JPH08243505A - Address reading device and method - Google Patents

Abstract

(57) [Summary] [Purpose] Fast and accurate recognition of the address display number in the address information of the mail, and easy maintenance such as registration of a new address display number notation pattern. The purpose is to do. [Structure] An address display number is created by registering a word expressed using a wildcard that represents an arbitrary number in a dictionary, and performing word matching on the candidate character group obtained by converting the result of character recognition into a wildcard and the dictionary. Recognize. In addition, when searching for words to be matched from the dictionary, the candidate character groups converted into wildcards are searched as an index, and only the words with the new and old address notation attributes of the address display number and the vertical and horizontal writing attributes are extracted. Match.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an address reader and method for reading address information written on a mail in order to automatically classify the mail.

[0002]

2. Description of the Related Art If, for example, "3-8-1, Nishikoigakubo, Kokubunji, Kokubunji, Tokyo" is described as address information, "3-8-1", Nishikoigakubo, Kokubunji, Kokubunji, Tokyo The part described by numbers is defined as the address display number information. Conventionally, as a mail automatic reading classification method, a method of classifying town area information is known. Since this has a hierarchical structure with "Tokyo", "Kokubunji City", and "Nishi-Koigakubo", even if one or two characters cannot be recognized, the hierarchical structure is used to interpolate it by knowledge processing. so,
This is because it is possible to recognize the entire town area information. On the other hand, the address display number information does not have such a hierarchical structure, and there are various description forms even for the address display numbers having the same meaning. For example, "3rd, 8th, 1st,""3rd,8th"
"-1,""3-8-1," and "1 of 3-8" mean the same address display number.

Conventionally, as a method of classifying such address display numbers, for example, Japanese Laid-Open Patent Publication No. 6-124366 or the Institute of Electronics and Communication Engineers, NCL92-26, PR.
U92-40 (October 1992) "Chome / block recognition method for address reading" is known. In this method, after recognizing each cut-out character pattern, labeling is performed according to the character type of the recognized result, and the address display number part is divided into a numerical value part and a delimiter information part.

For example, "3-8-1" is "NDDND"
N ”(N: label indicating arithmetic number, D: label indicating delimiter information). After that, prepare multiple processing functions according to all combinations of labels in advance, and call the individual processing functions according to the above-mentioned labeling,
It recognizes the address display number.

[0005]

However, the above method has two major problems. The first problem is that it cannot adequately deal with misrecognition. For example, when the address display number "3 chome 8-1" is entered and the recognition result of "chome" includes only recognition candidates other than numbers and characters representing delimiter information, labeling may be performed. I can't respond because I can't. Alternatively, if the area of the address display number cannot be detected, labeling cannot be performed and the address display number cannot be recognized.

Further, in the above method, the characters representing the delimiter information are not distinguished by being represented by one label. Therefore, for example, the "eye" of "3-8-1" is mistakenly recognized as "8".
If there is such a recognition candidate, the label will be labeled as "NDNNDN", and since the numbers are separated by delimiter information, the labeling will be apparently consistent.

This problem can be solved by separating the delimiter information in detail. However, in this device, for the purpose of high-speed processing, individual processing functions corresponding to labeling are provided and each processing is performed. Therefore, when trying to classify and process various notation patterns of address display numbers in detail, it is necessary to provide an extremely large number of processing functions.
It will be very difficult.

The second problem is that since it is an individual processing function system, it is necessary to generate a new function when registering a new notation pattern, and maintenance is not easy.

The present invention solves the above problems without sacrificing the processing speed, and when the address display number cannot be automatically recognized, the operator is required to input the correct address display number. The purpose is to provide assistance. Other purposes will be apparent from the description in the specification.

[0010]

In order to solve the above problems, the present invention basically provides five features. The first feature is that word matching is performed using wildcards that represent arbitrary numbers. As a result of character segmentation and recognition from the address information of the mail, a candidate character lattice that stores a candidate character group and the corresponding penalty is obtained.
After that, the beginning of the address display number area is detected by knowledge processing, and word matching is performed in order to extract the correct address display number from the candidate character lattice of the address display number area. However, since the numbers indicating the claws and numbers can be arbitrary numbers, it is practically impossible to hold the word of the address display number for all combinations of numbers in terms of storage capacity and processing speed. Therefore, a candidate character lattice is created by converting the candidate characters of the candidate character lattice into a wildcard that represents an arbitrary number. Then, if the word with the address notation number represented by a wild card and the wild card lattice are matched, the same word matching as the town area matching can be performed for the address notation number, and the word with the correct address display number is extracted. be able to. Then, as a result of the word matching, if the numeral portion of the wildcard word that has been selected as the upper candidate is restored to the original numeral while referring to the original candidate character lattice, a correct candidate can be obtained.

Details of the first characteristic item will be described. FIG. 6 shows the candidate character lattice for the address display number area. FIG. 6A is a candidate character table storing the character codes of the candidate characters, and FIG. 6B is the penalty for each candidate character. FIG. 7 shows an example of the conversion table to the wild card. In FIG. 7, for example, arithmetic numerals are converted to “n”, Chinese numerals are “k”, and delimiters are converted to “Ding”, “eye”, “ban”, “go”, “−”, “|”, etc. It All characters not used in the address notation number are converted to "e" as other characters.

FIG. 8 shows an example of the wild card lattice generated using such a table. FIG. 8A is a wild card table, and FIG. 8B is a corresponding cost table. Here, if the same wild card is duplicated for one entry character, only the highest wild card is written in the wild card table, and other wild cards are omitted. For example, looking at the candidate character with the written character number 12 in FIG. 6 (a), there are candidates for arithmetic numbers in the first, second, and fifth positions, respectively. Only the card "n" is written in the corresponding place in the wild card table of FIG. 8 (a). Then, at the location corresponding to that in the cost table of FIG. 8 (b), the first cost "0"
Write.

On the other hand, the word for matching is stored in the dictionary in a representation such as "n-chome n-ne" using a wild card. Therefore, a correct word can be obtained by generating an automaton for word matching from the wild card lattice and matching with the word represented by the wild card. Here, in the automaton, candidate characters and their corresponding costs are assigned to the transition paths between states, and the corresponding costs are accumulated while the word transitions between states. . As a result, the average cost divided by the number of characters for each word is obtained, and the word with the low cost is listed as the top candidate.

