JP3673034B2 - Mail address area detection device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a mail processing apparatus, and more particularly to a mail address area detection apparatus used when reading mail addresses and sorting them according to their delivery destinations.
[0002]
[Prior art]
In the mail processing field, mails sent in large quantities every day must be processed within a limited time. Therefore, the mechanization of postal services has been promoted, and the burden on post office staff has been reduced. One example is an optical sorting device that uses an optical character reader to read a postal code on a postal item, and sorts the postal item into a plurality of delivery post offices based on the reading result, or even on a postal item. It is a delivery sorting machine that reads an address and sorts mail items into a plurality of delivery areas based on the read result.
[0003]
In the delivery sorting machine, when reading the address written on the mail piece, conventionally, it is determined one area where the address is estimated to be written by image processing, and line detection, character detection, The address is automatically recognized by pipeline processing of character identification and town name identification.
[0004]
However, with regard to the current mail address, there are almost no restrictions on the format of the address description, description method, etc., so there are various forms of mail, and the address of advertisements etc. on the mail. There are other character information, pictures, and photographs, and it is very difficult to correctly extract the address information from a mixture of these image information. Therefore, considering that the amount of mails will continue to increase in the future, it is required to develop an address reading method with higher performance than conventional methods.
[0005]
[Problems to be solved by the invention]
Since the conventional technology is configured to read the information contained in the address area after determining the address area, if the wrong area is determined as the address area, the mail is rejected from the delivery sorting machine. Therefore, there is no choice but to re-input, and it is impossible to correctly recognize the address in the subsequent processing in the first delivery sorter. Moreover, since all the characters written in one determined area are recognized and then it is determined whether or not this area is a destination area, the processing time per area becomes longer.
[0006]
In general, in order to increase the success rate of address area detection, it is only necessary to continue the detection operation of other areas for mail that has been rejected once. However, since the processing time per mail is limited, It has been difficult to detect a plurality of regions by the method because it takes a lot of time.
[0007]
SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a mail address area detection apparatus that can easily and accurately detect an address area from an image on the surface of a mail object and perform mail process at high speed.
[0008]
[Means for Solving the Problems]
The mail address detecting device of the present invention is More photoelectric conversion and image signal Image input means to obtain; An image memory for storing the image signal obtained by the image input means, a black pixel connection area obtained by performing binary image determination on the image signal in the image memory, and circumscribed rectangle information circumscribing the black pixel connection area are extracted. From the black pixel connection region extraction unit, the line candidate detection unit that extracts the rectangle information of the line candidate from the circumscribed rectangle information extracted by the black pixel connection region extraction unit, and the rectangular information extracted by the line candidate detection unit An address area candidate detection unit for detecting an address area candidate; and Addressing area determining means for determining the addressing area from the character information included in the addressing area candidate.
[0009]
With the above configuration, character lines are first detected, integrated to detect a plurality of address area candidates, character lines that are estimated to be address writing portions are detected from each area candidate, and the character lines are detected. By performing character detection, character identification, and town name identification on the address, it is possible to determine at high speed whether or not the area including the character line is an address description area. . As a result of character recognition, if the character line does not include address information, it is determined that address information is not written in an area including the line, and the same process is performed for the next area candidate. Here, in the present invention, since it is only necessary to perform character recognition for one to several lines for each area, the processing time per area can be shortened, so that it is possible to check a plurality of area candidates.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of the optical character reading apparatus according to the first embodiment of the present invention, and FIG. 2 is a block diagram showing the internal configuration of the address area detection unit of FIG.
[0011]
In FIG. 1, an image of a postal matter M is converted into an electric signal by the photoelectric conversion circuit 2 of the optical character reader 1, and converted into a binary signal indicating white pixels and black pixels by the binarization processing unit 3. The binary signal obtained from the binarization processing unit 3 is temporarily stored in the image memory of the black pixel connection region extraction unit 4, and the black pixel connection region is extracted on the image memory. In general, the mail address is written in black, and the mail address is written on one or more lines. Therefore, when a black pixel connection area is detected from an image of a postal matter, there is a high possibility that the detected black pixel connection area contains a character or a part of a character constituting the addressed character string. From the detected black pixel connection area, rectangular information circumscribing the area is obtained.
