JPH0644401A - Postal matter processor - Google Patents

Abstract

PURPOSE:To provide the postal matter processor which can determine uniquely a recognition candidate without deteriorating the processing efficiency and the reliability. CONSTITUTION:A data base 21 for storing frequency information to which generation frequency of each dwelling number of each street name is weighted by a misreading rate is prepared in advance, and when plural numeral/character recognition candidates 24 are selected, the frequency information corresponding to each numeral/character recognition candidate is read out of the data base 21, and based on these frequency information, an evaluation value of each numeral/character recognition candidate is calculated, and based on these evaluation values, the numeral/character recognition candidate is determined uniquely.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a mail processing apparatus such as a mail address automatic reading / sorting machine.

[0002]

2. Description of the Related Art Generally, an automatic mail address reading / sorting machine recognizes mail addresses by optical means and automatically sorts them into sorting boxes according to the recognition result. To recognize an address by such an automatic mail address reading / sorting machine, it is necessary to recognize a town name and a house number.
Conventionally, recognition of these town names and house numbers has been performed by cutting out the parts in which they are described from the mail and calculating the recognition candidates for the cut out parts by, for example, the composite similarity method.

By the way, in the recognition method based on such a composite similarity method, a plurality of recognition candidates may be calculated especially for a house number which is a combination of up to four digits, and in this case, finally, in this case. Needs to uniquely determine recognition candidates.

However, conventionally, when a plurality of recognition candidates are calculated, a method of rejecting even at the expense of the classification rate and a method of appropriately selecting misreading are prepared.
There is a problem that processing efficiency and reliability are reduced.

[0005]

As described above, in the automatic mail address reading / sorting machine, when recognizing the town name and the house number, there are a plurality of recognition candidates for the house number, which is a combination of up to four digits. It may be calculated,
Conventionally, recognition candidates are uniquely determined by a method of rejecting even at the expense of the classification rate or a method of appropriately selecting misreading, and thus there is a problem that processing efficiency and reliability are reduced.

Therefore, an object of the present invention is to provide a mail processing apparatus capable of uniquely determining a recognition candidate without deteriorating processing efficiency and reliability.

[0007]

In order to solve the above problems, the present invention provides a numerical character recognition means for selecting a numerical character recognition candidate from a mail in a mail processing device for recognizing a house number of a mail. , A database that stores frequency information in which the frequency of occurrence of each house number of each town name is weighted by the misreading rate, and when a plurality of numeric character recognition candidates are selected by the numeric character recognition means, each numeric character recognition candidate is handled. An evaluation unit that calculates an evaluation value of each numerical character recognition candidate based on the frequency information and uniquely determines a numerical character recognition candidate based on these evaluation values, and the database every time the house number of the mail is recognized. And a means for updating the occurrence frequency of.

[0008]

In the present invention, a database for storing frequency information in which the frequency of occurrence of each house number of each town name is weighted by the misreading rate is prepared in advance, and when a plurality of numerical character recognition candidates are selected, each database is selected from the database. The frequency information corresponding to the numerical character recognition candidates is read, the evaluation value of each numerical character recognition candidate is calculated based on these frequency information, and the numerical character recognition candidate is uniquely determined based on these evaluation values. Therefore, compared with the conventional technology that uniquely determines recognition candidates by a method of rejecting even at the expense of the classification rate or a method of appropriately selecting misreading, it is possible to uniquely identify recognition candidates without lowering processing efficiency and reliability. Can be determined. In addition, by updating the frequency of occurrence of the database every time the house number of the mail is recognized, the evaluation performance improves as the mail is used.

[0009]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 2 is a front view of a mail address automatic reading / sorting machine according to an embodiment of the present invention.

In the figure, 1 is an operation panel for designating a reading mode, and 2 is a supply section for supplying mail 3.

In the supply unit 2, the postal items 3 taken out one by one from the supply table 4 by the take-out unit 5 are sent out to the reading unit 7 through the conveying path 6.

The reading unit 7 recognizes the address written on the mail piece 3, and the recognized mail piece 3 collects the mail piece 3.
Sent to.

In the accumulating section 8, the postal matter 3 passing through the conveying path 9
On the basis of the recognition result in the reading section 7, the boxes are classified according to the address.

FIG. 3 is a block diagram showing the configuration of the reading unit 7.

As shown in the figure, the entire surface image of the postal matter 3 imaged by the photoelectric conversion unit 11 is converted into an electric signal and transmitted to the image processing unit 12.

The image processing unit 12 detects the address area, further performs character detection cutout, and transmits the character detection cutout candidate to the character recognition unit 13.

The character recognition unit 13 performs recognition according to the transmitted character detection cutout candidates, evaluates and integrates them, finally determines a recognition result uniquely, and transmits the result to the MCPU 15 via the LCPU 14.