FIG. 9 shows an example of the automaton. In the automaton 191, circles indicate states, and numbers written therein indicate state numbers. In addition, the state-to-state correspondence corresponds to each written character position of the word, and the line between the states indicates a transition path. The character on the left side of the transition path indicates that when the character is input to the automaton in a certain state, the transition path is followed to transition to the next state. Further, "other" on the transition path represents all characters other than those explicitly specified as the characters corresponding to the transition path. The number in brackets [] on the transition route is the cost of having a transition along the route.

As an example, using the automaton 191,
Consider the cost calculation when the word "n-chome n-ne" 190 is entered. First, when the state 1 transits to the state 2, the cost [0] of “n” is added, and the state 2 to the state 3 are added.
When transitioning to, the cost [1] of "Ding" is added, and the transition proceeds in the same manner. Then, after all the transitions of the number of characters of the word are completed, the integrated cost is divided by the number of characters of the word to obtain the average cost 192 of the word.

The second feature is that an index is provided for high-speed word matching. It is necessary to register many words in order to deal with various notation patterns of the address display number, but if all the word groups are collated, the processing time becomes enormous. Therefore, an index is provided so that unnecessary words are not matched. The three indexes are shown below.

The first index is to use the first and second candidate characters of the entered character as an index. It is highly likely that the first and second candidate characters that include a word character are correct words, and those that leak to that are unlikely. Therefore, by matching only the words including the candidate characters corresponding to the first and second characters, the correct solution can be searched at high speed without matching all the words. The second index is a flag indicating the old and new address. First, a flag related to the notation of old and new addresses in a town area is registered in advance in the word dictionary for town area matching. On the other hand, the wildcard word is provided with a flag indicating which of the new and old address notation patterns it corresponds to. Then, when searching for a word, by checking those flags, it is possible to prevent unnecessary word matching. The third index is a vertical and horizontal writing flag. When performing character recognition, the flags related to vertical and horizontal writing are output. On the other hand, the flags related to vertical and horizontal writing are also provided for the wildcard words, and by comparing these flags, unnecessary word matching can be performed. Can be avoided.

A third feature is that word matching is performed at various entry character positions when the beginning of the address display number area cannot be detected by town area matching. First, the candidate character lattice starting from the beginning of the address is searched for all numerical candidates, and the entry character position including them is stored. This is because the address display number always starts with a number. Then, assuming that the searched character position is the beginning of the address display number, word matching is performed from all the assumed character positions. As a result, the address display number written in any position can be checked.

The fourth characteristic item is an address display number range dictionary which holds a range in which numbers can be hierarchically set for the number part of the address display number. The possible range of each numerical part such as chome, number, and number is registered in the dictionary in a hierarchical manner in advance. For example, Kokubunji City has only 4 chomes, 4
Information such as that the number of chome is only 8 or that the number of chome 4 is 8 is only hierarchically retained. And
By comparing this address dictionary with the candidates for address display numbers that have been restored from wildcards to numbers, it is possible to remove the impossible address display numbers from the candidates.

The fifth feature is to determine whether the address display number entered by the operator using the fourth feature is correct. If the correct address display number is not obtained as a result of the collation, the operator inputs the address display number while checking the address of the mail. At this time, in order to prevent input mistakes, it is determined using the address display number range dictionary whether the input address display number is correct.

The above-mentioned five basic features can be regarded as inventions in terms of both a device and a method.

[0022]

The present invention has the following five basic actions. First, the address display number can be recognized even if no correct candidate character is found in the character recognition. In the present invention, the notation pattern of the address display number is held as a dictionary word expressed by using a wildcard that represents an arbitrary number, and the candidate character group of the recognition result and the cost of the word are calculated and collated. ， Can recognize the address display number. Therefore, even when a correct candidate does not appear in the candidate character group of the recognition result of a part of the address display number, the address display number can be recognized by interpolating the correct candidate. For example, even if the character "Ding" does not appear as a recognition candidate for the character pattern corresponding to "Ding" in "3rd chome 8-1", if the cost of the corresponding word is small as a whole, " The address number can be recognized by interpolating "Ding".

In addition, since the notation pattern of the address display number is held in the form of a dictionary, the notation pattern can be viewed in more detail than preparing and processing an individual function, thus preventing erroneous recognition. it can.

Second, although the address display number can be examined in detail, it is processed at high speed.
First, when a word is searched from a dictionary, a candidate character group corresponding to each character pattern is searched as an index, so that the number of words to be matched can be reduced. Furthermore, since each word has an attribute corresponding to the old and new address notation and vertical and horizontal writing, it is possible to know in advance whether the address display number to be recognized belongs to the new or old address notation or the vertical and horizontal writing. If so, unnecessary word collation can be prevented and high-speed processing becomes possible.

Thirdly, since it is a dictionary system, when a new writing pattern occurs, it can be easily registered in the dictionary and maintenance is easy.

Fourthly, since there is an address display number range dictionary that holds a range in which the numbers of the address display numbers can be hierarchically held, it is possible to exclude the address display number candidates that cannot actually exist.

Fifth, as a result of the address display number collation, if no address display number candidate is obtained, the operator can correctly input the address display number while looking at the image of the address area of the mail. Since the address display number range dictionary that hierarchically holds the values of the range that can be taken by the numerical part of the address display number is stored, whether the numerical part of the address display number entered by the operator is within the correct range, It can be determined by referring to the address display number range dictionary. If the result of the determination is that it is out of the range, the operator is warned, so that an input error by the operator can be prevented.

[0028]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A first embodiment of the present invention will be described below with reference to FIGS.
Will be explained.

FIG. 1 is a block diagram of the entire apparatus of this embodiment. The postal matter 100 is sequentially sent to the supply unit 101. In the supply unit 101, the mail piece passes through a predetermined position, and an image of the surface of the mail piece is captured by the image input unit 102 during the passage. The address recognition unit 106 reads the address written on the surface of the postal matter and generates classification information. On the other hand, the postal matter whose surface image has been captured is sent to the delay conveyance path 103. In the delay conveyance path 103, the mail is sent for the predetermined time for generating the sorting information from the surface image,
To move. The sorting unit 104 sorts the mail according to the sorting information from the address recognition unit 106, and then sorts the sorting shelves 10
Store in 5. The image input unit 102 digitizes an image signal from a photoelectric conversion element such as a line sensor, and extracts a character line of an address from an image on the surface of a mail.

