JPH07271899A - Character recognition device - Google Patents

Abstract

PURPOSE:To provide a highly reliable character recognition device capable of obtaining the best recognized result considering of plural kinds of character recognition procedures. CONSTITUTION:A character on a postal matter P is optically read out, a character pattern obtained by the reading is collated with a reference pattern by a 1st recognition part 51 based upon composite similarity extracting processing to find out a 1st character candidate as a recognized result of the read character. The shape of the character pattern obtained by the reading is collated with reference shape data by a 2nd recognition part 52 based upon outline feature matching processing to find out a 2nd character candidate as a recognized result of the read character. These recognized results are allocated to a matrix table, a character candidate selecting index is read out and either one of the two recognized results is selected in accordance with the read index.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing a character written on a mail piece, for example.

[0002]

2. Description of the Related Art There is a mail sorting machine for automatically sorting mail by reading a zip code, a destination address, etc. from the mail.
This mail sorting machine is equipped with a character recognition device for recognizing the characters on the mail. This character recognition device optically reads on a mail piece, detects a character pattern for one character from this read image and cuts it out, and collates the cut out character pattern with the reference pattern of the reference character,
The character to be read is recognized based on the matching condition of both patterns. In some cases, the read character is recognized by collating the shape of the cut-out character pattern with the reference shape data.

[0003]

There are various character recognition methods, and in each case an accurate recognition result can be obtained according to the type of the character to be read (Kanji, Hiragana, Katakana, numbers). And sometimes it isn't.

The present invention takes the above circumstances into consideration,
It is an object of the present invention to provide a highly reliable character recognition device that can obtain the best recognition result in consideration of a plurality of types of character recognition methods.

[0005]

A character recognition device according to a first aspect of the present invention comprises a reading means for optically reading a character on an object to be read,
First recognition means for recognizing the read character from the character pattern read by the reading means, second recognition means for recognizing the read character from the character pattern read by the reading means, recognition result of the first recognition means and the above And means for obtaining a final recognition result from the recognition result of the second recognition means.

According to a second aspect of the present invention, there is provided a character recognizing device for optically reading a character on an object to be read, and a first character candidate as a recognition result for the read character from a character pattern read by the reading means. 1 recognizing means, a second recognizing means for obtaining a second character candidate as a recognition result for a read character from a character pattern read by the reading means, and a recognizing element of the first character candidate are arranged on one side for each character candidate, And a table in which the recognition elements of the second character candidates are arranged in the other for each character candidate, and the character candidate selection index based on the contrast of both arrangements is registered in advance, the recognition result of the first recognition means, and the second recognition means. Reading means for applying the recognition result of the above to the table to read the character candidate selection index, and the first recognizing hand according to the character candidate selection index read by the reading means. And a selecting means for selecting one of the recognition result either of the second recognition means.

A character recognition device according to a third aspect of the present invention recognizes a read character by a reading means for optically reading a character on an object to be read, and a character pattern read by the reading means and a reference reference pattern. As a result, the first recognizing means for obtaining the first character candidate and the second recognizing means for obtaining the second character candidate as the recognition result for the read character by collating the shape of the character pattern read by the reading means with the reference shape data. And the recognition element of the first character candidate is arranged on one side for each character candidate, and the recognition element of the second character candidate is arranged on the other side for each character candidate, and a character candidate selection index based on the contrast of both arrangements is set. A table registered in advance; a reading means for applying the output of the first recognizing means and the output of the second recognizing means to the table to read the character candidate selection index; Depending on the character candidate selection index to be read by the reading means and a selection means for selecting one of the recognition result of the first recognition means and the second recognition means.

