JPH08171615A - Address reader - Google Patents

Abstract

PURPOSE: To exactly judge printing and to enable user to exactly and efficiently read it in the case of a printing character string by deciding the character string of the final address read result based on a place name part candidate buffer and an address part candidate buffer. CONSTITUTION: When a place name part candidate like printing character exists, the combination of the character candidates of an address part suited to attributes in a printing attribute buffer 9 and suited to the conditions of the address part in a place name table memory 5 is extracted from a recognized result buffer 4 by a printing character priority address part retrieving means 11. When no place name part candidate like printing character exists, the combination of the character candidates of the address part in the place name table memory 5 is extracted from the recognized result buffer 4 by a general address part retrieving means 10. An address part candidate buffer 12 stores the result of the printing priority address part retrieving means 11 or the result of the general address part retrieving means 10. Based on a place name part candidate buffer 7 and an address part candidate buffer 12, a final deciding means 12 decides the character string of the final address read result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an address reading device for reading an address character string consisting of a place name part and an address part from input image data. As examples of the address character string, "4-1-1 Miyazaki, Miyazaki-ku, Kawasaki-shi, Kanagawa" and "5-1, 7 Shiba, Minato-ku" are mentioned, and the place name part in this specification is "Kawasaki, Kanagawa-ken.""Miyazaki,Ichimae-ku" and "Shiba, Minato-ku" are referred to, and the address section is "4-1-1" and "5-chome 7-1".

[0002]

2. Description of the Related Art FIG. 3 is a block diagram showing the configuration of a conventional address reading device.

The image input means 1 inputs image data in which an address character string consisting of a place name part and an address part is described. The image buffer 2 stores the input image data. The individual recognition means 3 cuts out a segment corresponding to one character from the image data, and individually recognizes each segment. The recognition result buffer 4 stores the result of the individual recognition means 3. The place name table memory 5 stores a list of place names to be read and conditions of an address part corresponding to each place name. The place name part search means 6 searches the place name table memory 5 and extracts place name part candidates corresponding to the combination of character candidates in the recognition result buffer 4. The place name part candidate buffer 7 stores the extracted place name part candidates. The general address part searching means 10 extracts, from the recognition result buffer 4, a combination of address part character candidates that match the conditions of the address part in the place name table memory 5. The address part candidate buffer 12 stores the result of the general address part search means 10. The final determination means 13 determines the final character string of the address read result based on the place name part candidate buffer 7 and the address part candidate buffer 12.

FIG. 4 is a block diagram showing another structure of a conventional address reading device.

The image input means 1 inputs image data in which an address character string consisting of a place name part and an address part is described. The image buffer 2 stores the input image data. The individual recognition means 3 cuts out a segment corresponding to one character from the image data, and individually recognizes each segment. The recognition result buffer 4 stores the result of the individual recognition means 3. The place name table memory 5 stores a list of place names to be read and conditions of an address part corresponding to each place name. The place name part search means 6 searches the place name table memory 5 and extracts place name part candidates corresponding to the combination of character candidates in the recognition result buffer 4. The place name part candidate buffer 7 stores the extracted place name part candidates. The general address part individual recognition means 10 cuts out a segment corresponding to one character of the address part from the image data, and performs individual character recognition for each segment. The address part recognition result buffer 16 stores the result of the general address part individual recognition means 14. The general address part searching means 10 extracts, from the recognition result buffer 4, a combination of address part character candidates that match the conditions of the address part in the place name table memory 5. The address part candidate buffer 12 stores the result of the general address part search means 10. The final determination means 13 includes a place name section candidate buffer 7 and an address section candidate buffer 1.
Based on 2, the final character string of the address reading result is determined.

In the conventional configuration of FIG. 3 and the conventional configuration of FIG. 4,
There is a difference in the character cutting and character recognition of the address part. In the configuration of FIG. 3, the address part has the same character segmentation / character recognition result (recognition result buffer 3) as the place name part, but in the configuration of FIG. 4, the address part is separate from the place name part. Character extraction and character recognition are executed again, and the result (address part recognition result buffer 16) is processed.