[0012]
The rectangular information circumscribing the black pixel connection area extracted by the black pixel connection area extraction unit 4 is sent to the address area detection unit 5 for detecting the address area. Reading of the address characters included in the area detected by the address area detection unit 5 is performed by detecting each character from the address character line by the next character detection / cut-out circuit 6, and for each of the cut-out characters. Do.
[0013]
Here, the internal configuration of the address area detection unit 5 will be described with reference to FIG. In FIG. 2, the rectangle information circumscribing the black pixel connected area extracted by the black pixel connected area extracting unit 4 is sent to the row candidate detecting unit 5a of the addressed area detecting unit 5 for checking the addressed area. The The address area detection unit 5 is provided with a reference address area database 5R in association with a line candidate detection unit 5a, an address area candidate detection unit 5b, a line, an area evaluation unit 5c, and an address writing head candidate detection unit described later. Candidates are detected at each stage by referring to reference information prepared in advance in each of 5d or by the degree of similarity based on comparison with reference information.
[0014]
First, in the line candidate detection unit 5a, information indicating the shape of the rectangular area, the position on the mail, etc. included in each circumscribed rectangle information extracted by the black pixel connection area extraction unit 4 is stored in the reference address area database 5R. Are compared with the corresponding reference information, and a plurality of circumscribed rectangular areas are integrated based on the comparison result to cut out a line candidate area, and the rectangular information of the line candidate is extracted.
[0015]
The detected line candidate information is then processed by the address area candidate detection unit 5b and compared with the corresponding reference information stored in the reference address area database 5R. Based on the comparison result, one or a plurality of locations are detected. Address area candidates are detected.
[0016]
For the detected address candidate area, how many lines are included in the detected address candidate area, or the number of lines is larger or smaller than the predetermined number n, or the average width of the detected lines is predetermined. Whether the distribution is larger or smaller than the size, the distribution of the detected line width, the distribution of the position of the lines included in the detected address candidate area, etc. Evaluation is performed by the area evaluation unit 5c. Details of the evaluation method by the line and region evaluation unit 5c will be described later.
[0017]
When the row / area evaluation unit 5c evaluates that the address area is an address area, each address area candidate detected by the address area candidate detection section 5b is similarly stored in the reference address area database 5R. It is compared with the corresponding reference information, and the address description format is determined from the distribution of the number, size, and position of the lines included in the reference information, and the next address head line candidate detection unit is determined according to the determined format. In 5d, a candidate for the first address row is detected.
[0018]
When there are a plurality of detected address candidate areas, the head candidate among the plurality of address candidate areas is detected by the address head line candidate detecting unit 5d and stored in a predetermined storage area.
[0019]
One of the address head line candidate candidates detected by the address area detection unit 5 in this way, for example, one line most likely to be the head of the head is selected, and this line is individually detected by the next character detection / cutout circuit 6. Are sequentially sent to the recognition circuit 7 and character recognition is performed using the character dictionary 8. Character recognition is performed on all the characters in the first address portion of the address, and the recognized characters are sent to the next knowledge processing circuit 9 to identify the city name using the knowledge dictionary 10. If the identified city name matches the actual address stored in the knowledge dictionary 10 in advance, all the lines in the address area including the processed address first line candidate are similarly identified and read. The address information thus sent is sent to the read result processing circuit 11, and a predetermined delivery classification process is performed.
[0020]
The mail portion M read by the photoelectric conversion circuit 2 in FIG. 1 is written in two lines L1 and L2 in horizontal writing at the center in the vertical direction as shown in FIG. The operation of the optical character reader 1 shown in FIG. 1 and FIG. It is assumed that the image of the postal matter M to be processed further includes images of advertisement character lines L3 and L4 and a stamp L5. In the following explanation, in a normal letter, the letter or postal code frame is placed on the top and the longitudinal side is vertical. We will refer to how to place a letter whose direction is vertical. A vertical line refers to a long line in the longitudinal direction of the letter, and a horizontal line refers to a long line in the lateral direction of the letter.