In the character recognition device of the fourth aspect of the present invention, the reading means for optically reading the character on the object to be read, and the character pattern read by the reading means are used as reference patterns in the composite similarity extraction processing. First recognition means for matching and obtaining a first character candidate as a recognition result for a read character from the similarity between both patterns, and a confidence value for the similarity between both patterns, and the shape of the character pattern read by the reading means. Is compared with the reference shape data in the contour feature matching process to obtain the second character candidate as the recognition result for the read character, and the reliability value of the similarity as the recognition element of the first character candidate. A character candidate selection index based on the contrast of both arrangements is arranged for each character candidate, and the shape feature serving as a recognition element of the second character candidate is arranged for each character candidate on the other side. Registered table, reading means for applying the output of the first recognizing means and the output of the second recognizing means to the table, and reading the character candidate selection index, and character candidate selection read by the reading means. And a selection unit for selecting one of the recognition results of the first recognition unit and the second recognition unit according to the index.

[0009]

In the character recognition device of the first invention, the character on the object to be read is optically read, and two recognition results for the read character are obtained from the character pattern obtained by this reading. The final recognition result is obtained from these recognition results.

In the character recognition device of the second invention, the character on the object to be read is optically read, and from the character pattern obtained by this reading, the first character candidate and the second character result are obtained as two recognition results for the read character. Character candidates are requested.
On the other hand, the recognition element of the first character candidate is arranged in one for each character candidate, and the recognition element of the second character candidate is arranged in the other for each character candidate, and the character candidate selection index based on the contrast of both arrangements is registered in advance. There is a table, and the character candidate selection index is read by applying the above two recognition results to this table. Either one of the above two recognition results is selected according to the read character candidate selection index.

In the character recognition device of the third invention, the character on the object to be read is optically read, and the character pattern obtained by this reading and the reference reference pattern are collated by the first recognizing means and read. The first character candidate is obtained as the recognition result for the character. Further, the shape of the character pattern obtained by reading and the shape data for reference are collated by the second recognizing means, and the second character candidate is obtained as the recognition result for the read character. On the other hand, the recognition element of the first character candidate is arranged in one for each character candidate, and the recognition element of the second character candidate is arranged in the other for each character candidate, and the character candidate selection index based on the contrast of both arrangements is registered in advance. There is a table, and the output of the first recognizing means and the output of the second recognizing means are applied to this table to read the character candidate selection index. Either one of the above two recognition results is selected according to the read character candidate selection index.

In the character recognition device of the fourth aspect of the invention, the character on the object to be read is optically read, and the character pattern obtained by this reading and the reference reference pattern are extracted by the first recognition means as the composite similarity. The first character candidate is obtained as a recognition result for the read character by collating in the processing. Further, the shape of the character pattern obtained by reading and the shape data for reference are collated by the second recognizing means in the contour feature matching process, and the second character candidate is obtained as the recognition result for the read character. On the other hand, the recognition element of the first character candidate is arranged in one for each character candidate, and the recognition element of the second character candidate is arranged in the other for each character candidate, and the character candidate selection index based on the contrast of both arrangements is registered in advance. There is a table, and the output of the first recognizing means and the output of the second recognizing means are applied to this table to read the character candidate selection index. Either one of the above two recognition results is selected according to the read character candidate selection index.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. First, FIG. 2 shows the configuration of a mail sorting machine according to the present invention. FIG. 2 is a front view of the mail sorting machine.

The postal sorter 1 is provided with a supply unit 3 from which postal items P, such as postcards and sealed letters, are supplied collectively and in a standing position. The postal matter P supplied from the supply section 3 is conveyed to the take-out section 5. The take-out section 5 takes out the postal matter P one by one and conveys it to the take-in conveyance path 7. The take-in / conveyance path 7 is composed of a belt and a plurality of rollers, and nips and conveys the postal matter P.

An address recognition device 9 is installed on the upstream side of the take-in / transport path 7. The address recognition device 9 is a character recognition device of the present invention to which a word recognition function and an address discrimination function are added. The address recognition device 9 reads the characters of the address information described on the mail P, recognizes the read contents, and classifies them. Output specified data. The details of the address recognition device 9 will be described later.