FIG. 5 shows an example of the contents of the place name table memory 5. Based on the address structure of Shinagawa-ku, Tokyo. The place name table 56 has a list 56 of place names and a condition 57 of an address part.

The list 56 of place names includes prefecture names 50,
Place names such as city name 51 and town name 52 are also registered together with their hierarchical relationships. Depending on the area where the address is read, it may include a county name, an abbreviated name, or a character name.

The condition 57 of the address part describes the range of values permitted as each chome, number, and number for each town name 52. For example, in the case of "EBARA" in FIG. 5, the number of chomes is from 1 to 7, and numbers 1 to 50 and numbers 1 to 9 are allowed.

In the example of FIG. 5, the address part is a claw, a number,
Although it is assumed to be composed of a number, depending on the town name 52, there is also a type of address / number. In such cases,
Add type distinction (chome / ban / go type or street / go type), or 1 of chome / ban / go
The method of not using the value of the item (for example, setting the value to 0) is adopted.

The place name part searching means 6 is provided in the recognition result buffer 4
The combination of the character candidates in the place name (a combination of character candidates for adjacent segments) is collated with the character string of the place name in the place name list 56 of the place name table memory 5 as described above, and the place name hierarchical relationship is obtained. The place name sequence that does not cause contradiction is output to the place name part candidate buffer 7 as a place name part candidate.

The general address part searching means 10 combines the character candidates in the recognition result buffer 4 (in the case of the configuration of FIG. 3) or the address part recognition result buffer 16 (in the case of the configuration of FIG. 4) to combine the characters of the address part. Form a row. At that time, by referring to the place name part candidates in the place name part candidate buffer 7, some constraint conditions are added to the formation of the character string of the address part.

The first constraint condition is to know the end position of the place name part candidate and control to form the address part from the position immediately after that.

The second constraint condition is that the condition of the address part corresponding to the candidate place name part is known from the place name table memory 5, and the number strings of chome, number and number are combined so as to satisfy the condition.

Further, although it does not depend on the place name part candidate,
Even if the arrangement or combination of the separators that delimit the number strings such as "chome", "ban", "go", "street number", "no", "no", "-" is not unnatural, it is used as a constraint condition in the character string formation of the address part. It A character string satisfying such a restriction condition of the address part is the address part candidate buffer 1 as the address part candidate.
2 is output.

The conventional address reading device and the method of realizing each component described above are described, for example, in "Automatic reading / sorting machine for mail address for Japan Post" (Ishikawa et al., NEC Technical Report, Vol. 44, No. 3). , 1991), "Automatic reading / sorting machine for mailing name TR-17" (Torimoto et al., Toshiba review,
Vol. 45, No. 2, 1990), JP-A-5-32489
Japanese Unexamined Patent Application Publication No. 9- "Postal item address recognition device", JP-A-3-18
No. 9780, "Address Recognition Device", JP-A-6-1243.
No. 66, "Address Reader", Japanese Patent Laid-Open No. 5-169033
It is described in "Address Reader" or the like.

[0017]

It is relatively easy to cut out a character string of an address character string entered in a character frame provided in advance in a form or the like. But,
There are many things such as mail that do not have a character frame,
From the standpoint of address entry side, it is more convenient to be able to enter the address without being restricted by letter boxes. Therefore, it is necessary to accurately perform character segmentation on an address character string that is freely written without providing a character frame.

On the other hand, the ratio of printing type is increasing due to the spread of word processors and the progress of computerization of various operations. In the case of a printed character string, it is easier to cut out a character than a handwritten character string because of the uniformity of character size and pitch. A conventional method for character extraction is described in Japanese Patent Laid-Open No. 5-166099 “Character cutting / recognition method and apparatus”, Japanese Patent Laid-Open No. 5-128307 “Character recognition apparatus” and the like.

However, it is general that the handwriting and the printing are processed in a mixed state, and there are few cases in which the processing can be switched by designating handwriting or printing in advance. .

On the other hand, in the conventional address reading device as described above, handwriting and printing are processed without distinction. Therefore, with respect to the printed character string, the character segmentation is performed assuming various character sizes and pitches similarly to the handwritten character string.