The address recognition unit 106 comprises a control unit 107, an image processing unit 108, a character recognition unit 109, and a knowledge processing unit 110. The control unit 107 includes an image processing unit 108, a character recognition unit 109, and a knowledge processing unit 110, respectively. Control. The knowledge processing unit 110 includes a town area collation unit 111 and an address display number collation unit 112. The town area collation unit 111 accesses the town area word dictionary 113, and the address display number collation unit 112 accesses the house display number dictionary group 114. Then, an error in the recognition result of the character recognition unit 109 is automatically corrected.

The address display number dictionary group 114 includes an address display number word index dictionary 115, an address display number word dictionary 116, and an address display number range dictionary 117. The address display number word dictionary 116 is a dictionary that stores the notation pattern of the block, and the address display number word index dictionary 115.
Is a dictionary that stores indexes for selectively searching for required words from the address display number word dictionary 116.
The address display number range dictionary 117 is a dictionary in which each range of chome, block, and house display number in each address display number is hierarchically described.

FIG. 2 shows the flow of the entire processing of the address recognition unit 106 in FIG. The operation of the address recognition unit 106 will be described with reference to FIG.

In step 120, the image input unit 102 captures an image of the surface of the postal matter on which the address is written, and inputs the image to the address recognition unit 106. In step 121, the image processing unit 108 extracts the address area from the entire image of the address written from the image input unit 102. In step 122, the image processing unit 108 extracts an image for each character line from the image of the address area obtained in step 121. In step 123, the character recognition unit 109 causes step 122
After the character image is cut out for each character from the image of the character line obtained in step 1, a character is recognized for each cut out character and converted into a character code. Here, as a result of character recognition, similarities corresponding to a plurality of candidate character codes are obtained.
In addition, information on the direction of characters, that is, vertical writing or horizontal writing can be obtained.

In step 124, the town area collating section 111 in the knowledge processing section 110 generates a candidate character lattice based on the character recognition result obtained in step 123. Here, an example of the candidate character lattice is shown in FIG. These are, for example, when the address "3-8-1, Nishi-Kogokubo, Kokubunji-shi, Tokyo, Hitachi Dormitory" is written in the address of the mail, it is the part of "Nishi-Kogakubo, Kokubunji-shi, Tokyo" that represents the town area information. It is an example of a candidate character lattice. FIG. 4A is an example of a candidate character table in the candidate character lattice, and FIG. 4B is an example of a cost table corresponding to it. In the candidate character table of FIG. 4 (a), the first to mth positions (m is arbitrary) for each entry character number.
Up to, the character codes of the candidate characters are arranged. FIG.
In the cost table of (b), the cost for the candidate character is stored at a position corresponding to the position where the character code of the candidate character table of FIG. 4 (a) is stored. For example,
The cost of the first candidate character “East” of the written character number 1 is “0”.

In step 125, the town area collation unit 111 in the knowledge processing unit 110 uses the town area word dictionary 113 to convert the candidate character lattice obtained in step 124 into town area information of an address and town area information. Extract the corresponding old and new address notation information. As a method of extracting town area information, a method is known in which an automaton is generated from a candidate character lattice and word matching is performed. The details are described in Japanese Patent Laid-Open No. 3-125288.

Next, the construction of the town area word dictionary 113 and the method of extracting new and old address notation information corresponding to town area information will be described with reference to FIG. The town area word dictionary 113 stores words hierarchically. For example, as prefecture level words, Saitama prefecture 142, Tokyo 143, Kanagawa prefecture 144.
There are municipality-level words below Tokyo 143 such as Koganei city 145, Kokubunji city 146, and national city 147. Below the Kokubunji city 146, there are town level words, Hiyoshi-cho 148, Nishikoigakubo 149, and Higashikoigakubo 150. Further, in the word at the town area level, 7-digit town area division numbers 151 to 153 for identifying town area information nationwide, and flags 154 to 156 indicating which of the old and new address areas each town area corresponds to. Are stored together.
The flag of the new and old address notation takes a value such as 1 if the town area corresponds to the new address notation and 2 if the town area corresponds to the old address notation. Therefore, when word information is hierarchically obtained and town area information is obtained, by referring to the corresponding part of the dictionary,
A 7-digit town area classification number for identifying town area information and a flag indicating the old and new address in the dictionary can be obtained at the same time. The 7-digit town area division number is used in the address display number range dictionary 117 in order to determine the possible range of the numerical portion of the address display number in the subsequent processing.
It is used as an index when searching for and as a part of the control information for sorting the mail by the sorting unit 105. Further, the new and old address notation flags are used as an index for limiting the number of words to be matched from the address display number word dictionary in order to perform address display number matching. The town area information, town area division number, and new and old address notation flag are obtained by the town area collation processing in step 125 described above.

In step 126, the address display number collation unit 112 in the knowledge processing unit 110 causes the address display number dictionary group 1
14 is used to extract the address display number from the candidate character lattice obtained in step 124. Details of step 126 will be described in detail with reference to FIGS. 3 and 5 to 11. FIG.
In steps 132 and 135, the feature of the present invention is.

In step 130, step 12 in FIG.
Address display number collation unit 1
Enter in 12. In step 131, the entry character number at the head of the area in which the address display number of the candidate character lattice is written is detected. This is utilized because the end of the written area of the town area can be detected by the town area collation performed in step 125 of FIG. In step 132, a wildcard lattice is generated from the candidate character lattice obtained in step 130. Here, the wildcard lattice is a lattice in which the numbers in the candidate character lattice are replaced with wildcards representing arbitrary numbers in order to match the address display numbers.

Details of step 132 are shown in FIG. 6, FIG. 7 and FIG.
Will be explained. FIG. 7 is a conversion table used to generate the wild card lattice, which will be described in detail below.

The table of "numerals" in the classification is a table relating to numerals representing chome, block, and house display number. Converted to wildcards "n" and "k".

The "delimiter" table is a table relating to the characters "chome" and "-" used to separate the numbers of the chome and the block when, for example, "3 chome 8-1" appears in the address. Is. Here, "Ding", "eyes",
"Number", "ground", "no", and "no" are not converted and the characters are used as they are. On the other hand, "-", "~",
"/" Is converted to a wildcard "-".