The postal matter P passing through the address recognition device 9 is then conveyed to the sorting section 11. The sorting unit 11 is composed of, for example, a plurality of sorting boxes 13 ... Divided according to addresses and postal codes. Mailboxes P are sorted and accumulated in the sorting boxes 13 ... In accordance with the sorting designation data from the address recognition device 9 described above.

The postal matter P is transported to the sorting boxes 13 ... By the sorting transport paths 15a, 15b ,. The configuration of the address recognition device 9 is shown in FIG.

The address recognition device 9 comprises a character recognition device 20, a word recognition circuit 30, a word dictionary 31, an address recognition circuit 40, and an address dictionary 41 of the present invention. Character recognition device 2
Reference numeral 0 includes a photoelectric conversion unit 21, a binarization circuit 22, an address area detection unit 23, a selection circuit 24, a character detection / cutout circuit 25, a normalization circuit 26, a character recognition circuit 27, and a character dictionary 28.

The photoelectric conversion unit 21 irradiates the postal matter P with light by a light source (not shown), receives the reflected light from the postal matter P, and converts it into an electric signal. The output signal of the photoelectric conversion unit 21 is input to the binarization circuit 22.

The binarization circuit 22 binarizes the output signal of the photoelectric conversion unit 21 (a read signal corresponding to the image of the entire surface of the postal matter P) with a predetermined threshold value. The output signal (binarization signal) of the binarization circuit 22 is input to the address area detection unit 23 and the selection circuit 24.

The address area detection unit 23 specifies the address area of the image of the mail P by coordinates by the projection method or the like,
To detect. After the detection, a signal indicating that the predetermined range is the address area is output. The output signal of the address area detection unit 23 is input to the selection circuit 24.

The selection circuit 24 receives the binary data from the binarization circuit 22.
Among the binarized signals, the binarized signal corresponding to the addressed area is selected and output according to the detection result of the addressed area detection unit 23. The character detection / cutout circuit 25 performs line detection and cutout on the binarized signal of the address area input from the selection circuit 24,
Further, a character pattern for one character is detected and cut out based on the estimated pitch. The character pattern detected and cut out by the character detection / cutout circuit 25 is sent to the normalization circuit 26.

The normalization circuit 26 is a character detection / cutout circuit 25.
The character pattern sent from is normalized to a certain size. The character pattern normalized here is sent to the character recognition circuit 27.

In the character dictionary 28, a standard pattern corresponding to the whole image of the reference character and shape data representing the shape feature of the reference character are registered. Character recognition circuit 2
7 has the following functional means [1] to [5].

[1] The input character pattern is compared with the reference pattern for reference in the composite similarity extraction processing, the first character candidate is obtained as the recognition result for the read character from the similarity of both patterns, and the similarity of both patterns is obtained. A first recognition means for obtaining a confidence value for the degree.

[2] The shape of the input character pattern is collated with the reference shape data in the contour feature matching process, and the second character candidate is obtained as the recognition result for the read character.
Recognition means.

[3] The reliability value of the similarity, which is the recognition element of the first character candidate, is arranged on one side for each character candidate, and the shape feature, which is the recognition element of the second character candidate, is arranged on the other side for each character candidate. Table means in which character candidate selection indexes based on the contrast of both arrays are registered in advance.

[4] Read-out means for applying the output of the first recognizing means and the output of the second recognizing means to the table means to read the character candidate selection index. [5] Selection means for selecting one of the recognition results of the first recognition means and the second recognition means according to the character candidate selection index read by the reading means. This selection output becomes the final recognition result.

The recognition result of the character recognition circuit 27 is
It is sent to the word recognition circuit 30. The word recognition circuit 30 collates the combination of the read characters successively recognized by the character recognition circuit 27 with the word data registered in the word dictionary 31, and recognizes the word by the combination of the read characters. This recognition result is sent to the address recognition circuit 40.

The address recognition circuit 40 collates the word recognized by the word recognition circuit 30 with the address data registered in the address dictionary 41, and recognizes the address represented by the word. This recognition result is output as classification designation data. This sorting designation data is used for sorting and stacking the sorting boxes 13 in the mail sorting machine 1 as described above.