In such a case, the following problem occurs in the character string of the address part of the address.

The first problem is the vertical writing of "two" and "three".
For example, a lot of possibilities occur, and it is impossible to uniquely determine the reading result of the address part due to the competition. For example, if "3" is written vertically, there is a possibility that "11", "12", "21", "3", etc. will occur. If it is known that it is an Inprint character string, using the conditions of character size and pitch, reject candidates such as "11-11", "12", "21", and uniquely set to "3". I can decide.

The second problem is that in print characters (especially word processing characters), the horizontal bars above and below "two" and "three" are designed to be positioned at the upper and lower ends of the character size. In many cases, if the character segmentation is performed simply by paying attention to only the space between line segments, it is easy to combine the characters with the preceding and succeeding characters. FIG. 6 shows an example of erroneous clipping due to this problem.
Figure 6 for the stamp character string "Ebara 2-chome, Shinagawa-ku"
(A) is a correct cutting result. On the other hand, in Fig. 6 (b), the horizontal bar above "2" is included in the immediately preceding "original" segment, and in the example of Fig. 6 (c), the horizontal bar below "2" is included. The stick is included in the "Ding" segment immediately after. As a result, in the case of (b) and (c), the result of reading the address will be an erroneous result of "Ebara 1-chome, Shinagawa-ku". It is very likely that "hara" and vertically written "ichi" are collectively recognized as "cho."

With respect to the problem when the handwriting and the printing are mixed, the following two countermeasures can be considered in the configuration of the conventional address reading device as shown in FIGS. 3 and 4.

The first measure is that the individual recognition means 3
This is a method of determining whether it is handwriting or printing and reflecting it in the character segmentation.

The second measure is to operate the individual recognizing means 3, the place name part retrieving means 6, the general address part individual recognizing means 14, the general address part retrieving means 10, etc. without worrying about handwriting / printing, and finally. This is a method of devising a method of selecting a place name part candidate or an address part candidate at the stage of the determination means 13.

However, these two measures have the following drawbacks.

The drawback of the first measure is that there is a limit to the accurate judgment of handwriting or printing at the stage of the individual recognition means 3. If it is judged at the stage of the individual recognition means 3 whether it is handwriting or printing, if it is possible to cut out the characters with equal size and pitch, it is judged as printing, and if not, it is handwritten. It will be decided.
However, if only the physical conditions such as equal size and equal pitch are used for judgment, the handwritten character string will be forcibly cut into small segments of equal size and equal pitch, or even multiple characters may be cut out from a printed character string. Occasionally, it may cut out at the wrong point. FIG.
While (a) is the correct cutting method, (b)
(C) is an example in which erroneous cutting is performed even with the same size and the same pitch.

The disadvantage of the second measure is that even in the case of a print character string, a lot of place name part candidates and address part candidates are generated, and it takes time to process. When a large number of candidates for character extraction occur, the subsequent processing targeting the result generally increases the processing amount in a combinatorial manner. In the case of reading the address of the mail, the time limit is given to one address read, and if the number of combinations increases, the processing cannot be performed within the time limit.

In view of the above, the present invention eliminates the drawbacks of the conventional address reading device as described above, and makes accurate printing judgment even if the designation of a handwritten character string or a stamp character string is not given from the outside. If the character string is a print character string, it is an object to realize accurate and efficient reading by utilizing its characteristics (equal size, equal pitch, etc.).

[0031]