[0021]
The image data of the postal matter M read by the photoelectric conversion circuit 2 is subjected to differentiation / binarization processing by the binarization processing unit 3, and then the black pixel connection region extraction unit 4 in FIG. The rectangle information circumscribing the connected black pixel area corresponding to the rows L1, L2, L3, and L4 as shown in FIG. This rectangular information corresponds to each character, and this information is shown in FIG. 3C by integrating the circumscribed rectangular information adjacent in the horizontal direction in the line candidate detecting unit 5a in the address area detecting unit 5. Such lines L1, L2, L3, L4 are detected.
[0022]
The information of these detected rows L1, L2, L3, and L4 is obtained by integrating the adjacent rows in the address area candidate detection unit 5b and corresponding to the rows L1 and L2 as shown in FIG. The first address area candidate A1 and the second address area candidate A2 are detected. The respective rows L1 to L4 in each of the areas A1 and A2 are subjected to the evaluation criteria prepared in advance and stored in the reference address area database 5R in FIG. , A2 having a high similarity or high evaluation value is selected as an address writing line candidate, that is, an address head line part. The row selected as the leading row candidate in the areas A1 and A2 is indicated by hatching as shown in FIG.
[0023]
A plurality of address writing line candidates are selected and subjected to the feedback processing shown in FIG. 3 (g) in descending order of evaluation value. In the feedback process, first, the line L1 is selected from the address writing line candidates in FIG. 3 (h), each character of the line L1 is detected in FIG. 3 (i), and each character in FIG. 3 (j) is detected. 3 is identified, and the identified character string is subjected to municipality identification, for example, town name identification processing in FIG. Next, in FIG. 3L, it is checked whether the identification result can be read as address information. If the identified character string does not make sense as a result of referring to the knowledge dictionary 10, the first row in another area A2, for example L3, is selected, and the same processing is repeated. If it makes sense in FIG. 3 (l), the line before or after the line L1, here the line L2 after that, is subjected to recognition processing, and the address information is completed.
[0024]
Next, a description will be given of a process for determining in what format the address is described in the row and area evaluation unit 5c in FIG. In this process, first, some main forms in which addresses are described in mail pieces are stored as information, and a process is performed to determine in which form a given letter is likely to be written. FIG. 4 is an example of a classification system for formats in which addresses are described. In FIG. 4, paying attention to the direction and the position of the address line, vertical horizontal writing (a), vertical horizontal writing (b), horizontal horizontal writing forward direction (c) and reverse direction (d) In this example, the address format is classified into four ways. Other than this example, the address line direction / position may be additionally registered in the address description format. Further, the address description format may be further classified and defined in consideration of the address line size in addition to the address direction and description position.
[0025]
As an example of a method for determining the address description format, if a horizontal line is found from the top of the image when the line is cut out from the letter image, it is determined that the address is written in the vertical and horizontal direction. Can be considered. Further, in this case, it is conceivable to perform a process of increasing the priority applied to the address line writing candidate detection process as the horizontal line written in the upper part of the letter.
[0026]
Hereinafter, specific examples of row and region evaluation in the row and region evaluation unit 5c in FIG. 2 will be described in detail with reference to FIGS.
In the example of FIG. 5, as shown in FIG. 5A, it is a case of a mail piece in which the address is described in two lines of vertical letters on the letter placed vertically, and FIG. 5B is an image of this mail piece. The flowchart which shows the process which determines an address description area | region from A, and detects an address line when an address is displayed by handwriting is shown. After cutting out the lines included in the area detected by the address area candidate detection unit 5b in FIG. 2, the number of lines detected in the first step ST1 in FIG. 5B is n lines or less, for example, 4 lines or less. It is determined whether or not it is present, and if it is not less than n rows, it is determined that the area is not an address area, and another determination process in step ST2 is performed.
[0027]
Since the address is n lines or less, the address is handwritten, the handwriting flag is set in step ST3, and the first address line detection process prepared for the handwritten address is performed in step ST4. This head address line detection process is performed in the address head head candidate line detector 5d shown in FIG.
