JP2012064057A - Character recognition device, sorting apparatus, sorting control device, and character recognition method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character recognition device with excellence in recognition of missing characters.SOLUTION: A character recognition device in one embodiment includes detection means and recognition means. The detection means detects each of character candidates from an image. The recognition means recognizes each of the character candidates based on a plurality of character recognition dictionaries corresponding to the degree of multiple different character losses.

Description

  Embodiments described herein relate generally to a character recognition device such as an OCR (optical character reader) that reads characters in an image. Embodiments of the present invention also relate to a sorting apparatus that sorts sorting objects such as mail. Embodiments of the present invention also relate to a sorting control device that controls sorting of sorting objects such as mail. Embodiments of the present invention also relate to a character recognition method for reading characters in an image.

  In a sealed letter with a window, the destination address is located at the edge of the window, a part of characters constituting the destination address may be hidden, and characters may be missing. Various techniques for recognizing such missing characters have been proposed in addition to the sealed letter window.

  For example, a technique has been proposed in which the height of characters in a hidden line is estimated from the height of characters in a non-hidden line, and the character is recognized using the estimation result. In addition, a technique has been proposed in which the height of a character missing when removing an underline is estimated from the height of a character in a line without an underline, and the character is recognized using the estimation result.

JP-A-4-148391

  However, the above-described technology still has insufficient character recognition accuracy, and a more accurate character recognition technology is desired.

  An object of the present invention is to provide a character recognition device and a character recognition method that are excellent in recognition of missing characters. Another object of the present invention is to provide a sorting apparatus and a sorting control apparatus that are excellent in recognition of missing characters and excellent in sorting accuracy.

  The character recognition device of the embodiment includes a detection unit and a recognition unit. The detection means detects each character candidate from the image. The recognition means recognizes each character candidate based on a plurality of character recognition dictionaries corresponding to a plurality of different character missing degrees.

It is a block diagram which shows an example of the classification system (character recognition apparatus) common in 1st and 2nd embodiment. It is a block diagram which shows the modification of the division system common in 1st and 2nd embodiment. It is a flowchart which shows an example of the character string recognition process common in 1st and 2nd embodiment. It is a figure which shows an example of the division | segmentation target object which concerns on 1st Embodiment. It is a figure which shows an example of the detection of the character candidate which concerns on 1st Embodiment. It is a figure which shows an example of the estimation of the character lower end position which concerns on 1st Embodiment. It is a figure which shows an example of selection of the character recognition candidate which concerns on 1st Embodiment. It is a figure which shows an example of the division | segmentation target object which concerns on 2nd Embodiment. It is a figure which shows an example of the detection of the character candidate which concerns on 2nd Embodiment. It is a figure which shows an example of the estimation of the character lower end position which concerns on 2nd Embodiment. It is a figure which shows an example of selection of the character recognition candidate which concerns on 2nd Embodiment.

  The first and second embodiments will be described below with reference to the drawings.

  FIG. 1 is a block diagram showing an example of a sorting system common to the first and second embodiments.

  As shown in FIG. 1, the sorting system 1 includes a transport unit 11, a reading unit 12, a recognition unit 13, a character recognition dictionary database 14, an alignment verification unit 15, a character string verification unit 16, a character string database 17, and a sorting unit 18. I have. The transport unit 11, the reading unit 12, and the sorting unit 18 constitute a sorting processing unit (sorting device) 1A, and also includes a recognition unit 13, a character recognition dictionary database 14, an alignment verification unit 15, a character string verification unit 16, The character string database 17 constitutes a character recognition processing unit (character recognition device) 1B.

  As shown in FIG. 2, the sorting system 1 can also be configured. That is, the sorting system 1 can be configured by a plurality of sorting processing units 1A, a single character recognition processing unit 1B, and a communication unit 1C. The single character recognition processing unit 1B and the communication unit 1C constitute a sorting control processing unit (sorting control device) 1D. The communication unit 1C transmits information (image data of the classification target object) from the plurality of classification processing units 1A to the character recognition processing unit 1B, and information from the character recognition processing unit 1B (addresses read and recognized from the image) Information (classification information)) is transmitted to a plurality of classification processing units 1A. By configuring the sorting system 1 as shown in FIG. 2, it is possible to distribute sorting processing and concentrate discrimination processing (recognition processing), and to improve the overall processing efficiency.

