.
FIG. 6 is a diagram showing an example of a vertical edge detection filter.
FIG. 7 is a flowchart for explaining detection of a cellophane window area of a mail in the area detection unit.
FIG. 8 is a diagram for explaining a minimum combination example that can determine the position of a cellophane window candidate;
FIG. 9 is a diagram for explaining cellophane window candidate integration;
FIG. 10 is a diagram for explaining the distinction between a cellophane window and a surrounding frame.
[Explanation of symbols]
P ... Mail Pa ... Cellophane window 1 ... Image input unit 2 ... Area detection unit 3 ... Line detection unit 4 ... Character detection unit 5 ... Character recognition unit 6 ... Character recognition dictionary 7 ... Address recognition unit 8 ... Address dictionary

Claims (4)

In an address reader that reads an address of a medium to be read,
  Capture means for capturing an image from the medium;
  First detection means for detecting information on the four corners of the cellophane window candidate and edge information in the horizontal and vertical directions of the cellophane window candidate from the image captured by the capture means;
  Second detection means for detecting a cellophane window candidate region based on information on the four corners of the cellophane window candidate detected by the first detection means and information on horizontal and vertical edges of the cellophane window candidate;
  When the density inside and outside the cellophane window candidate area detected by the second detection means in the image captured by the capture means is different, the paper quality of the medium is different from the paper quality of the contents of the medium. Determining means for determining the cellophane window region as being,
  Reading means for reading the address from the image of the cellophane window area determined by the determining means in the image captured by the capturing means;
  An address reading device comprising: In an address reader that reads an address of a medium to be read,
  Capture means for capturing an image from the medium;
  First detection means for detecting information on the four corners of the cellophane window candidate and edge information in the horizontal and vertical directions of the cellophane window candidate from the image captured by the capture means;
  Second detection means for detecting a cellophane window candidate region based on information on the four corners of the cellophane window candidate detected by the first detection means and information on horizontal and vertical edges of the cellophane window candidate;
  Determination of determining the cellophane window area based on the number of changes from white to black when the inside and outside textures of the cellophane window candidate area detected by the second detection means in the image captured by the capture means are different. Means,
  Reading means for reading the address from the image of the cellophane window area determined by the determining means in the image captured by the capturing means;
  An address reading device comprising: The reading means determines the character string included in the cellophane window area determined by the determining means as the destination address area, and the image of the determined destination address area in the image captured by the capturing means. The address reader according to claim 1 or 2, wherein the address is read. Image reading means for outputting a digital image pattern based on density data by optically reading and scanning the surface of the medium on which the address is described;
  Based on the digital image pattern from this image reading means, an area detecting means for detecting a description area of the address and outputting a binarized pattern based on the digital image pattern of the address description area;
  Line detection means for detecting each line constituting the address in the address description area based on the binarization pattern from the area detection means, by means of slanting, etc .;
  Character detection means for cutting out a character pattern for each character by projection or the like based on the binarized data for each line detected by the line detection means;
  A character recognition means for recognizing a character candidate by matching a character pattern for each character supplied from the character detection means with a reference pattern for alphabets, numbers, and kanji in a character recognition dictionary;
  Address recognition means for recognizing the address by collating the character recognition result supplied from the character recognition means with the address dictionary,
  The region detecting means is
  With respect to the digital image pattern based on the density data for one screen from the image reading means, four corner detection edges, horizontal edge detection filters, and vertical edge detection filters are used, respectively. And a cellophane window area candidate detecting means for detecting a position candidate of a vertical edge and detecting a cellophane window area candidate by integrating these detected position candidates,
  For the cellophane window area candidate detected by the cellophane window area candidate detection means, Selection means for evaluating each cellophane window position candidate by each filter output, density difference, position, size, etc., and selecting only a candidate that is likely to be a cellophane window
  An address reader characterized by that.

Images