[0028]
The determination condition as to whether the address line is 4 lines or less may be used in combination with another condition. As an example, FIG. 6 shows a case where the address is combined with the condition that the address is in the longitudinal direction of the letter. After cutting out lines from the letter image shown in FIG. 6A, it is determined whether or not the number of lines in the longitudinal direction of the letter among the lines detected in step ST11 is n or less. In other words, the process proceeds to another determination process (ST12). If it is n lines or less, it is determined that the address is handwritten in the longitudinal direction of the letter (ST13), and address line detection processing prepared for the case where the address line is a handwritten vertical line is performed (ST14). For example, if the number of vertical lines detected from the letter is 4 or less, it is determined that the address is handwritten. The row direction condition may be the letter short direction instead of the letter longitudinal direction.
[0029]
FIG. 7 shows another example of a method for determining the address description area. After cutting out lines from the letter image shown in FIG. 7 (a), the average width of the lines detected from the letter is calculated in step ST21 of FIG. 7 (b), and whether the line average width is equal to or greater than W millimeters. If it is not greater than W millimeters, another determination process (ST23) is performed.
[0030]
If it is greater than or equal to W millimeters, it is determined that the address is handwritten (ST24), and the first address line detection process prepared for the handwritten address is performed (ST25). For example, if the average width of a line detected from a letter is 20 mm or more, it is determined that the address is handwritten. The row average width threshold may be given in units of pixels instead of in millimeters. In the above example, all lines detected from the letter image are used to obtain the average width. However, the average width may be calculated using only lines written in a specific area of the letter image.
[0031]
The above determination conditions may be used in combination with other conditions. As an example, FIG. 8 shows a case where the address line is combined with the condition that the address line is in the longitudinal direction of the letter. After cutting out lines from the letter image shown in FIG. 8A, the average width of the vertical lines among the lines detected from the letter in step ST31 of FIG. 8B is calculated, and the average line width is W millimeters. It is determined whether or not this is the case (ST32). If it is greater than or equal to W millimeters, it is determined that the address is handwritten in the longitudinal direction of the letter (ST34), and the first address line detection process prepared for the case where the address line is a handwritten vertical line is performed (ST35). For example, if the average width of the vertical lines detected from the letter is 20 mm or more, it is determined that the address is handwritten. The row average width threshold may be given in units of pixels instead of in millimeters. The row direction condition may be the letter short direction instead of the letter longitudinal direction. Further, in the above example, all lines detected from the letter image are used to obtain the average width, but the average width may be calculated using only lines written in a specific area of the letter image.
[0032]
In order to estimate the address description format using the average line width detected from the image, the average line width needs to be a value close to the address line width. FIG. 9 shows an example of the head addressed line detection method when there is a variation in the width size of the line detected from the image. After a line is cut out from the letter image shown in FIG. 9A, a line with a high possibility of a destination line is selected as shown in FIG. 9B surrounded by a rectangle, and the address of the address is selected according to the average width of the selected line. Determine the format.
[0033]
First, the distribution of the width sizes of the lines detected from the letter is checked (ST41), and the average width of the lines is calculated (ST42). Next, among the detected lines, a line having a width that is wider than 1 / m of the average width and not more than n times is selected (ST43), and the average width of the selected lines is calculated to determine the address format. (ST44). It is determined whether or not the reference width is greater than or equal to W millimeter (ST45). If it is not greater than or equal to W millimeter, the process proceeds to another determination process (ST46).
[0034]
If it is greater than or equal to W millimeters, it is determined that the address is handwritten (ST47), and the first address line detection process prepared for the handwritten address is performed (ST48). For example, if m = 2, n = 1.5, and the reference width is 20 mm or more, it is determined that the address is handwritten. The row average width threshold may be given in units of pixels instead of in millimeters. In the above example, all the lines detected from the letter image are used to obtain the reference width. However, the reference width may be calculated using only the lines written in a specific area of the letter image.
[0035]
The above determination conditions may be used in combination with other conditions. As an example, FIG. 10 shows a case where the address is combined with the condition that the address is in the longitudinal direction of the letter. After cutting out lines from the letter image shown in FIG. 10 (a), select a vertical line with a high possibility of the address line as shown in FIG. 10 (b), and select it in each step of FIG. 10 (c). The address description format is determined by the average width of the selected lines.