According to a first aspect of the present invention, in an address reading device for reading an address character string consisting of a place name part and an address part from input image data, a segment corresponding to one character in the image data. An individual recognition unit that cuts out and recognizes individual characters for each segment, a recognition result buffer that stores the result of the individual recognition unit, a list of place names to be read, and a condition of an address part corresponding to the place name are registered. And the place name table memory
A place name part search means for searching the place name table memory to extract a place name part candidate corresponding to a combination of character candidates in the recognition result buffer, a place name part candidate buffer for storing the place name part candidate, and the place name part candidate If there is a place name part candidate that seems to be in print by determining the print impression of the character string that constitutes, the print activity determination means that prioritizes the place name part candidate that seems to be the print place among the place name part candidates, Stamp attribute buffer that stores the attributes of the likely place name part candidates, and the conditions of the address part in the place name table memory and the attributes in the stamp attribute buffer from the recognition result buffer when the likely place name part candidates exist Inquiry priority address part search means for extracting a combination of character candidates of the address part that conforms to,
A general address part searching means for extracting a combination of address part character candidates that meet the conditions of the address part in the place name table memory from the recognition result buffer when there is no place name part candidate that seems to be stamped; An address part candidate buffer for storing the result of the live priority address part search means or the result of the general address part search means, and a character string of the final address read result based on the place name part candidate buffer and the address part candidate buffer An address reading device comprising a final determination means for determining.

A second invention is an address reading device for reading an address character string consisting of a place name part and an address part from input image data, and a segment corresponding to one character is cut out from the image data and each segment is segmented. An individual recognition means for individually recognizing characters, a recognition result buffer for storing the result of the individual recognition means, a list of place names to be read, and a place name table memory in which conditions of an address part corresponding to the place names are registered, A place name part search means for searching the place name table memory to extract a place name part candidate corresponding to a combination of character candidates in the recognition result buffer, a place name part candidate buffer for storing the place name part candidate, and the place name part candidate If there is a place name part candidate that is likely to be printed by judging the printing impression of the character string forming the A seal activation determining unit that gives priority to a lively place name part candidate, a print job attribute buffer that stores an attribute of the printable place name part candidate, and a stamp from the image data when the printable place name part candidate exists. Indicating priority address individual recognizing means for cutting out a segment corresponding to one character of the address so as to match the attribute in the live attribute buffer and recognizing individual characters for each segment, and the place name part candidate likely to be in printing If there is not, a segment corresponding to one character of the address part is cut out from the image data and individual character recognition is performed for each segment, and the printing priority address part individual recognition means. Address result recognition buffer storing the result of the above or the result of the general address individual recognition means, and the address table from the address recognition result buffer General address part search means for extracting a combination of address part character candidates that match the conditions of the address part in memory, an address part candidate buffer for storing the result of the general address part search means, the place name part candidate buffer, and the An address reading device, comprising: final determination means for determining a character string of a final address reading result based on an address part candidate buffer.

[0033]

In the present invention, the place name part candidate is once extracted, and then the impression of the place name part candidate is judged. And
If there is an Inzai-like candidate for a place name, it will be given priority. Furthermore, when a place name club candidate that seems to be Inzai was obtained,
The address part is processed by using the attributes such as the character size and pitch regarding the place name part candidate.

In order to narrow down the candidates for character cutting out of the address part by using the information on the impression of the place name part candidates, handwriting / writing
The above-mentioned two points which have been pointed out as the conventional defects when the printing is not distinguished are removed. In other words, the print letter "three"
Will no longer be divided into "11", "12", and "21", and the cutout of the address part that does not match the character size and pitch of the place name part as shown in Fig. 6 (c) will be rejected. it can.

Further, since the place name portion candidates are obtained, the ones that are likely to be printed are prioritized, so that the problem of the first conventional measure can be eliminated. That is, even with the same size and the same pitch, a wrong character cutout as shown in (b) and (c) of FIG. 7 can be eliminated because a place name part candidate cannot be formed as a combination of character candidates. . Figure 6
Even if the place name part candidate is readable for the cut-out of the place name part as shown in (b), the character size / pitch is disturbed in the "Haraichi" part, so as shown in FIG. 6 (a). If there is a place name part candidate of equal size and pitch, (b)
Will be excluded and priority will be given to (a).

Furthermore, since the formation of extra candidates for the address part is suppressed based on the sign activity at the stage of obtaining the candidates for the place name part,
The poor processing efficiency, which was the drawback of the conventional second measure, can be eliminated.

[0037]

1 is a block diagram showing the configuration of an embodiment of the first invention.