A specific example of the character recognition circuit 27 and the character dictionary 28 is shown in FIG. The normalized character pattern is input to the first recognition unit 51 and the second recognition unit 52.

The first recognition unit 51 compares the input character pattern with the reference reference pattern registered in the first memory 61 of the character dictionary 28 in the composite similarity extraction process, and reads it from the similarity of both patterns. The first recognition result for characters
Character candidates are obtained, and a confidence value for the similarity between both patterns is obtained.

The second recognizing unit 52 collates the shape of the input character pattern with the reference shape data registered in the second memory 62 of the character dictionary 28 by the contour feature matching process, and as a recognition result for the read character. Find a second character candidate.

The outputs of the first recognition section 51 and the second recognition section 52 are sent to the comprehensive evaluation section 53. This comprehensive evaluation unit 53
Applies the output of the first recognition unit 51 and the output of the second recognition unit 52 to the matrix table 54 to read the character candidate selection index, and according to the character candidate selection index, the first recognition unit 51 and the second recognition unit. One of the recognition results 52 is selected. This selection output becomes the final recognition result.

As shown in FIG. 4, the matrix table 54 has a reliability value (“0”, “1”, “2”, “3”) of similarity as a recognition element of the first character candidate in the first recognition unit 51.
Are arranged in the X direction for each character candidate (table columns 101 and 102), and the shape feature (entry number “0” “1” “2” “3”) that serves as a recognition element for the second character candidate in the second recognition unit 52.
Are arranged in the Y direction for each character candidate (table columns 103 and 104), and the character candidate selection index based on the contrast between both sequences is registered in the table column 105 in advance.

Here, the reliability values arrayed in the X direction are created from the actually calculated similarity using a statistical method. The reliability value “0” indicates that the degree of similarity has not reached the predetermined value, and is not suitable as a character candidate. The reliability value "1" indicates that the similarity reaches a predetermined value, but the difference from the similarity of the next character candidate does not reach a sufficient set value. The reliability value “2” indicates that the similarity reaches a predetermined value, but the difference from the similarity of the next character candidate reaches a sufficient set value by applying a predetermined bias. The reliability value “3” indicates that the similarity reaches a predetermined value and the difference from the similarity of the next character candidate reaches a sufficient set value.

The shape features arranged in the Y direction are created by a statistical method in consideration of the fact that the reliability of the shape differs depending on the type of the target character, and the entry numbers "0", "1" and "2" are used. "3" ... will be ranked as it is.

The character candidate selection index is, as shown in FIG.
It consists of a character candidate selection code and a comprehensive evaluation value. The character candidate selection code includes the recognition result of the first recognition unit 51 and the second recognition result.
This is for designating which of the recognition results of the recognition unit 52 should be selected. Overall evaluation value is "0""1"
It is a numerical value of "2", "3" ..., and becomes larger as the reliability of both recognition results is higher.

A large number of character candidate selection indexes are prepared, and the comprehensive evaluation values included in these character candidate selection indexes are arranged in the table column 105 as a representative. Next, the operation of the above configuration will be described.

The image on the postal matter P is optically read.
A portion of the read image corresponding to the address area is detected, and a character pattern for one character is detected and cut out from the image of the address area. The cut-out character pattern is normalized to a certain size.

The normalized character pattern is sent to the first recognition section 51 and the second recognition section 52. First recognition unit 51
Then, the input character pattern is the first in the composite similarity extraction processing.
The reference character pattern in the memory 61 is collated, the first character candidate is obtained as a recognition result for the read character from the similarity of both patterns, and the confidence value for the similarity of both patterns is obtained.

In the second recognition section 52, the shape of the input character pattern is collated with the reference shape data in the second memory 62 by the contour feature matching process, and the second character candidate is obtained as the recognition result for the read character. .