Then, the MCPU 15 controls the drive unit 16 of each unit of the apparatus to sort the mail piece 3 into a specific sorting box 10.

FIG. 4 is a block diagram showing the structure of the character recognition unit 13.

In the address recognition unit 17 shown in FIG. 3, the address character recognition candidates are calculated by the composite similarity method according to the character detection cutout candidates transmitted from the image processing unit 12, and the address information database 18 is further referred to. , The address character recognition candidates are narrowed down to certain reliable recognition candidates. The recognition result by the address recognition unit 17 is transmitted to the number recognition unit 19.

The number recognition unit 19 calculates a number character recognition candidate from the number character portion subsequent to the recognition candidate of the address recognition unit 17 according to the character detection cutout candidate by using the combined similarity method and the structural feature matching method in combination. Then, referring to the house number information database 20, the number character recognition candidates are narrowed down to the number character recognition candidates of the existing numbers.

Then, the numeral recognition section 19 refers to the house number frequency database 21 to uniquely determine the final answer.

FIG. 1 is a block diagram showing the configuration of the numeral recognition unit 19.

The recognition object 22 shown in the figure is recognized by the numeric character recognition unit 23. Here, since there are a plurality of detection cutout candidates, a plurality of recognition results are calculated.

Further, the numeral character recognition section 23 refers to the house number information database 20 and calculates only the house number that actually exists as the recognition result 24.

Since the numeric character recognition unit 23 cannot uniquely determine the recognition result, the recognition result is transmitted to the recognition result evaluation unit 25.

The recognition result evaluation section 25 refers to the house number frequency database 21 to uniquely determine the final answer. The house number frequency database 21 stores frequency information in which the frequency of occurrence of each house number of each town name is weighted by the misreading rate. The house number frequency database 21 is automatically updated with the recognition result of the recognition result evaluation unit 25.

The recognition result evaluation unit 25 uses the evaluation function f (x).
With reference to the house number frequency database 21, the evaluation value is acquired.

Then, for example, when the evaluation calculation result for the recognition result 24 is f (c) <f (b) <f (a), and f (a) ≧ TH (TH is a threshold value), The a in the recognition result 24 becomes the recognition result 26 that is uniquely determined.

FIG. 5 is a diagram showing the structure of the house number frequency database 21.

As shown in the figure, town name information 27 is entered in the house number frequency database 21, and a predetermined town name is selected based on the recognition result by the address recognition unit 17. The house number range information 28 is linked to each town name in the town name information 27.
A predetermined range of house numbers is selected based on the recognition result of 9.

The house number frequency information 29 is linked to each house number range in the house number range information 28, and the occurrence frequency and the evaluation value of the evaluation function f (x) are calculated based on the house number frequency information 29. It When the corresponding house number frequency information 29 is referenced at the same time, one vote is given for the number of occurrences. Furthermore, all town names are automatically updated even if they are not referenced. It should be noted that the error rate is information given in advance, and when the expected evaluation value cannot be obtained,
You should reconsider this value.

As described above, in this embodiment, the evaluation value of each numerical character recognition candidate is calculated based on the frequency information stored in the house number frequency database 21, and the numerical character recognition candidate is uniquely determined based on these evaluation values. Since it is determined, the recognition candidate can be uniquely determined without deteriorating the processing efficiency and reliability.

[0035]

As described above, according to the present invention, the numerical character recognition candidate is uniquely determined by referring to the frequency information in which the occurrence frequency is weighted by the misreading rate, so that the processing efficiency and reliability are reduced. It is possible to uniquely determine the recognition candidate without causing any trouble.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a numeral recognition unit according to an embodiment of the present invention.

FIG. 2 is a front view of the mail address automatic reading / sorting machine according to the embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of a reading unit according to an embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a character recognition unit according to an embodiment of the present invention.

FIG. 5 is a diagram showing a configuration of a house number frequency database according to an embodiment of the present invention.

[Explanation of symbols]

20 ... House number information database, 21 ... House number frequency database, 22 ... Recognition object, 23 ... Numeric character recognition part, 24 ... Recognition result, 25 ... Recognition result evaluation part, 26 ... Evaluation result.

Claims (2)

[Claims] 1. A mail processing device for recognizing a house number of a mail, wherein the character recognition means for selecting a candidate for character recognition from the mail and the frequency of occurrence of each house number of each town name are weighted by a misreading rate. When a plurality of numerical character recognition candidates are selected by the numerical character recognition means and a database that stores the frequency information, the evaluation value of each numerical character recognition candidate is calculated based on the frequency information corresponding to each numerical character recognition candidate. Then
A mail processing apparatus comprising: an evaluation unit that uniquely determines a numerical character recognition candidate based on these evaluation values. 2. The mail processing apparatus according to claim 1, further comprising means for updating the occurrence frequency of the database each time the house number of the mail is recognized.