The image input means 1 inputs image data in which an address character string including a place name part and an address part is described. The image buffer 2 stores the input image data. The individual recognition means 3 cuts out a segment corresponding to one character from the image data, and individually recognizes each segment. The recognition result buffer 4 stores the result of the individual recognition means 3. The place name table memory 5 stores a list of place names to be read and conditions of an address part corresponding to each place name. The place name part search means 6 searches the place name table memory 5 and extracts place name part candidates corresponding to the combination of character candidates in the recognition result buffer 4. The place name part candidate buffer 7 stores the extracted place name part candidates. The printing activity determination means 8 determines the printing activity of the character string that constitutes the printing site name candidate in the printing site name candidate buffer 7, and if there is a printing site name candidate that seems to be printing activity, it is one of the printing site name candidates. The contents of the place name part candidate buffer 7 are rewritten so that the place name part candidates that are likely to be printed are prioritized. The stamp activity attribute buffer 8 stores the attribute of the place name portion candidate which seems to be the stamp activity detected by the stamp activity determination means 7. When there is a place name part candidate that seems to be a seal, the print priority address part search means 11 matches the attribute in the print attribute buffer 9 from the recognition result buffer 4 and stores the address part in the place name table memory 5. A combination of character candidates of the address part that also meets the conditions is extracted. The general address part searching means 10 extracts a combination of address part character candidates that meet the conditions of the address part in the place name table memory 5 from the recognition result buffer 4 when there is no place name part candidate that seems to be printing. . The address part candidate buffer 12 stores the result of the printing priority address part searching means 11 or the result of the general address part searching means 10. The final determination means 13 determines the final character string of the address read result based on the place name part candidate buffer 7 and the address part candidate buffer 12.

Among these components, the image input means 1, the image buffer 2, the individual recognition means 3, the recognition result buffer 4, the place name table memory 5, the place name part searching means 6, the place name part candidate buffer 7, and the general address part search. Means 10,
Each of the address part candidate buffer 12 and the final determination means 13 is
The components are the same as those of the conventional address reading device shown in FIG. The three new components are the stamp activation determination means 8, the stamp attribute buffer 9, and the stamp priority address part searching means 11.

These three components will be described.

First, the print activation determining means 8 determines the print activity of the character strings forming the place name part candidates in the place name part candidate buffer 7. This impression of impression is
For example, the following points of the characters (segments) forming the place name department candidates may be focused on. If it is likely to be an impression, it is determined to be YES for each of these viewpoints. (A) Are the center points of the segments aligned in the character string direction? (B) Are the intervals (pitch) between the center points of the segments uniform? (C) Is the segment size stable? (D) Is the segment width stable?

FIG. 8A shows an example in which it is determined that the printing is likely to be made, and the meanings of the center point 41 of the segment 40, the pitch 42, the size 43, the width 44, etc. are also shown. Since the coordinate information of each segment is obtained at the stage of character extraction, the center point, pitch, size, width, etc. can be easily calculated.

In (A), it is sufficient to calculate whether or not the difference between the maximum and minimum coordinate values in the direction perpendicular to the character string direction is within a fixed value for the center point of each character. FIG.
(B) is an example in which the difference between the maximum and the minimum is large, and it cannot be determined that the printing is likely to be performed.

In (B), the difference between the coordinates of the center point in the character string direction is sequentially calculated, the pitch is calculated, and it is determined whether the difference between the maximum and the minimum of the pitch falls within a fixed value. In FIG. 8C, the variation in pitch is large,
This is an example in the case where it cannot be determined that it seems to be Inzai.

Similarly, in (C) and (D), the size and width are calculated from the difference between the coordinates of the vertices of the segment, and the variation is checked. However, in the case of size and width, there are characters such as "one" and "1" that have smaller values, so it is necessary to handle them exceptionally.

Further, as a criterion for determining the impression of printing, it is not always necessary to satisfy all the points (A), (B), (C) and (D). The determination items may be narrowed down from the beginning, or some of the determination items may be satisfied. In addition, the above (A) (B)
Conditions other than (C) and (D) may be introduced.