  The conveyance part 11 is comprised by the conveyance path etc., and conveys division | segmentation target objects, such as a letter and a parcel, along a conveyance path. The reading unit 12 reads an image of the sorting object in the middle of the conveyance path. For example, the sorting object is a sealed letter with a window, and a part of a character string constituting the destination is hidden at the edge of the window.

  The character recognition dictionary database 14 stores a plurality of character recognition dictionaries corresponding to a plurality of different character missing degrees. Further, the character recognition dictionary database 14 also stores a character recognition dictionary of complete characters with no missing characters. For example, the character recognition dictionary database 14 stores N (N: natural number) character recognition dictionaries of character recognition dictionaries D1, D2,.

  For example, the character recognition dictionary D1 (deficient dictionary) is a character recognition dictionary that is generated from a plurality of characters that have no missing characters. The character recognition dictionary D2 (10% missing dictionary) is a character recognition dictionary generated from a plurality of characters missing from 1% to 20%. The character recognition dictionary D3 (30% missing dictionary) is a character recognition dictionary generated from a plurality of characters missing 21% to 40%. The character recognition dictionary D4 (50% missing dictionary) is a character recognition dictionary generated from a plurality of characters lacking 41% to 60%.

  The recognition unit 13 detects each character candidate from the image of the classified object shown in FIG. For example, the recognition unit 13 detects an image that seems to be a character line and a character string from the image of the classification target object, detects an image that seems to be a plurality of characters from the image that seems to be a character line and a character string, and Detect multiple character candidates.

  Further, the recognition unit 13 recognizes each character candidate based on a plurality of character recognition dictionaries D1, D2,..., DN stored in the character recognition dictionary database. That is, the recognition unit 13 can recognize a destination address (a destination address configured by each character candidate) described in the classification target object.

  The arrangement verification unit 15 verifies whether or not the lower ends (lower sides) are arranged in an orderly manner when the missing characters are restored (see FIG. 6). In other words, the alignment verification unit 15 verifies whether or not the lower end (lower side) is aligned along a line of characters when the lack of each character candidate is restored.

  The character string verification unit 16 uses the character string data (address data) stored in the character string database 17 to verify whether the character string exists as character string data (address data). Note that the character string data (address data) is a character string pattern that may be described in the classification object.

  The sorting unit 18 sorts the sorting object transported by the transporting unit 11 based on destination information (sorting information) corresponding to the character recognition result by the recognition unit 13.

(First embodiment)
Next, an example of the character string recognition process of the first embodiment will be described with reference to FIG. In the first embodiment, an example of a Japanese character string recognition process will be described.

  First, the reading unit 12 reads an image of a sorting object in the middle of the conveyance path (S1). FIG. 4 is a diagram illustrating an example of the sorting target object according to the first embodiment. As shown in FIG. 4, for example, the sorting target is a sealed letter with a window, and a part of the character string “Toshiba Taro” constituting the destination is hidden at the edge of the window.

  Subsequently, the recognition unit 13 extracts the character candidates C1, C2, C3, and C4 as shown in FIG. 5 from the image of the classified object shown in FIG. 4 (S2). A plurality of character-like image regions surrounded by dotted rectangles shown in FIG. 5 are character candidates C1, C2, C3, and C4.

  Note that the recognition unit 13 can also extract a plurality of character candidates from an image region that seems to be a single character. For example, as shown in FIG. 5, the recognition unit 13 can also extract a plurality of character candidates C4a and C4b from an image area that seems to be a single character. For example, the character candidate C4a corresponds to a character edge (left part of the character), and the character candidate C4b corresponds to a character formation (right part of the character).

