JPH02308384A - Address recognizing device - Google Patents

Abstract

PURPOSE:To shorten the processing time and to recognize an address at high speed by defining a word, which starts the recognition of the address, as a keyword out of words to be used for recognition processing. CONSTITUTION:Out of the words to be prepared in a word dictionary 15 in advance, the word to be the starting point of the address recognition, namely, the word in the least-significant hierarchy of the address to be blocked is registered as the keyword. Out of word candidates recognized by a word recognizing means 14, the word registered as the keyword is retrieved and this retrieved word is defined as the starting point of collation. Then, the word candidate is successively collated toward the high-order hierarchy and a series of the words, which can be collated, are recognized as the address by an address recognizing means 16. Thus, the number of the words to be collated is decreased and the address can be recognized at high speed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to an address recognition device that reads and recognizes an address, which is a destination indicated on, for example, a delivery item.

(Prior Art) Conventionally, in an address recognition device, a character string constituting an address on a delivery item as classification information is read by, for example, an optical character reader, and a word dictionary is created from the read character string. It then recognizes the word by referring to it, then recognizes the address formed by combining the recognized words by referring to the address dictionary, and classifies the delivery based on this recognized address. ing.

In such an address recognition device, when recognizing an address, an address dictionary is searched in order from words in the upper layer of the address to words in the lower layer to perform recognition processing. For example, when recognizing the address "Senju Akebono-cho, Adachi-ku, Tokyo" as shown below, "Tokyo" = "Adachi-ku" -
Address recognition was performed by searching the address dictionary in the order of ``Senju'' → ``Akebono-cho'' and finding a match in the address dictionary. However, with this method, it is necessary to check all of the word candidates obtained through word recognition against the address dictionary, and even if it is known that it is impossible to read all the way to the end, the process is checked halfway, which increases the processing time. be.

(Problem to be Solved by the Invention) As mentioned above, it has been found that the method that performs word recognition by sequentially comparing addresses from the upper level requires checking all of the word candidates with the address dictionary, making it impossible to read all the way to the end. To provide an address recognition device capable of shortening the processing time and performing address recognition at high speed, by eliminating the drawback that processing time becomes long because verification is performed halfway even if the processing is completed.

[Structure of the Invention] (Means for Solving the Problems) An address recognition device of the present invention detects character information and recognizes an address by comparing it with information in a dictionary. A character reading means for reading the string, a character recognition means for performing character recognition on the character string read by the character reading means and outputting character candidates, and a character recognition means for outputting character candidates among the words used in the recognition process, which should start address recognition. a word dictionary storing words as keywords; a word recognition means for performing word recognition and outputting word candidates by comparing character candidates output from the character recognition means with the word dictionary; Address recognition means for searching for a word stored as the keyword among the output word candidates, and performing address recognition by sequentially collating the word candidates using the searched word as a starting point. Features.

(Operation) This invention registers in advance as a keyword a word that should be the starting point for address recognition among words prepared in a word dictionary, that is, a word at the lowest level of the address to be classified.
Search for words that are registered as keywords among the word candidates recognized by the word recognition means, use the searched words as a starting point for matching, and sequentially match the word candidates in the upper hierarchy. It is designed to recognize words as addresses. This reduces the number of checks that need to be made and allows for faster address recognition.

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

FIG. 1 schematically shows an address recognition device. That is, reference numeral 11 denotes a scanning section (photoelectric conversion section), which is provided so as to face the delivery P that is conveyed in the direction of arrow a in the figure along a conveyance path (not shown).

This scanning unit 11 obtains a pattern signal by optically scanning the delivery item P and photoelectrically converting it. For example, it uses a light source that illuminates the delivery item P, and receives the reflected light and converts it into an electrical signal. It is composed of a self-scanning CCD image sensor and the like. The reading signal outputted from the scanning section 11 is supplied to the binarization circuit 20 in the address area detection section 12 and the character recognition section 13.

The address area detection unit 12 is a well-known device that detects an area where address information is written from among various information written on the delivery item P based on the shape, position, etc. of the area. For example, the coordinate values of an area detected as an address area by the address area detection unit 12 are output to a selection circuit 21 in the character recognition unit 13.