The "other" table is a table relating to characters other than the "numeric" table and the "delimiter" table, and all the characters other than those mentioned above are converted into a wildcard "e". That is, "other" refers to all but the characters related to the chome, block, and house display number.

FIG. 8 shows an example in which a wild card lattice is generated from the candidate character lattice shown in FIG. 6 using the conversion table shown in FIG. Wild card lattice is Figure 8
The wild card table shown in (a) and the cost table shown in FIG. 8 (b) are used.

The generation of the wild card lattice shown in FIG.
In the candidate character table of FIG. 6A, the candidate characters are ranked in descending order. First, the first-ranked candidate character is converted into a wildcard and written in the first-ranked place of the wildcard table shown in FIG. At the same time, the cost corresponding to the converted candidate character is written in the corresponding place in the cost table of FIG. 8 (b).

Next, the second-ranked candidate character is examined, and if it belongs to the same wildcard as the first-ranked candidate character, it is omitted because it is duplicated. If they belong to different wild cards, the costs corresponding to the wild cards are written in the wild card table of FIG. 8A and the cost table of FIG. 8B, respectively. Hereinafter, the same thing is repeated for all the candidate characters.

For example, consider conversion of the column of the entry character number 12 in the candidate character table of FIG. 6 (a). First, the first
The candidate character "8" for the rank is converted into a wild card "n" and written in the first place of the entry character number 12 in the wild card table of FIG. 8 (a). At the same time,
The cost "0" corresponding to the candidate character "8" is written in the corresponding place of the cost table shown in FIG. 8 (b).

Next, looking at the second place, since the candidate character "3" belongs to the same wildcard "n", it is omitted in FIG.
Nothing is written in the wildcard table in (b). Further, looking at the 3rd place, the candidate character "day" belongs to the wildcard "e" different from "n", so it is placed in the 2nd place that is vacant in the wildcard table of FIG. 8 (a). Write. At the same time, the cost "2" of the candidate character "day" is written in the corresponding location of the cost table of FIG. 8 (b). Hereinafter, the same processing is performed on the candidate characters of all ranks.

Through the above processing of step 132, a wildcard lattice is generated from the candidate character lattice obtained in step 130.

In step 133, an automaton for word matching is generated from the wild card lattice generated in step 132. Details of step 133 will be described with reference to FIG.

FIG. 9 shows a process of extracting a word from a dictionary and matching address display number words by using an automaton. First, the automaton 191 will be described. A finite state automaton 191 is generated from the wild card lattice shown in FIG. 8 in order to match the dictionary word representing the address display number pattern with the wild card lattice. Then, the automaton 191 sequentially inputs the dictionary words represented as a character string, and calculates what the average cost of the words is.

In the automaton 191, a circle indicates a state, and a number written therein indicates a state number. In addition, between states corresponds to each written character position of the word,
Lines between states indicate transition paths. The character on the left side of the transition path indicates that when the character is input to the automaton in a certain state, the transition path is followed to transition to the next state. Further, "other" on the transition path represents all characters other than those explicitly specified as the characters corresponding to the transition path.
The number in brackets [] on the transition route is the cost of having a transition along the route.

As an example, using the automaton 191,
Consider the cost calculation when the word "n-chome n-ne" 190 is entered. First, when the state 1 transits to the state 2, the cost [0] of “n” is added, and the state 2 to the state 3 are added.
When transitioning to, the cost [1] of "Ding" is added, and the transition proceeds in the same manner. Then, after all the transitions of the number of characters of the word are completed, the integrated cost is divided by the number of characters of the word to obtain the average cost 192 of the word.

Next, the generation of the finite state automaton 191 from the wildcard lattice of FIG. 8 will be described. First, a transition path between the state 1 and the state 2 of the automaton 191 of FIG. 9 is generated, and the candidate character of the entry character number 12 of the wildcard table of FIG. At the same time, the costs in the cost table shown in FIG. 8B corresponding to the candidate characters are similarly assigned to the transition paths between the states 1 and 2, respectively. Next, "other" is generated as a transition path that represents all characters other than the specified one, and the cost is set to 15. Similarly, states and transition paths are generated one after another from the wild card lattice.
Here, the cost of the transition route does not necessarily have to be the above value, and may be an arbitrary number.

By the processing of the above step 133, the automaton 191 is generated from the wild card lattice.

In step 134, the automaton matching of the address display number word is performed using the automaton 191 generated in step 133 and the address display number word index dictionary 115 and the address display number word dictionary 116 of FIG.
Details of step 134 will be described with reference to FIGS. 8, 9, 10, and 11. FIG. 10 is a PAD showing the flow of the automaton matching process. FIG. 11 is a diagram showing the configurations of the address display number word dictionary and the address display number word index dictionary.

First, the structure of the dictionary shown in FIG. 11 will be described. The address display number word dictionary 116 is a word representing a notation pattern of chome, block, and house display number using wild cards,
And a dictionary that stores the search information and attributes thereof. For example, “n-chome n-ne” and “n | n | ne” can be used as words that represent the notation patterns of chome, block, and house display number.
Etc. are stored. The address display number word index dictionary 115 is a dictionary that stores an index for selectively retrieving a word necessary for matching from the address display number word dictionary. The index is the first of the dictionary words
The first and second characters are used.

The address display number word dictionary 116 includes a word 228 representing a block notation pattern, a relative address 229 between words having the same second character as the second character, a flag 230 indicating old and new address notation, and vertical and horizontal writing. The flag 231 indicates.

The flag 230 indicating the new and old address notation corresponds to either 1 if the word representing the block notation is the notation pattern of the new address notation or 2 if it is the notation pattern of the old address notation. The number 3 is stored. The flag 231 indicating vertical / horizontal writing stores a number of 1 if the word representing the block notation belongs to horizontal writing, 2 if it belongs to vertical writing, and 3 if it belongs to both vertical and horizontal writing.

Address display number word index dictionary 115
Is composed of two tables, a first character index table 220 and a second character index table 224.
The first character index table 220 includes a table 221, which stores the character number of the first character of the dictionary word, the number 222 of words having the same first character, and a pointer table 223 to the address display number word dictionary 116. Similarly, the second character index table 224 is a table 225 storing the second character number of the dictionary word, the number 226 of words having the same second character, and the address display number word dictionary 11.
6 to the pointer table 227.