Further, when there is a place name part candidate that seems to be Inzai, the mark activation determining means 8 changes the contents of the place name part candidate buffer 7 so as to give priority to the place name part candidate that seems to be Inkaku among the place name part candidates. rewrite. This rewriting method is, for example,
The following methods are possible. (1) Only the place name part candidates that are likely to be Inzai are left, and other place name part candidates are deleted. (2) For each place name division candidate, an area is prepared in the place name division candidate buffer 7 so that the priority can be stored,
When there is a place name part candidate that seems to be Inzai, a value is set so that the priority of the candidate is relatively higher than the priority of other candidates.

For example, the place name part of FIG. 6A and the place name part of FIG.
It is assumed that there are two types of candidates such as the place name part in (b). At this time, the seal activation determination means 8 determines that (a) is likely to be active, but (b) is not likely to be active, and if the rewriting method as described in (1) above is adopted, the place name portion candidate buffer is used. 7 to (b) candidates will be deleted.

In the stamp activation attribute buffer 9, among the attributes obtained by the stamp activation determining means 8, the stamp priority address searching means 11 (or the stamp priority address individual recognizing means 15 of the second invention).
Stores the attributes used in. In the present embodiment, for example, the coordinates of the center points of the last characters of the place name part candidates determined to be printing impressions, the intervals (pitch) of the center points, and the segment size are stored as attributes.

Similar to the general address part searching means 10, the printing priority address part searching means 11 extracts from the recognition result buffer 4 a combination of character candidates of the address parts which meet the conditions of the address parts. At that time, the general address part searching means 10 considers only the compatibility with the condition of the address part in the place name table memory 5, whereas the printing priority address part searching part 11
Further, the compatibility with the attributes in the print attribute buffer 9 is also taken into consideration. Therefore, the printing priority address part searching means 11 can be realized by adding the following modification to the general address part searching means 10, for example.

First, as preprocessing, the recognition result buffer 4
For each segment, the compatibility with the attribute in the print attribute buffer 9 is determined. This suitability determination method is performed, for example, by calculating the coordinates of the center point of the last character of the place name part candidate that seems to be In printing (the estimated center point coordinates) and the coordinates of the center point of each segment. , You can make the comparison. As one example, the distance from the center point coordinate of the segment to the closest estimated center point coordinate can be regarded as the compatibility with the printing attribute. In this preprocessing, such a goodness of fit is calculated and given to each segment. Alternatively, a segment whose adaptability does not reach the threshold value may be deleted from the recognition result buffer 4.

After performing such preprocessing, processing similar to that of the general address part searching means 10 is performed. If the general address part searching means 10 defines the suitability of the character candidates of the address part as something like a cost value, the cost value reflects the compatibility with the printing attribute. Just do it. If not, the character candidates of the segment whose adaptability does not reach the threshold value may be ignored (deleted in advance), and the process of the general address part searching means 10 may be executed.

As a result, for example, as shown in FIGS.
For a place name part candidate such as "2" in Fig. 6 (a)
The segments corresponding to "Ding" and "eyes" match the printing attributes, but the segments corresponding to "One" and "One Ding" in FIG. 6C do not match the printing attributes. Figure 6
It is possible to avoid erroneous address part candidate extraction as in (c).

The seal activation priority address searching means 11 and the general address searching means 10 are switched and executed by the seal activation determining means 8. And the address part candidate buffer 12
Stores the result of one of the stamp activation priority address portion searching means 11 and the general address portion searching means 10 which is selected by the stamp activation determining means 8.

FIG. 2 is a block diagram showing the configuration of an embodiment of the second invention.