The character recognition unit 13 includes a binarization circuit 20 that binarizes the read signal from the scanning unit 11, and a binarized signal within the area specified by the coordinate values supplied from the address area detection unit 12, that is, the scanning unit a selection circuit 21 that selects and outputs only the address area read signal from among the read signals from 11;
1 character of the address information corresponding to the address area supplied from
A character detection and extraction circuit 22 detects and extracts each character, a normalization circuit 23 normalizes and samples the detected and extracted character information supplied from the character detection and extraction circuit 22, and a normalization circuit 23 that is supplied from the normalization circuit 23. It is constituted by a recognition circuit 24 that recognizes character information by, for example, a method of matching each character in a character dictionary 25 with a reference pattern, and outputs a plurality of character candidates. The output from the recognition circuit 24 of the character recognition section 13 is supplied to the word recognition section 14.

The word recognition unit 14 recognizes the character candidates supplied from the character recognition unit 13 using the words read out from the word dictionary 15, and outputs a plurality of recognized word candidates. Furthermore, among the words stored in the word dictionary 15, words that must be recognized in the address recognition process, that is, words in the lowest hierarchy that are used as classification information based on the recognition results, are set in advance as keywords. Identification information indicating that it should be used is attached. This word dictionary 15 is a well-known one constructed of, for example, a ROM or the like. The output from the word recognition section 14, that is, the plurality of word candidates, is supplied to the address recognition section 16.

The address recognition unit 16 performs predetermined processing on the word candidates supplied from the word recognition unit 14 to recognize an address, and generates classification designation data for the recognized address, that is, the address recognition unit is provided at a subsequent stage of the address recognition device. The compartment pocket (
(not shown).

Next, the operation of the above configuration will be explained with reference to FIGS. 2 to 4.

For example, when the delivery P shown in FIG.
The delivered item P is optically scanned and photoelectrically converted, and the pattern signal is supplied to the address area detection section 12. Then, the address area detection unit 12 detects an address area in which address information is written from among the pieces of information written in the delivery item P.

Further, the output of the scanning section 11 is supplied to the binarization circuit 20 in the character recognition section 13, and after being binarized, the output is sent to the selection circuit 20.
is output to.

Thereby, the selection circuit 21 selects the address area detection section 12.
The binarization circuit 20 corresponding to the coordinate information supplied from
, that is, only the read signal of the address area is selectively output to the character detection and cutout circuit 22. The character detection and cutout circuit 22 detects and cuts out the characters of the address information corresponding to the supplied address area one by one, and outputs them to the normalization circuit 23.

The normalization circuit 23 normalizes the detected and extracted character information, and outputs it to the recognition circuit 24 as a sampled and binarized character string. In this embodiment, an example in which a string of eight characters is input as shown in FIG. 2(a) will be described.

The recognition circuit 24 performs character recognition on the supplied character information by, for example, a method of matching the characters in the character dictionary 25 with a reference pattern, and outputs the result to the word recognition section 14 . At this time, since it is difficult to determine a unique recognition result for freely handwritten kanji, etc., character candidates that are a plurality of answer candidates are output for one input character. In the illustrated example, four character candidates are output.

The word recognition section 14 recognizes the word from the word dictionary 15 read out from the character candidates supplied from the character recognition section 13, and outputs the word string to the address recognition section 16. That is, word candidates are obtained by extracting a part of the input character string, combining each character candidate therein, and comparing the combination with a word dictionary. For example, if you combine the 4th, 3rd, and 4th answer candidates for each character string in the 2nd to 4th character part, you will get "Sekiya-machi", which matches the word included in the word dictionary, and as shown in Figure 2 ( As shown in b), the word candidate "Sekiya-machi" is obtained. In FIG. 2(b), words marked with "*" are keywords. This keyword is a word that must be recognized in order to perform address recognition processing, and for example, a word at the lowest level when classifying by address is selected as a keyword.

FIG. 3 is an image of the word candidates shown in FIG. 2(b) in memory. In the figure, "Left word (LW) J is
Indicates the word to the left of "word name (NM) J" that is a candidate address. For example, in the case of the address "Adachi, Adachi-ku, Tokyo", the left side of "Sedate" indicates "Sedate-ku". . Note that when there is no left word, that is, in the case of the word in the top hierarchy such as "Tokyo" in the above example, rENDJ is stored as the left word.

Furthermore, the "left end position (L)" and "right end position (R)" indicate the word candidate position within the input character string. For example, if the input character string is "Senju Miyamoto-cho, Seitatsu-ku," then "the left end of Miyamoto-cho j is represented by "6" and the right end is represented by "8."

Next, when such word candidates are obtained from the word recognition unit 14, the address recognition unit 16 starts address recognition processing as shown in the flowchart of FIG.