Next, the flow of processing when searching for words in the address display number word dictionary 116 using the address display number word index dictionary 115 will be described. The arrow shown by a solid line indicates a search flow when searching for a word in which the first character is the same character using the first character index table 220. The arrow indicated by the dotted line shows the flow of search when searching for a word in which the second character is the same character using the second character index table 224.

For example, when a word in which the first character of the word is "n" is searched from the dictionary, the number of words in which the first character of the first character index table 220 is the same character
The pointer P1 (1) of m1 and "n" is referred to. Since the pointer P1 (1) stores the first pointer of the word in which the first character starts with "n" in the address display number word dictionary, that word is referred to. In the word dictionary, words with the same first character are arranged in a row, so if you find the first word starting with "n", you can search words one after another by simply incrementing the pointer. it can. Then, when the search is performed m1 times, the end of the word starting with "n" is reached, and the word search is terminated at that point.

The same applies when the second character of a word is searched as an index. For example, when searching for a word in which the second character of the word is "Ding" from the dictionary, the number of words q1 in which the first character of the second character index table 224 is the same character, and a pointer of "Ding" See P2 (1). Since the pointer P2 (1) stores a pointer of a word whose second character is "Ding" in the address display number word dictionary, the word is referred to. After that, by referring to the relative address 229 between words in which the second character of the word has the same character and shifting the pointer, it is possible to retrieve the word of "Ding" having the same second character. Then, when the search is performed q1 times, the second character ends at the end of the word with the same "ding", so the word search ends there.

Here, since it is known whether the address is in vertical writing or horizontal writing in the character segmentation / character recognition in step 123 of FIG. 2, by referring to the flag 231 indicating vertical / horizontal writing, from among the searched words, You can narrow down and extract the relevant words.

In addition, since it is known by the town area collation in step 125 of FIG. 2 which address notation the new or old address belongs to, the flag 230 indicating the old and new address notation is referred to, and You can narrow down and extract the corresponding words from.

Next, the specific processing contents of step 134 will be described with reference to the processing flow of FIG. 10 and FIGS. 8, 9 and 11.

In step 200, a flag indicating vertical and horizontal writing of the address obtained in step 123 of FIG. 2 and a flag representing the old and new address notation obtained in step 125 are set for the address to be collated. At step 202, the first candidate character of the entry character number 12 in the wild card table of FIG.
By referring to the character index table 220, the dictionary word number 222 and the word dictionary pointer 223 are acquired. In step 204, the word pointed to by the pointer is searched, and whether the word is consistent with the old and new address notation flags set in step 200 and the vertical and horizontal writing flags is referred to by the corresponding tables 230 and 231 in the dictionary. To check. If there is no contradiction, go to step 205. In step 205, the retrieved word is input to the automaton 191 of FIG. 9 and the average cost is calculated while transitioning the states. In step 203, step 204
The process from step 205 to step 205 is repeated by the number of dictionary words obtained in step 202.

In step 201, the candidate character of the next rank of the entered character address 12 in FIG. 8A is taken out, and the processing in steps 202 to 205 is performed.
Repeat the candidate character several times. In step 207, FIG.
The first candidate character of the entry character number 13 in the wildcard table of (a) is taken out, and the number of dictionary words 226 and the pointer 227 of the word dictionary are acquired by referring to the second character index table 224 of FIG. . Step 2
At 09, the word pointed to by the pointer is searched, and it is checked by referring to the corresponding table in the dictionary whether this word is consistent with the old and new address notation flags set in step 200 and the vertical and horizontal writing flags. If there is no contradiction, go to step 210.

In step 210, the retrieved word is searched for in FIG.
To the automaton 191 to calculate the average cost while transitioning the states. In step 208, the processes of steps 209 to 210 are repeated for the number of dictionary words obtained in step 207. In step 206, the candidate character of the next rank of the entered character address 13 in FIG. 8A is taken out, and the processing from step 207 to step 210 is repeated for the number of candidate characters in the entered character address 13.

In step 211, the words obtained by the processing in steps 200 to 210 and their costs are rearranged in the order of ascending power. In step 212, the upper L (L> 1) words rearranged in step 211 are selected.

Steps 200 to 212 above
By the processing of (1), the automaton collation of step 134 in FIG. 3 is performed, and the upper L words having a small average cost and the costs thereof are obtained.

In step 135, for the words represented by the top L wildcards obtained in step 134, the wildcards representing the numbers "n" and "k" are restored to the original numbers and candidates are selected. To generate. Here, FIG. 12 shows a result of restoring the word “n-chome n-ne” 190 of FIG. 9 into a numeral with L = 1. Figure 6 shows the process of digit restoration.
Candidate character table 160 and the word “n-chome nn” shown in FIG.
190 is aligned. After that, the candidate number of the place corresponding to the number “n” and its cost are shown in FIG.
(a) Generate an actual chome or block by extracting it from the candidate character table. Further, the costs corresponding to the numbers are extracted from the cost table of FIG. 6 (b), and the costs for the restored address display number words are accumulated.

At step 136, the address display number word candidates restored to numbers at step 135 are selected as follows.
It is determined whether or not there is a contradiction in the numerical portion of the block and residence display number with the address display number range dictionary 117 of FIG. Here, the address display number range dictionary 117 is a dictionary that hierarchically stores range information indicating which range of the chome, block, and house display number can be set for each town area. Details of the address display number range dictionary 117 will be described with reference to FIG.

The address display number range dictionary 117 includes an index table 260 and an address display number range table 263.
Consists of The index table 260 includes a town area division number table 261 for identifying town areas and a pointer table 262 for an address display number range table 263.
The address display number range table 263 is a chome table 264 which stores chome numbers as indexes, and a block table 265 which stores clique numbers as indexes.
The house display number table 266 stores the maximum value of the house display number.