The image input means 1 inputs image data in which an address character string including a place name part and an address part is described. The image buffer 2 stores the input image data. The individual recognition means 3 cuts out a segment corresponding to one character from the image data, and recognizes another character for each segment. The recognition result buffer 4 stores the result of the individual recognition means 3. The place name table memory 5 stores a list of place names to be read and conditions of an address part corresponding to each place name. Place name part search means 6
Searches the place name table memory 5 and extracts place name part candidates corresponding to the combination of character candidates in the recognition result buffer 4. The place name part candidate buffer 7 stores the extracted place name part candidates. The printing activity determination means 8 determines the printing activity of the character string that constitutes the printing site name candidate in the printing site name candidate buffer 7, and if there is a printing site name candidate that seems to be printing activity, it is one of the printing site name candidates. The contents of the place name part candidate buffer 7 are rewritten so that the place name part candidates that are likely to be printed are prioritized. The stamp activity attribute buffer 8 stores the attribute of the place name portion candidate which seems to be the stamp activity detected by the stamp activity determination means 7. When there is a place name portion candidate that seems to be a seal stamp, the stamp priority priority address portion individual recognizing means 15 uses the address portion 1 to match the attribute in the seal stamp attribute buffer 9 from the image data in the image data buffer 2. Cut out the segment corresponding to the character,
Individual character recognition is performed for each segment. The general address part individual recognition means 10 cuts out a segment corresponding to one character of the address part from the image data in the image data buffer 2 when there is no place name part candidate that is likely to be printed, and individually for each segment. Character recognition. The address part recognition result buffer 16 stores the result of the printing priority address part individual recognition means 15 or the result of the general address part individual recognition means 14. The general address part searching means 10 extracts, from the recognition result buffer 4, a combination of address part character candidates that match the conditions of the address part in the place name table memory 5. The address part candidate buffer 12 stores the result of the general address part search means 10. The final determination means 13 determines the final character string of the address read result based on the place name part candidate buffer 7 and the address part candidate buffer 12.

Of these constituent elements, the image input means 1, the image buffer 2, the individual recognition means 3, the recognition result buffer 4, the place name table memory 5, the place name part search means 6, the place name part candidate buffer 7, the general address part individual Recognition means 1
4, the address part recognition result buffer 16, the general address part search means 10, the address part candidate buffer 12, and the final determination part 13 are the same constituent elements as the conventional address reading device of FIG. The three new components are the stamp activation determination means 8, the stamp attribute buffer 9, and the stamp priority address part individual recognition means 15. Among them, the print activation judging means 8 and the print attribute buffer 9 are common to the configuration of the embodiment of the first invention (FIG. 1) and have been already described.

The printing priority address part individual recognition means 15 will be described.

The print priority address individual recognizing means 15 is similar to the general address individual recognizing means 14 in the image buffer 2.
A segment corresponding to one character in the address section is cut out and the character is individually recognized. In the cutout processing, the stamp impression priority address part individual recognition means 15 is different in that the compatibility with the attribute in the stamp impression attribute buffer 9 is also taken into consideration. Therefore, the printing priority address individual search unit 15 can be realized by adding the following modification to the character segmentation process of the general address individual recognition unit 14, for example.

From the attributes in the printing attribute buffer 9, for example, the coordinates and pitch of the center point of the last character of the place name portion candidate that seems to be printing can be known. Therefore, a coordinate (estimated center point coordinate) obtained by shifting the coordinates of the center point of the last character of the place name portion candidate according to the pitch is calculated. And the printing attribute buffer 9
If you know the size of the character (segment) from the attributes inside,
A segment may be cut out with a fixed size around the estimated center point coordinates.

The printing activation priority address individual recognizing means 15 and the general address individual recognizing means 14 are switched and executed by the printing activation determining means 8. Then, the address part recognition result buffer 16 stores the result of one of the printing priority address part individual recognition means 15 and the general address part individual recognition means 14 which is selected by the printing activity determination part 8.

[0062]

As described above, according to the present invention, since the printing activity is determined after the place name portion candidates are obtained, the printing activity can be accurately determined. Then, when it is determined that it seems to be an impression, it is used for the processing of the address part, so the possibility of character cutting out of the address part can be appropriately narrowed down,
The reading of the address part of the print character string is accurate and efficient. That is, according to the present invention, even if the designation of the handwritten character string or the stamp character string is not given from the outside, the stamp mark is accurately determined, and if the stamp character string is detected, the characteristics (equal size)・ We can provide an accurate and efficient address reading device that reads by making the best use of equal pitches.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the first invention.

FIG. 2 is a block diagram showing the configuration of an embodiment of the second invention.

FIG. 3 is a block diagram showing a configuration of a conventional address reading device.