First, a word that is a keyword is searched among word candidates, and an address candidate containing the word is selected.

In other words, the variable n set in a counter (not shown) is
1" (step S1), the keyword flag (K) column of the nth word candidate is referred to to determine whether the word is a keyword (step S2).

If it is determined that it is not a keyword, the process branches to step Sll, and the variable n is incremented to determine whether or not the word obtained as a word candidate has been examined to the end.
In the illustrated example, it is determined whether the variable n has become "9" (step 512). If it is determined that the last word has not been investigated, the process returns to step S2, and steps S2, S11, and S12 are repeatedly executed until a word candidate defined as a keyword appears.

In the process of this repeated execution, when it is determined in step S2 that a word candidate serving as a keyword has been found, a register LNG (not shown) that stores the length of the matched word, that is, the length of the address candidate is 0'', and the variable n at that time is set as the variable m in a register (not shown) (step S3). In the illustrated example, the word candidate that is the first keyword found is the fourth word candidate "Sekiya town." Therefore, "4", which is the variable n at that time, is set as the variable m.

Next, the left word LW of the m-th word candidate is L W m
, the left end position is set as Lm, and the right end position R is set as Rm (step S4). In other words, the left word LW of the fourth word candidate "Sekiya-machi" is "Seitateku", and the leftmost position is "
2", and "4" is obtained as the right end position R.

Next, the word length of the address candidate is determined. In other words, the value calculated by rRm-Lm+IJ is set in register LNG as "4 to 2.
+1-3J is added (step S5). Then, whether or not the left word of the word candidate is rENDJ, that is,
It is checked whether the word candidate is a word in the highest hierarchy (step S6). Then, if it is determined that it is not "END", the word names N and M are the left words L W m .

In other words, it is a "established ward" and the right end position R is "Lm-1
'', that is, the word r2-1-IJ is searched from the word candidates (step S7). If it is determined in step S8 that there is a corresponding word candidate, the number of the word candidate is set as a new variable m (step S9). On the other hand, if it is determined that there is no matching address, the variable n and the length LNG of the matched address candidate are left in a memory (not shown) as an address candidate (step 510). In the illustrated example, there is no corresponding item, so the variable n "4" and L-NG are changed to "
3" in memory. In other words, in the above process, '
This means that only "Sekiya Town" could be matched as a potential address.

Next, increment the variable n (step 511),
In the same manner as described above, steps S2, Sll, and 512 are repeatedly executed until a word candidate defined as the next keyword appears. When a word candidate serving as a keyword is found in step S2 in the process of this repeated execution, the register LNG is cleared to "0" and the variable n at that time is stored as a variable m in a register not shown (step S3). . In the illustrated example, the word candidate that is the second keyword to be found is the sixth word candidate, ``front piece''. Therefore, the variable n at that time is "6"
is set as the variable m. The subsequent processing is the same as in the case of the keyword "Sekiya-machi", so the explanation will be omitted, but in this case, only "front piece" can be collated as an address candidate.

Next, increment the variable n (step 511),
In the same manner as described above, steps S2, Sll, and S12 are repeatedly executed until a word candidate defined as the next keyword appears.

In the process of this repeated execution, step S
When a word candidate serving as a keyword is found in step 2, the register LNG is cleared to "0" and the variable n at that time is set as a variable m in a register (not shown) (step S3). In the illustrated example, the third word candidate found is the eighth word candidate "Miyamoto-cho."

Therefore, "8", which is the variable n at that time, is set as the variable m. Next, the left word L of the m-th word candidate
W is L W m, left end position is Lm, right end position R is R
m (step S4). That is, "Senju" is obtained as the left word LW of "Miyamotocho", which is the eighth word candidate, "6" is obtained as the left end position, and "8" is obtained as the right end position R.

Next, the length of the address candidate is determined. That is, register L
The value rRm-Lm+IJ is calculated for NG r8-6+1-
3J is added (step S5).

Then, it is checked whether the left word of the word candidate is rENDJ, that is, whether the word candidate is a word in the highest hierarchy (step S6), and if it is determined that it is not rENDJ, the word name Search the word candidates for a word in which NM is the left word L W m, that is, "Senju J, and the right end position R is rLm-IJ, that is, r6-1-5" (
Step S7). If it is determined in step S8 that there is a corresponding word candidate, the number of that word candidate is set as a new variable m (step S9). In this case, since there is a word that matches the fifth word candidate, the process advances to step S9, sets the number of the matched word candidate, that is, "5", as a new variable m, and returns to step S4 again. In the second round of processing, the left word LW of "Senju", which is the fifth word candidate, is
, "established ward" is obtained, "4" is obtained as the left end position, and "5" is obtained as the right end position R.