Next, the flow of processing when referring to the address display number range will be described. For example, in order to refer to the address display number range of "Nishi-Koigakubo, Kokubunji-shi, Tokyo", first, correspond to "Nishi-Koigakubo, Kokubunji-shi, Tokyo" that was obtained using the town area word dictionary shown in Fig. 5 in step 125 of Fig. 2. Index table 2 for town area classification number "1850002"
The town area classification number table 261 of 60 is referred to. "18
The pointer Pa corresponding to "50002" is "Nishikoigakubo, Kokubunji-shi, Tokyo" in the address display number range table 263.
The beginning of the area in which the range data is stored is referenced. In the area, the maximum value of the house display number is stored in the house display number table 266 using the chome table 264 and the block block table 265 as indexes. Therefore, by searching the corresponding chome and block index, for example, the maximum value of the house display number of "3-8" is 9
By searching all areas of "Nishi-Koigakubo, Kokubunji-shi, Tokyo", you can see that the maximum value of chome is only 4. For example, if there are only 4 chomes in "Nishi-Koigakubo, Kokubunji-shi, Tokyo" and 3 chomes have only house display numbers 5, the candidate group in FIG. 12 is narrowed down to the candidates shown in FIG. By the process of step 136 described above, the range of each chome of the address display number, the block, and the house display number is determined, and the candidates determined to be out of the range are deleted from the candidate group of FIG.

Here, the address display number range dictionary is not limited to this embodiment. For example, the house display number portion may have not only the upper limit value but also the lower limit value at the same time. Alternatively, all the house display numbers actually targeted for mail delivery may be registered, excluding the house display numbers of parking lots.

In step 137, P (P> 1) candidates having a small cost are selected from the candidates narrowed down in step 136 as the address display number collation result. Here, P
= 2, "3 chome 8-1" and "3 chome 3-" in FIG.
"1" is selected as the matching result.

Steps 130 to 137 above
By the processing up to step 127, the address display number collation in step 127 of FIG. 2 is performed, and as a collation result, the address display number candidate “3” is displayed.
Chome 8-1 "and" 3 chome 3-1 "are obtained.

At step 127, the town area candidate “Nishi-Koigakubo, Kokubunji-shi, Tokyo” obtained by the town area collation at step 125 is obtained.
Then, the address display number candidates “3 chome 8-1” and “3 chome 3-1” obtained by the address display number collation in step 126 are connected to generate an address candidate. In this example, "3-8-1, Nishikoigakubo, Kokubunji-shi, Tokyo" and "3-3-1 Nishikoigakubo, Kokubunji-shi, Tokyo" are obtained. Further, using this address information, control information for controlling the sorting unit 104 in FIG. 1 is generated.

A second embodiment of the present invention will be described with reference to FIGS. 1, 2, 9 and 15. Here, consider the case where the town area information cannot be obtained by the town area collation when the 7-digit town area classification number is printed on the address.

In FIG. 2, steps 120 to 122 are the same as those in the first embodiment.

In step 123, the character recognition unit 109 recognizes each character from the image of the character line obtained in step 122 and converts it into a character code, as in the first embodiment. However, here, not only the address information but also the town area classification number printed on the address is recognized and converted into a character code. In step 124, the town area matching unit 111 generates a candidate character lattice based on the character recognition result obtained in step 123, as in the first embodiment. In step 125, the town area matching unit 111 uses the town area word dictionary 113 to perform town area matching, as in the first embodiment. However, in the present embodiment, it is assumed that, as a result of the town area collation, the town area information and the new and old address notation information corresponding to the town area information cannot be extracted. In step 126, the address display number collation unit 112 in the knowledge processing unit 110 of FIG.
4, the address display number is extracted from the candidate character lattice obtained in step 124. Details of step 126 will be described in detail with reference to FIGS. 9 and 15. In FIG. 15, steps 302 and 304 are characteristic of the present invention.

In step 300, step 12 in FIG.
Input the candidate character lattice generated in 4.

In step 301, a wild card lattice is generated from the candidate character lattice obtained in step 130 by the same method as in the first embodiment. However, since the beginning of the address display number cannot be detected because the town area information cannot be obtained, the wild card lattice is generated from the beginning of the address. In step 302, all entered character numbers including candidate numbers are detected from the wild card lattice. That is, "n" or "k" in the wildcard table
Extract the entry character number that includes. In step 303, the entered character number 1 is obtained from the wild card lattice generated in step 301 by the same method as in the first embodiment.
Generate an automaton for word matching from. Then, for example, it is assumed that the automaton 191 of FIG. 9 is obtained. Here, only the automaton of the address display number part is displayed.

In step 304, the automaton 191 generated in step 303 and the address display number word index dictionary 115 and the address display number word dictionary 116 shown in FIG.
However, the difference from the method of the first embodiment is that, assuming that the number of states of the generated automaton is K (K> 1), it cuts at K cutting points, and for the rear partial automaton starting from each cutting point. The word matching of the first embodiment is performed.
For example, in FIG. 9, word matching was performed on the automaton starting from state number 1, but it was changed to state number 2,
Similar word matching is performed for automata starting from 3 ,. As a result, the word existing at an arbitrary position can be extracted. When selecting a word from the dictionary, since the attributes of the old and new address notation cannot be obtained in step 125 of FIG. 2, the word is selected without looking at the flag of the old and new address notation in the dictionary.

At step 305, the upper L (L> L) obtained at step 304 is processed by the same method as in the first embodiment.
1) For each word represented by wildcards,
Wild cards representing the numbers "n" and "k" are restored to the original numbers to generate candidates for address display numbers.

In step 306, P candidates having a small cost are selected as the address display number collation result from the candidates generated in step 305 by the same method as in the first embodiment. The address display number collation of step 126 in FIG. 2 is performed by the above processing from step 300 to step 306.

In step 127, the town area division number obtained in step 123 and the address display number obtained in step 126 are combined to generate control information for controlling the division section 104 in FIG.

By the processing from step 120 to step 127, it is possible to obtain the control information for controlling the sorting unit 104 even when the head of the address display number area is not found by the town area collation.

A third embodiment of the present invention will be described with reference to FIGS. 2, 16, 17, and 18.

In FIG. 16, the feature of the present invention is the non-reading correction section 310 and the address display number range dictionary 117. FIG.
The operation of the device will be described with reference to FIG.

Steps 340 to 345 correspond to steps 120 to 1 in FIG. 2 in the first embodiment.
The same processing as that up to 25 is performed. Step 346
Then, the address display number is collated by the same method as in the first embodiment. However, consider the case where no candidate is obtained as a result of the matching. In step 347, the control unit 107 of FIG.
The address display number is input by the operator's input work in the unreadable character correction unit 310. FIG. 18 is an example of a display screen for input work. FIG. 16 shows 360.
It is an address image of a mail item captured by the image input unit 102 of FIG. 361 is a result of the knowledge processing unit 110. Since there is no address display number candidate, the part corresponding to the numerical value is "?".
Indicated by. Numerals 362, 363, and 364 are frames for the operator to input the numerical values of the chome, block, and house display number in the address display number, respectively. Here, the knowledge processing unit 110 uses the address display number range dictionary 117 to determine whether the value input by the operator is within the correct range. If the result of the judgment is that it is outside the range, the fact is displayed to the operator. The address display number is input by the processing in step 347 described above.