[0036]
First, the distribution of width sizes of the vertical lines detected from the letter is checked (ST51), and the average width of the vertical lines is calculated (ST52). Next, among the detected vertical lines, a line having a width that is wider than 1 / m of the average width and not more than n times is selected (ST53), the average width of the selected vertical lines is calculated, and the address format A reference width for determination is used (ST54). It is determined whether or not the reference width is equal to or greater than W millimeter (ST55), and if it is not equal to or greater than W millimeter, the process proceeds to another determination process (ST56). If it is greater than or equal to W millimeters, it is determined that the address is handwritten in the longitudinal direction of the letter (ST57), and the first address line detection process prepared for the case where the address line is a handwritten vertical line is performed (ST58). For example, if m = 2, n = 1.5, and the reference width is 20 mm or more, it is determined that the address is handwritten. The threshold value of the row average width may be given in units of the number of pixels instead of in millimeters, or the row direction condition may be in the short side direction of the letter instead of the long side of the letter. In the above example, all the lines detected from the letter image are used to obtain the reference width. However, the reference width may be calculated using only the lines written in a specific area of the letter image.
[0037]
FIG. 11 shows another example of a method for determining the address description format. In step ST61 of FIG. 11A, the distribution of the written positions of the lines detected from the letter is checked, and among the detected lines, the entire line area or the line in the right half area when the letter is placed vertically in step ST62. Investigate how many lines are covered by part of the area. If not more than n% of the total number of rows, the process proceeds to another determination process (step ST71) in FIG.
[0038]
If it is n% or more, it is determined that the address is written vertically or horizontally on the right half of the letter as shown in FIG. 11B or 11D (ST63). For example, in the letter shown in FIG. 11 (b) or FIG. 11 (d), each letter has 70% or more of the total number of lines written in the right half of the letter. Or it is judged that it is written horizontally.
[0039]
Conversely, when 70% or more of the total number of lines is written in the left half of the letter as shown in FIG. 12B, among the lines detected in step ST71 of FIG. Determine how many lines are covered by the entire line area or a part of the line area in the left half area when the letter is placed vertically, and if it is not more than n% (for example, n = 70) of the total number of lines The process proceeds to another determination process (ST72).
[0040]
If it is 70% or more, it is determined that the address is written horizontally in the left half of the letter from the bottom to the top (ST73). For example, when the letter of FIG. 12 (b) is given and 70% or more of the total number of lines is written in the left half of the letter, the lower half of the letter is directed upward from the bottom of the letter as shown in FIG. 12 (c). Judge that it is written horizontally in horizontal writing.
[0041]
Here, a specific example of the process of detecting the address description area from the mail image will be further described with reference to FIG. The line information shown in FIG. 13B is detected from the letter image shown in FIG. Next, by defining the default area and address line direction for detecting the address line for each address format, select the line that has the same direction as the address line and the line area in the line default area, and select the address line candidate. And For example, when it is assumed that the address is written in vertical and horizontal writing, when the horizontal line in the hatched area in FIG. 13C is selected and it is assumed that the address is written in horizontal writing or vertical writing Selects the vertical line that covers the shaded area in FIG. 13 (d) or (e). In the case of the letter shown in FIG. 13A, as shown in FIGS. 13F and 13G, address area candidates and address line candidates for vertical horizontal writing and horizontal horizontal writing in the forward direction are detected. As shown in FIG. 13 (e), when the hatched portion is biased to the right end in the reverse direction, the horizontally-addressed addressee candidates are not detected in FIG. 13 (h).
[0042]
For each description format detected as described above, a process for finding an address line from the selected lines is performed. Hereinafter, processing for detecting and reading an address line from a mail image will be described. FIG. 14 shows an example of a process in which a line of address writing is first detected from the postal image, and subsequent lines of the address are sequentially subjected to character recognition.
[0043]
In step ST181 of FIG. 14, a mail image is input, line information is detected in step ST182 from the input mail image, and the description position and line size distribution of each line detected in step ST183 are examined and analyzed.
[0044]
The description format of the most likely address as a result of the analysis is determined (ST184), address line detection processing corresponding to the address description format is performed, and one line or a plurality of address writing, that is, the first line candidate is selected (ST185). ). The selected address writing line candidates are subjected to character detection and character recognition processing in descending order of priority, and it is determined whether they can be read as part of a town name, for example (ST186).