The outputs of the first recognizing unit 51 and the second recognizing unit 52 are applied to the matrix table 54, whereby the character candidate selection index is read. For example, when the first recognition unit 51 obtains the first character candidate “A” and the confidence value “3” and the second recognition unit 52 obtains the second character candidate “A”, the second character candidate The total evaluation value "4" existing at the intersection of the X-direction line (broken line in the drawing) along the entry number (shape feature) related to "A" and the Y-direction line (broken line in the drawing) along the reliability value "3" It is read.

When the comprehensive evaluation value "4" is read, the character candidate selection code attached to it is determined. Then, according to the character candidate selection code, either the first character candidate recognized by the first recognition unit 51 or the second character candidate recognized by the second recognition unit 52 is selected and output. This selection output becomes the final recognition result.

This recognition result is sent to the word recognition circuit 30, where word recognition is performed. Further, the address recognition circuit 40 recognizes the address based on the result of the word recognition.
In this way, the recognition result by the composite similarity degree extraction process and the recognition result by the contour feature matching process are compared on a matrix table created in advance by a statistical method, and thereby the character candidate selection index is read, Since either one of the two recognition results is selected according to the character candidate selection index, the best recognition result considering each feature of the composite similarity extraction process and the contour feature matching process, which are character recognition methods, is obtained. can get. The composite similarity extraction process and the contour feature matching process have been described as examples of character recognition types, but the types can be appropriately selected.

[0046]

As described above, the character recognition device of the first invention optically reads a character on the object to be read and obtains two recognition results for the read character from the character pattern obtained by this reading. Since the final recognition result is obtained from these recognition results, the reliability is excellent in that the best recognition result can be obtained in consideration of a plurality of types of character recognition methods.

The character recognition device of the second invention optically reads a character on the object to be read, and the first character candidate and the second character candidate are obtained as two recognition results for the read character from the character pattern obtained by this reading. On the other hand, the recognition element of the first character candidate is arranged on one side for each character candidate, and the recognition element of the second character candidate is arranged on the other side for each character candidate, and a character candidate selection index based on the contrast of both arrangements is obtained. Is prepared in advance, the above two recognition results are applied to this table, the character candidate selection index is read out, and either one of the above two recognition results is selected according to the read character candidate selection index. Since the configuration is adopted, the reliability of obtaining the best recognition result in consideration of plural kinds of character recognition methods is excellent.

In the character recognition device of the third invention, the character on the object to be read is optically read, the character pattern obtained by this reading and the reference reference pattern are collated by the first recognition means, and the read character is read. The first character candidate is obtained as a recognition result for, and the shape of the character pattern obtained by reading is compared with the reference shape data by the second recognition means, and the second character candidate is obtained as a recognition result for the read character. , The recognition element of the first character candidate is arranged in one for each character candidate, and the recognition element of the second character candidate is arranged in the other for each character candidate, and the character candidate selection index based on the contrast of both arrangements is registered in advance. A table is prepared, the output of the first recognizing means and the output of the second recognizing means are applied to this table to read out the character candidate selection index, and the character candidate selection index is read out. Since a configuration for selecting one of the two recognition results Te, it becomes the best considering the approach of a plurality of types of character recognition of the recognition result is excellent reliability obtained.

In the character recognition device of the fourth invention, the character on the object to be read is optically read, and the character pattern obtained by this reading and the reference pattern for reference are subjected to the composite similarity extraction processing by the first recognition means. And the first character candidate is obtained as a recognition result for the read character, and the shape of the character pattern obtained by the reading and the shape data for reference are collated by the contour recognizing process by the second recognizing means and read. The second character candidate is obtained as a recognition result for the character, while the recognition element of the first character candidate is arranged on one side for each character candidate, and the recognition element of the second character candidate is arranged on the other side for each character candidate. A table is prepared in which character candidate selection indexes based on arrangement contrasts are registered in advance, and the output of the first recognizing means and the output of the second recognizing means are applied to this table and the sentence Since the candidate selection index is read and either one of the above two recognition results is selected according to the read character candidate selection index, the best recognition result in consideration of a plurality of types of character recognition methods can be obtained. It will be highly reliable.