FIG. 4 is a block diagram showing a configuration of a conventional address reading device.

5 is an example of contents of a place name table memory 5. FIG.

FIG. 6 is an example of a character cutout result for a printed character string.

FIG. 7 is an example of a character cutout result for a printed character string.

FIG. 8 is a diagram for explaining a determination example by a seal activation determination means 8.

[Explanation of symbols]

 1 image input means 2 image buffer 3 individual recognition means 4 recognition result buffer 5 place name table memory 6 place name part search means 7 place name part candidate buffer 8 stamp activation judgment means 9 stamp attribute buffer 10 general address search means 11 stamp priority address Part search means 12 Address part candidate buffer 13 Final decision means 14 General address part individual recognition means 15 Printing priority address part individual recognition means 16 Address part recognition result buffer 40 Segment 41 Segment center point 42 Segment pitch 43 Segment size 44 Width of segment 50 Prefecture name 51 City name 52 Town name 53 Chome 54 No. 55 No. 56 List of place names 57 Condition of address part

Claims (2)

[Claims] 1. In an address reading device for reading an address character string consisting of a place name part and an address part from input image data, a segment corresponding to one character is cut out from the image data, and an individual character is segmented for each segment. Individual recognition means for recognition, a recognition result buffer for storing the result of the individual recognition means, a place name table memory in which a list of place names to be read and conditions of an address part corresponding to the place name are registered, and the place name table memory A place name part searching means for extracting a place name part candidate corresponding to a combination of character candidates in the recognition result buffer, a place name part candidate buffer storing the place name part candidate, and a character forming the place name part candidate. If there is a place name club candidate that seems to be an In print after judging the print impression of the row, the place name that seems to be the Inzai among the place name club candidates. A priority indicia activation determining means candidate,
An impression stamp buffer that stores the attributes of the place name portion candidate that seems to be printing, and an attribute in the printing attribute buffer and the address in the place name table memory from the recognition result buffer when the candidate printing place name portion exists Print priority address part search means for extracting a combination of character candidates of the address part that match the conditions of the copy part, and an address in the place name table memory from the recognition result buffer when the place name part candidate that seems to be the print position does not exist A general address part search means for extracting a combination of address part character candidates that match the conditions of the copy part, and an address part candidate buffer for storing the result of the printing priority address part search means or the result of the general address part search means , And a final determination means for determining a final character string of the address reading result based on the address part candidate buffer and the address part candidate buffer. Address reading device, characterized in that. 2. In an address reading device for reading an address character string consisting of a place name part and an address part from input image data, a segment corresponding to one character is cut out from the image data, and an individual character for each segment is cut out. Individual recognition means for recognition, a recognition result buffer for storing the result of the individual recognition means, a place name table memory in which a list of place names to be read and conditions of an address part corresponding to the place name are registered, and the place name table memory A place name part searching means for extracting a place name part candidate corresponding to a combination of character candidates in the recognition result buffer, a place name part candidate buffer storing the place name part candidate, and a character forming the place name part candidate. If there is a place name club candidate that seems to be an In print after judging the print impression of the row, the place name that seems to be the Inzai among the place name club candidates. A priority indicia activation determining means candidate,
A print job attribute buffer that stores the attributes of the print job place name portion candidate, and an address portion that matches the attribute in the print job attribute buffer from the image data when the print job likely place name portion candidate exists. Indicative priority address part individual recognition means for cutting out a segment corresponding to one character and recognizing an individual character for each segment, and an address part from the image data when there is no place name part candidate that seems to be the impression. A general address part individual recognition means for cutting out a segment corresponding to one character and performing individual character recognition for each segment, and a result of the printing priority address part individual recognition means or a result of the general address part individual recognition means is stored. Address part recognition result buffer, and an address part character that meets the condition of the address part in the place name table memory from the address part recognition result buffer A general address part searching means for extracting a complementary combination, an address part candidate buffer for storing the result of the general address part searching means, and a final address reading based on the place name part candidate buffer and the address part candidate buffer. An address reading device, comprising: final determination means for determining a resulting character string.