In step 348, the town area information obtained in step 345 and the address display number input in step 346 are combined to generate an address candidate. As described above, by the processing from step 340 to step 348, even if no candidate is obtained by the address display number collation, the operator can input the correct address display number with the assistance of the apparatus, and the correct address can be obtained.

In the above embodiment, the example of the address display number has been described by using the new address notation, but the same processing is applied to the address display number of the old address notation.

[0094]

The present invention has the following five effects. First, the address display number can be recognized even if no correct candidate character is found in the character recognition. In the present invention, the notation pattern of the address display number is held as a dictionary word expressed by using a wildcard that represents an arbitrary number, and the candidate character group of the recognition result and the cost of the word are calculated and collated. ， Can recognize the address display number. Therefore, even when a correct candidate does not appear in the candidate character group of the recognition result of a part of the address display number, the address display number can be recognized by interpolating the correct candidate. For example, even if the character "Ding" does not appear as a recognition candidate for the character pattern corresponding to "Ding" in "3rd chome 8-1", if the cost of the corresponding word is small as a whole, " The address number can be recognized by interpolating "Ding".

In addition, since the notation pattern of the address display number is held in the form of a dictionary, the notation pattern can be seen in more detail than preparing and processing individual functions, so that erroneous recognition can be prevented. it can.

Second, although the address display number can be examined in detail, it is processed at high speed.
First, when a word is searched from a dictionary, a candidate character group corresponding to each character pattern is searched as an index, so that the number of words to be matched can be reduced. Furthermore, since each word has an attribute corresponding to the old and new address notation and vertical and horizontal writing, it is possible to know in advance whether the address display number to be recognized belongs to the new or old address notation or the vertical and horizontal writing. If so, unnecessary word collation can be prevented and high-speed processing becomes possible.

Thirdly, since it is a dictionary system, when a new writing pattern occurs, it can be easily registered in the dictionary and maintenance is easy.

Fourthly, since there is an address display number range dictionary that holds a range in which the numerical portion of the address display number can be hierarchically held, it is possible to exclude the address display number candidates that cannot actually exist.

Fifth, as a result of the address display number collation, if the address display number candidate is not obtained, the operator can correctly input the address display number while looking at the image of the address area of the mail. Since the address display number range dictionary that hierarchically holds the values of the range that can be taken by the numerical part of the address display number is stored, whether the numerical part of the address display number entered by the operator is within the correct range, It can be determined by referring to the address display number range dictionary. If the result of the determination is that it is out of the range, the operator is warned, so that an input error by the operator can be prevented.

[0100]

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing the flow of the entire processing of the apparatus of the first embodiment of the present invention.

FIG. 3 is a diagram showing a flow of processing for performing address display number matching.

FIG. 4 is a diagram showing an example of a candidate character lattice for performing town area matching.

FIG. 5 is a diagram showing an example of the configuration of a town area dictionary storing town area information.

FIG. 6 is a diagram showing an example of a candidate character lattice of an address display number area.

FIG. 7 is a diagram showing a conversion table for generating a wildcard lattice for performing address display number matching from candidate character lattices.

8 is a diagram showing an example of a wild card lattice generated from the candidate character lattice of FIG. 6 using the conversion table of FIG.

FIG. 9 is a diagram showing an outline of processing of automaton word matching.

FIG. 10 is a flowchart P showing a flow of processing of automaton word matching.
It is AD.

FIG. 11 is a diagram showing configurations of an address display number word index dictionary and an address display number word dictionary.

FIG. 12 is a diagram showing an example of address display number candidates in which a numerical part is restored from the address display number word as a result of matching.

FIG. 13 is a diagram showing an example of the configuration of an address display number range dictionary.

FIG. 14 is a diagram showing candidates remaining after determining a possible range for each numeral part from the candidates of the address display number of FIG. 12;

FIG. 15 is a diagram showing the flow of address display number matching processing in the second embodiment of the present invention.

FIG. 16 is an overall configuration diagram of an apparatus according to a third embodiment of the present invention.

FIG. 17 is a diagram showing the overall flow of processing in a third embodiment of the present invention.

FIG. 18 is a diagram showing a display example of a screen for inputting an unread address display number.

[Explanation of symbols]

100 ... Mail, 101 ... Mail supply section, 102 ... Image input section, 103 ... Delayed transport section, 104 ... Sorting section, 105
... Division shelf, 106 ... Address recognition unit, 107 ... Control unit, 10
8 ... Image processing unit, 109 ... Character recognition unit, 110 ... Knowledge processing unit, 111 ... Town area collation unit, 112 ... Address display number collation unit, 113 ... Town area word dictionary, 114 ... Address display number dictionary group, 115 ... Address display number word index dictionary, 11
6 ... Address display number word dictionary, 117 ... Address display number range dictionary.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masafumi Koga 1-280, Higashi Koikeku, Kokubunji, Tokyo Inside Central Research Laboratory, Hitachi, Ltd. (72) Tatsuhiko Kagehiro 1-280, Higashi Koikeku, Kokubunji, Tokyo Hitachi, Ltd. Central Research Laboratory (72) Inventor Masato Teramoto 1 Ikegami, Haruoka-cho, Owariasahi-shi, Aichi Hitachi Ltd. Office Systems Division (72) Inventor Shigeru Watanabe 1 Ikegami, Haruoka-cho, Owariasahi, Aichi Hitachi Ltd. Office Systems Division (72) Inventor Hiromichi Fujisawa 1-280, Higashi Koikekubo, Kokubunji City, Tokyo Inside Central Research Laboratory, Hitachi, Ltd.