[Brief description of drawings]

FIG. 1 is a block diagram showing a specific configuration of a main part of an embodiment of the present invention.

FIG. 2 is a configuration diagram of a mail sorting machine according to the embodiment.

FIG. 3 is a block diagram showing the overall configuration of the same embodiment.

FIG. 4 is a diagram showing a format of a matrix table in the embodiment.

FIG. 5 is a diagram showing a format of a character candidate selection index in the embodiment.

[Explanation of symbols]

1 ... Mail sorter, P ... Mail, 21 ... Photoelectric converter, 2
5 ... Character detection / cutout circuit, 26 ... Normalization circuit, 27 ... Character recognition circuit, 28 ... Character dictionary, 51 ... First recognition unit, 52 ...
2nd recognition part, 53 ... Comprehensive evaluation part, 54 ... Matrix table, 61 ... 1st memory, 62 ... 2nd memory.

Claims (4)

[Claims] 1. A reading unit for optically reading a character on an object to be read, a first recognizing unit for recognizing a read character from a character pattern read by the reading unit, and a read from a character pattern read by the reading unit. A character recognition device comprising: a second recognition unit that recognizes a character; and a unit that obtains a final recognition result from the recognition result of the first recognition unit and the recognition result of the second recognition unit. 2. A reading means for optically reading a character on an object to be read, a first recognition means for obtaining a first character candidate as a recognition result for the read character from a character pattern read by the reading means, and the reading means. Second recognition means for obtaining a second character candidate as a recognition result for the read character from the character pattern read by, and recognizing the second character candidate by arranging the recognition elements of the first character candidate on one side for each character candidate. A table in which elements are arranged on the other side for each character candidate and a character candidate selection index based on the contrast of both arrangements is registered in advance, and the recognition result of the first recognition means and the recognition result of the second recognition means are applied to the table. Reading means for reading the character candidate selection index, and the first recognizing means and the second recognizing means according to the character candidate selection index read by the reading means. Character recognition apparatus characterized by comprising selecting means for selecting either one of the recognition result. 3. A first character candidate is obtained as a recognition result for the read character by comparing a reading means for optically reading a character on the object to be read and a character pattern read by the reading means with a reference reference pattern. A first recognizing means, a second recognizing means for obtaining a second character candidate as a recognition result for the read character by collating the shape of the character pattern read by the reading means with reference shape data; A table in which the recognition elements are arranged in one for each character candidate, the recognition elements in the second character candidate are arranged in the other for each character candidate, and a character candidate selection index based on the contrast of both arrays is registered in advance; Read means for applying the output of the first recognizing means and the output of the second recognizing means to the table to read the character candidate selection index, and a sentence read by the read means. Character recognition apparatus characterized by comprising a selection means for selecting either one of the recognition result of the first recognition means and the second recognizing means in response to the candidate selection index. 4. A reading means for optically reading a character on an object to be read, and a character pattern read by this reading means is collated with a reference pattern for reference in a composite similarity extraction processing, and the similarity between both patterns is compared. From the first recognition means for obtaining the first character candidate as a recognition result for the read character and the confidence value for the similarity of both patterns, and the shape of the character pattern read by the reading means is referred to by the contour feature matching process. Second matching means for finding the second character candidate as a recognition result for the read character and the reliability value of the similarity as the recognition element of the first character candidate are arranged on one side for each character candidate. , And the shape feature, which is the recognition element of the second character candidate, is arranged in the other for each character candidate,
A table in which character candidate selection indexes based on the contrast of both sequences are registered in advance; and a reading means for applying the outputs of the first recognition means and the second recognition means to the table to read the character candidate selection indexes. Character selecting means for selecting one of the recognition results of the first recognizing means and the second recognizing means in accordance with the character candidate selection index read by the reading means. apparatus.