[0045]
If there is a line that can be read as a town name in the address writing line candidates, a subsequent line is found near the position on the postal image of the line, and character detection and character recognition processing are sequentially performed as the address line (ST187). . If there is no address writing line candidate that can be read as a town name, the second most likely address description format is selected (ST188), address line detection processing corresponding to the address description format is performed, and one or more address addresses are selected. A writing line candidate is selected (ST189). Hereinafter, the address writing format having high possibility is sequentially selected until the address writing line candidate that can be read as a part of the town name is found (ST190) (ST191), and the address line detection processing corresponding to the address writing format is performed. Alternatively, a plurality of address writing line candidates are selected (ST192), and the selected address writing line candidates are subjected to character detection and character recognition processing up to n-th in descending order of priority to determine whether they can be read as town names (ST193). ) Repeat the process.
[0046]
If the address writing line is not found after processing up to the nth address description process, the process is aborted and rejected. Here, if the possibility is very low, an error often occurs, so the value of n may be a fixed value such as 3, for example. Further, the value of n may be changed according to the address description format selected first. As an example, if it is first determined that the vertical writing is horizontal writing, the address can often be read with the first candidate, so that n = 1, and in other cases, n = 4 can be considered.
[0047]
Next, FIG. 15 shows an example of processing in the case where the description format of the address is determined by examining the position and size distribution of the lines written in the mail image, and the description format cannot be narrowed down to one. Since most processes are the same as those in FIG. 14, the corresponding processes are given the same numbers. In FIG. 15, line information is detected from the mail image (181) (ST182), and the description position and line size distribution of each detected line are examined and analyzed (ST183), and several possible address description formats are selected. (ST184A).
[0048]
If there are n possible address description formats, first select one of the n most likely address description formats (ST184B), perform address line detection processing corresponding to that format, and perform addressing. A line candidate is detected (ST185). The detected address line candidates are sequentially subjected to character detection processing and character recognition processing to determine whether the line can be read as part of the address (ST186). If it can be read as the address, the subsequent lines are sequentially subjected to the address reading process (ST187).
[0049]
If no readable line is found, an address description format similar to the first address description format is preferentially selected (ST188A), and address line detection processing corresponding to the address description format is performed to detect address line candidates. (ST189) Then, a line that can be read as a part of the address is searched from the detected candidates (ST190). If the address line is not found, the same processing is repeated with another address description format (ST190), (ST191), ( ST192), (ST193). If the address line is not found even in the nth address description format, it is rejected.
[0050]
Next, FIG. 16 shows another example of processing in the case where the description format of the address is determined by examining the position and size distribution of the lines written in the mail image, and the description format cannot be narrowed down to one. This is because, for example, as shown in FIGS. 16 (a) and 16 (b), in the two letters whose addresses are written in the right half of the letter, the positions and sizes of the lines are very similar. In this case, it is difficult to determine whether the address format is vertical writing or horizontal writing until character recognition processing is performed. In such a case, when a letter written in one addressing format reaches more post offices than a letter written in the other addressing format, the address line detection process corresponding to the most frequently addressed format is performed. It is effective to perform the address line detection processing corresponding to the address description format that reaches the second most frequently when it is not found. In the following explanation, the address line detection process is performed first when the address is written vertically in the right half of the letter in the vertical writing direction.
[0051]
In ST202, from the letter determined that the address is written in the right half of the letter in the first step ST201 in FIG. Next, as shown in (a), address line detection processing (ST203) is performed when the address is written vertically in the right half of the letter in the vertical writing direction, and address line candidates are selected from the lines to be searched for the address. select.
[0052]
Address line candidates are sequentially subjected to character detection and character recognition processing to check whether there is a line that can be read as a part of the address (ST204). When the address line is found, the lines near the line are sequentially subjected to address reading processing (ST205). If the address line is not found, as shown in FIG. 16B, address line detection processing (ST206) is performed when the address is written horizontally and vertically in the right half of the letter. Select a candidate address line from the target lines. The address line candidates are sequentially subjected to character detection and character recognition processing to check whether there is a line that can be read as a part of the address (ST207). When the address line is found, the lines near the line are sequentially subjected to address reading processing (ST208).
[0053]
If the address line is still not found, the address line detection process is performed when another address description format is selected (ST209).