Claims (6)

[Claims] 1. An image input by converting an image on a mail into an electric signal in an address reading device that detects character information on the mail and reads address information including town area information and address display number information. An input unit, a character recognition unit that cuts out and recognizes character information from the input image, and outputs a recognition candidate character group for each cut out character pattern; and a recognition candidate character group that is output from the character recognition unit. A town area collating means for recognizing a town area by collating with a town area dictionary storing town area information, and an address display number area detecting means for detecting the head of the address display number area based on the recognition result of the town area collating means. And a candidate in which the number in the candidate character group corresponding to each character pattern after the beginning of the address display number region is replaced with a wild card representing any number based on the output from the address display number region detection means. A wild card conversion means for converting to a character group, an address display number word dictionary that holds various notation patterns of the address display number expressed by the above wild card as a word, an output result of the wild card conversion means and an address display number word dictionary The address display number collating means for recognizing the notation pattern of the address display number and the output result of the address display number collating means
And a numeral restoring means for collating with the candidate character group output from the character recognizing means, restoring the numbers in the candidate character group replaced by the wild card, and outputting the candidates for the address display number. Address reading device. 2. The address reading device according to claim 1, wherein in addition to the character direction detecting means for detecting whether the character group representing the address information on the mail is vertical writing or horizontal writing, and the town area information. , The town area classification number for identifying the town area, the town area dictionary that stores the information whether the town area is the new address notation or the old address notation, and the recognition result and the address of the town area matching means Based on the notation dictionary, the address notation identifying means for identifying whether the town area is the new address notation or the old address notation, and the address indication in which each word has an attribute concerning the old and new address notation and an attribute concerning the vertical and horizontal writing A number word dictionary and a word search means for searching a word from the house display number word dictionary using the character code of the candidate character group as an index for the candidate character group of each character pattern at any position replaced by a wild card. ， Character direction Based on the output of the output means and the address notation identifying means, from the words searched by the word searching means, only the word in which the attributes of the old and new address notation and the vertical and horizontal writing match from the house display number word dictionary is read out. And an address reading device. 3. The address reading device according to claim 1, wherein when the head of the address display number area cannot be detected based on the recognition result of the town area collating means, the recognition result includes a numerical candidate. Based on the output from the head presumption means that assumes an arbitrary character pattern to be the head of the address display number area and the position assumed to be the head of the address display number area, the output result of the wildcard conversion means Address display number collating means for recognizing the notation pattern of the address display number by collating the words in the house display number word dictionary with an automaton, the output result of the address display number collating means, and the candidate output from the character recognizing means And a numeral restoring means for collating with the character group, restoring the numbers in the candidate character group replaced by the wild card, and outputting the candidates for the address display number. Address reading device, characterized in that. 4. An address reading device that detects character information on a postal matter, reads address information consisting of town area information and address display number information, and determines a range that can be taken by the numerical part of the address display number. Image input means for converting an image on an object into an electric signal and inputting it, and character recognizing means for cutting out character information from the input image and recognizing it and outputting a recognition candidate character group for each cut out character pattern. Based on the recognition result of the town area collating means for recognizing the town area by collating the recognition candidate character group output from the character recognizing means with the town area dictionary storing the town area information, the address Address display number area detection means for detecting the beginning of the display number area,
Based on the output from the address display number area detection means, the numbers in the candidate character group corresponding to each character pattern after the beginning of the address display number area are converted into a candidate character group in which a wildcard representing any number is replaced. The wildcard conversion means, the address display number word dictionary that holds various notation patterns of the address display number expressed by the wildcard as words, the output result of the wildcard conversion means, and the words of the address display number word dictionary are automata. Using the address display number collation means for recognizing the notation pattern of the address display number, the output result of the address display number collation means, and the candidate character group output from the character recognition means.
A number restoring means for restoring the numbers in the candidate character group replaced by the wild card and outputting the candidates for the address display number, and an address display number for hierarchically holding the value of the range of the numerical part of the address display number. With respect to the range dictionary and the candidate of the address display number output from the number restoration means, it is determined whether the value of the numerical part is within the range of the address display number by referring to the address display number range dictionary. Address display number range determining means for narrowing down candidates based on the address display number. 5. An address reading device that detects character information on a postal matter, reads address information consisting of town area information and address display number information, and determines the range that can be taken by the numerical part of the address display number. Image input means for converting an image on an object into an electric signal and inputting it, and character recognition means for cutting out character information from the input image for recognition and outputting a recognition candidate character group for each cut out character pattern And a recognition candidate character group output from the character recognition means, and a recognition candidate character group output from the character recognition means, which recognizes the town area by matching the recognition candidate character group output from the character recognition means with a town area dictionary storing town area information. Based on the recognition result of the town area collation means for recognizing the town area by collating with the town area dictionary storing the town area information, and the address display number area detection for detecting the head of the address display number area Means and , Converting the numbers in the candidate character group corresponding to each character pattern from the beginning of the address display number area to the candidate character group in which the wild card representing any number is replaced based on the output from the address display number area detecting means The wildcard conversion means, the house display number word dictionary that holds various notation patterns of the address display number expressed by the wildcard as words, the output result of the wildcard conversion means, and the words of the house display number word dictionary. If the address display number collating means for recognizing the notation pattern of the address display number by collating using the automaton and the output result of the address display number collating means cannot obtain the address display number candidate, the operator can A display device for entering the address display number while looking at the image of the address area, and a range of possible values for the numerical part of the address display number are displayed. A layered address display number range dictionary and a range determination means for determining whether the numerical part of the address display number input by the operator is within the correct range by referring to the address display number range dictionary, and the determination result. An address reading device comprising: means for issuing an alarm to an operator when it is determined to be out of the range. 6. An address reading method for detecting character information on a postal matter and reading address information including town area information and address display number information, converting an image on the postal matter into an electric signal, and inputting the converted signal. The character information is cut out from the input image and recognized, the recognition candidate character group is output for each cut out character pattern, and the output recognition candidate character group is compared with the town area dictionary storing the town area information. By recognizing the town area, the head of the address display number area is detected based on the recognition result, and the number in the candidate character group corresponding to each character pattern after the head of the address display number area is detected based on the detection result. ,
The result of wildcard conversion is provided with an address display number word dictionary that holds the various notation patterns of the address display numbers represented by the above wildcards as words by converting to a candidate character group that has been replaced with a wildcard that represents an arbitrary number. And the address display number word dictionary are matched using an automaton, the notation pattern of the address display number is recognized, the recognition result is compared with the candidate character group output from the character recognition means, and a wild card is used. The numbers in the replaced candidate character group are restored, and the candidates for the address display number are output.
An address reading method characterized by the above.