Next, the length of the address candidate is determined. That is, register L
NG: "Rm-Lm+IJ calculated value r5-4+1-
2J is added (step S5).

As a result, the contents of register LNG become "5".

Then, it is checked whether the left word of the word candidate is rENDJ, that is, whether the word candidate is a word in the highest hierarchy (step S6), and if it is determined that it is not rENDJ, the word name NM is the left word L W m, that is, "established ward", and the right end position R is "Lm-IJ,"
In other words, the word r4-1-3j is searched from among the word candidates (step S7). In this case, since there is a matching word candidate, the process proceeds to step S9, sets the number of the matching word candidate, that is, "2", as a new variable m, and returns to step S4 again. Then, in the third round of processing, the left word LW of the second word candidate "Established Ward" is "Tokyo", the left end position is "1", and the right end position R is "Tokyo".
3”.

Next, the length of the address candidate is determined. That is, register L
The value rRm-Lm+lJ is calculated for NG r3-1+1-
34 is added (step S5).

As a result, the contents of register LNG become "8".

Then, it is checked whether the left word of the word candidate is rENDJ, that is, whether the word candidate is a word in the highest hierarchy (step S6), and if it is determined that it is not rENDJ, the word name NM is the left word L W m , that is, "Tokyo", and the right end position R is "Lm-IJ
, that is, a word rl-1-OJ is searched from among the word candidates (step S7). In this case, since there is no matching item, the process branches to step S10, and the values of "8" for variable n and "8" for LNG are left as address candidates. That is,
This means that "Senju Miyamotocho, established ward" has been verified.

Furthermore, when processing is performed using "Motomachi" as a keyword, only "Motomachi" can be collated in the same way as above. Then, in step S12, it is determined that the process has ended, and the series of address recognition processes ends.

Address candidates recognized in this way are shown in Figure 2 (
As shown in c), several are obtained. In this case, for example, by referring to the contents of register LNG,
The one with the largest number of matched characters (circled) is output as the recognition result. That is, in the illustrated example, the recognition result is "Senju Miyamoto-cho, Seitate-ku." Then, the classification designation data corresponding to the recognized address, that is, the data indicating the classification pocket in the subsequent classification section, is output to the subsequent classification section. As a result, the delivery items P are sorted into sorting pockets corresponding to each address.

As described above, among the words prepared in the word dictionary 1t15, words that should be the starting point for address recognition, that is, words at the lowest level of the address to be classified, are registered as keywords, and the word recognition unit 14 Search for words that have been registered as keywords among the word candidates recognized by Since the word is recognized as an address,
In the conventional method, it is necessary to match all nine word candidates, but according to this embodiment, only four matches are required, which is less than half, and the number of word candidates to be matched is reduced, allowing for faster address recognition. ing. In addition, since the matching process is performed sequentially starting from the word in the lowest hierarchy, there is no need to perform matching that is known to be unreadable to the end.
This enables efficient verification processing.

[Effects of the Invention] As detailed above, according to the present invention, it is possible to provide an address recognition device that can shorten processing time and perform address recognition at high speed.

[Brief explanation of the drawing]

The drawings show one embodiment of the present invention, and FIG. 1 is a diagram schematically showing the overall configuration, FIG. 2 is a diagram for explaining the processing up to obtaining address candidates, and FIG. 3 is a diagram showing the word FIG. 4 is an explanatory diagram showing a candidate memory image, and FIG. 4 is a flowchart for explaining the operation. P...Delivery item, 11...Scanning section (character reading means),
12... Address area detection section, 13... Character recognition section (character recognition means), 14... Word recognition section (word recognition means)
, 15... Word dictionary, 16... Address recognition unit (address recognition means).

Claims (1)

[Claims] An address recognition device that recognizes an address by detecting character information and comparing it with information in a dictionary, comprising: a character reading means for reading a character string constituting an address; and a literature reading means read by the literature reading means. a character recognition means that performs character recognition on a character string and outputs character candidates; a word dictionary that stores words as keywords for starting address recognition among words used in recognition processing; and the character recognition means. word recognition means for performing word recognition and outputting word candidates by comparing character candidates output from the word dictionary with the word dictionary; 1. An address recognition device comprising: address recognition means for performing address recognition by searching and sequentially collating the word candidates using the searched word as a starting point.