In this example, the case where the address is written on the right half of the letter is described, but it may be used in combination with other constraints. For example, the part of (ST201) may be “a letter determined that the address is handwritten and written on the right half of the letter”.
[0054]
Next, still another example of the processing for detecting and reading the address line from the mail image will be described. FIG. 17 shows an example of processing in which a line including a hyphen “-” is detected from a postal image and a line written in the vicinity of the line is used as a candidate address line. For example, the line information included in the address based on the hyphen “-” 211 included in the middle of the postal code description frame column 210 on the top line of the mail image shown in FIG. The hyphen 213 in 212 is detected.
[0055]
In the first step ST211 of FIG. 17, the mail image as shown in FIG. 4A is taken in, and the horizontal character line 210 is detected at the top of the image in step ST212.
[0056]
Next, in step ST213, the row number counter i is set to 0, and if i is smaller than the total number of rows detected in step ST214, that is, if there is one or more horizontal rows, a hyphen in the row 210 is detected in step ST215, for example. It is checked whether or not “−” 211 exists. The presence of the hyphen 211 is determined by using the projection information of the binarized image in the row 210 or by examining the shape of the black pixel circumscribed rectangle in the row 210. If there is no hyphen in line 210, the value of i is incremented by 1 in step ST216, and the process returns to ST214.
[0057]
If the hyphen 211 is found, the flag of the line 210 is set in step ST217, and the process returns to ST214. The process ends when the value of i becomes equal to the number of rows (in this case, i = 3). Similarly, a hyphen 213 is detected in the third row 212. After all the lines have been evaluated, a line set composed of one line (214) or a plurality of lines sandwiched between two lines (210, 212) in which a flag is set on the letter is detected in ST218.
[0058]
Then, 0 is set to the row set count count j (ST219), and if j is smaller than the total number of row sets detected (ST220), the two rows (210, 212) with the flag set are set to the zip code line and A row including block information and a row (214) belonging to row set j sandwiched between two rows is sequentially subjected to character detection and character recognition processing (ST221) to check whether it can be read as an address (ST222). If it is possible to determine which of the two lines with the flag set is the postal code line, the address reading process is performed in order from the line closer to that line in the line set j. If it is not possible to determine which of the two lines is a zip code line, the process when one of the two lines is a zip code line is performed for both lines. When the row set j can be read as the addressee as a result of the address reading process, the reading result is output. If the address cannot be read, the value of j is incremented by 1 (ST223), and the process returns to (ST220).
If the value of j is equal to the total number of rows and no address is found, the process proceeds to another address detection process or a reject process.
[0059]
【The invention's effect】
As described above in detail, according to the present invention, addressable area candidates can be quickly detected, and confirmation of whether or not the addressable area can be performed quickly by performing character recognition on a small portion of each area candidate. Judgment processing of whether or not the address area per area can be executed at high speed. Therefore, a plurality of address area candidates can be taken from the mail image, and it can be confirmed whether or not each address area is addressed sequentially, and address reading performance is improved. It becomes possible to improve. Further, in each address area candidate, the reading order is set for the lines in the area, so that even if the line selected as the address writing line is not, the next candidate can be selected immediately. Due to the above effects, it is easy to select a line to be recognized when reading address information after the address writing line. Since the format of the address is estimated by evaluating the size, distribution, and description position of the line on the letter, it is easy to detect the address line. Furthermore, since an address line search area is prepared for each address description format and address line detection is performed from the area, address line detection processing time per format is short. Since a plurality of address description formats are applied to the letter, even if the first estimated address description format is not correct, the address information can be read if there is a correct answer in the second and subsequent address description formats.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a circuit configuration of a mail address area detection apparatus according to an embodiment of the present invention.
FIG. 2 is a block diagram showing in detail the interior of a part of the circuit of FIG. 1;
FIG. 3 is a flowchart showing a flow of mail address area detection processing according to an embodiment of the present invention;
FIG. 4 is a diagram showing examples of various address description formats.
FIG. 5 is a diagram showing an example of an address description format determination process.
FIG. 6 is a diagram showing another example of address description format determination processing;
FIG. 7 is a diagram showing another example of address description format determination processing;
FIG. 8 is a diagram showing still another example of the address description format determination process.
FIG. 9 is a diagram showing still another example of address description format determination processing;
FIG. 10 is a diagram showing still another example of the address description format determination process.
FIG. 11 is a diagram showing still another example of the address description format determination process.
FIG. 12 is a diagram showing still another example of address description format determination processing;
FIG. 13 is a diagram showing various examples of address description area detection.
FIG. 14 is a flowchart showing a flow of mail address line detection processing according to an embodiment of the present invention;
FIG. 15 is a flowchart showing a flow of mail address line detection processing according to another embodiment of the present invention;
FIG. 16 is a flowchart showing the flow of mail address line detection processing according to still another embodiment of the present invention;
FIG. 17 is a flowchart showing a flow of mail address line detection processing according to still another embodiment of the present invention;
[Explanation of symbols]
1 ... Optical character reader
2 ... Photoelectric conversion circuit
3 ... Binarization processing unit
4 ... Black pixel connection area extraction unit
5 ... Address area detector
5a ... line candidate detection unit
5b ... Address area candidate detection unit
5c: Row and region evaluation section
5d ... Address head part candidate detection part
5R ... Standard address area database
6 ... Character detection cutout unit
7 ... Recognition circuit
8 ... Character dictionary
9 ... Knowledge processing circuit
10 ... Knowledge Dictionary
11 ... Read result processing circuit
M ... Mail

Claims (7)

Image input means for obtaining an image signal by photoelectric conversion from the mail surface;
An image memory for storing the image signal obtained by the image input means;
A black pixel connection region obtained by performing binary image determination on the image signal of the image memory, and a black pixel connection region extraction unit that extracts circumscribed rectangle information circumscribing the black pixel connection region;
A line candidate detection unit that extracts rectangular information of line candidates from the circumscribed rectangle information extracted by the black pixel connection region extraction unit;
An address area candidate detection section for detecting address area candidates from the rectangular information extracted by the line candidate detection section;
Address area determination means for determining the address area from the character information included in the address area candidate;
A mail address area detection apparatus comprising: The line candidate detection unit
Means for typifying and storing various feature patterns of line images included in the address area on the mail surface;
Means for detecting a similarity between the feature of the detected row image and various feature patterns of the stored row image;
Means for determining address area candidates in descending order of the detected similarity;
The mail address area detection apparatus according to claim 1, further comprising:   The address area determination means includes area information detection means for detecting a description position of a line image representing a character line including the detected plurality of characters and a direction of the address character with respect to a long side or a short side of a postal matter. The mail address area detection apparatus according to claim 1. The area information detecting means includes a database in which statistical data on the number, size, distribution state, and position of the character lines is stored in advance for a plurality of pieces of mail, and the number, size, distribution state, and position of the detected character lines. 4. The mail according to claim 3 , further comprising means for outputting at least two candidates for the address description position and the address character direction from the correspondence between the data indicating at least one of the data and the database. Device address area detection device. Means for obtaining binary image information on the mail surface;
Means for detecting character line information in the postal matter from the binary image information;
Means for detecting the address description position and direction according to the character line information in the mail surface image;
Means for determining that the address is handwritten when the number of lines detected from the mail surface image is less than a threshold value prepared in advance;
Means for executing recognition processing of handwritten address characters according to the result of the determination means;
Means for detecting that the detected address is concentrated on the right or left side of the vertical mail;
If the address is written on the right side of the vertically placed mail, the address is recognized as vertical writing, and if the lines are concentrated on the left side, it is recognized and processed as horizontal writing.
A mail address area detection apparatus comprising: 6. The mail according to claim 5 , wherein the detecting means includes means for detecting that a predetermined percentage of the detected lines are concentrated on the right side or the left side of the vertically placed mail piece. Device address area detection device. Means for obtaining binary image information on the mail surface;
Means for detecting character line information in the postal matter from the binary image information;
Means for detecting the address description position and direction according to the distribution state of the character line information in the mail surface image;
Means for checking whether a hyphen is included in each detected line;
When there are two lines including hyphens by this means, means for determining each of them as a postal code line and a block line,
Address area detecting apparatus of mail, characterized in that it comprises, means for determining the address line of one or more lines sandwiched by two lines including the hyphen.

Images