  Subsequently, the recognition unit 13 selects one or more character recognition candidates corresponding to the character candidates C1, C2, C3, and C4 based on the plurality of character recognition dictionaries D1, D2,..., DN (S3). Note that the recognition unit 13 selects one or more character recognition candidates that satisfy a condition of a predetermined similarity or higher (or a predetermined evaluation value or higher) for each character candidate. For example, when a predetermined character candidate completely matches the character recognition dictionary DM (1 ≦ M ≦ N), the predetermined character candidate has a similarity 1000 (or evaluation value 1000) with respect to the character recognition dictionary DM. And The recognition unit 13 selects one or more character recognition candidates that satisfy a condition of similarity 700 (or evaluation value 700) or more for each character candidate based on the plurality of character recognition dictionaries D1, D2,. Hereinafter, the description of “similarity” may be read as “evaluation value”. In the present embodiment, a method for recognizing a character based on similarity and a method for recognizing a character based on an evaluation value are described as examples. However, other character recognition methods may be used.

  For example, as shown in FIG. 7, the recognition unit 13 selects character recognition candidates C11, C12, and C13 corresponding to the character candidate C1, selects a character recognition candidate C21 corresponding to the character candidate C2, and sets it as the character candidate C3. Corresponding character recognition candidates C31 and C32 are selected, and character recognition candidates C41 and C42 corresponding to the character candidate C4 are selected.

  That is, the character recognition candidate C11 corresponds to “car” having a similarity of 950 (first place) with respect to the character recognition dictionary D3 (30% missing dictionary), and the character recognition candidate C12 is the character recognition dictionary D4 (50% missing). The character recognition candidate C13 corresponds to “east” having a similarity of 900 (second place) with respect to (dictionary), and “seedling” with a similarity of 800 (third place) with respect to character recognition dictionary D2 (10% missing dictionary). Corresponding to

  The character recognition candidate C21 corresponds to “turf” having a similarity of 900 (first place) with respect to the character recognition dictionary D4 (50% missing dictionary).

  The character recognition candidate C31 corresponds to “Large” having a similarity of 850 (first place) with respect to the character recognition dictionary D4 (50% missing dictionary), and the character recognition candidate C32 corresponds to the character recognition dictionary D4 (50% missing). Corresponds to “thick” with a similarity of 850 (second place).

  Further, the character recognition candidate C41 corresponds to “Buro” having a similarity of 950 (first place) with respect to the character recognition dictionary D1 (missing dictionary), and the character recognition candidate C42 is the character recognition dictionary D1 (missing dictionary). Corresponds to “Ryo” with a similarity of 750 (2nd place).

  Note that one or more character recognition candidates (character recognition candidates C11, C12, C13) corresponding to the character candidate C1 are referred to as a first character recognition candidate group G1, and one or more character recognition candidates (characters) corresponding to the character candidate C2. The recognition candidate C21) is referred to as a second character recognition candidate group G2, and one or more character recognition candidates (character recognition candidates C31, C32) corresponding to the character candidate C3 are referred to as a third character recognition candidate group G3. One or more character recognition candidates (character recognition candidates C41, C42) corresponding to C4 are referred to as a fourth character recognition candidate group G4.

  Subsequently, the character string verification unit 16 selects one character recognition candidate from each character recognition candidate group G1, G2, G3, and G4, and generates one or more character string candidates by combining the selected character recognition candidates ( S4).

  For example, the character string verification unit 16 selects the character recognition candidate C11 from the first character recognition candidate group G1, selects the character recognition candidate C21 from the second character recognition candidate group G2, and then selects the third character recognition candidate group. A character recognition candidate C31 is selected from G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a first character string candidate (Karo Shiba) is generated.

  Similarly, the character string verification unit 16 selects the character recognition candidate C11 from the first character recognition candidate group G1, selects the character recognition candidate C21 from the second character recognition candidate group G2, and then selects the third character recognition candidate. A character recognition candidate C32 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a second character string candidate (Taro Kashiba) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C12 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and then selects a third character recognition candidate. A character recognition candidate C31 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a third character string candidate (Taro Toshiba) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C12 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and then selects a third character recognition candidate. A character recognition candidate C32 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a fourth character string candidate (Taro Toshiba) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C13 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and selects a third character recognition candidate. A character recognition candidate C31 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a fifth character string candidate (Nao Shiba) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C13 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and selects a third character recognition candidate. A character recognition candidate C32 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a sixth character string candidate (Taro Naeshiba) is generated.

  Subsequently, the alignment verifying unit 15 verifies whether or not the lower ends (lower sides) are arranged in an orderly manner when the missing character string candidates are restored (S5). For example, the character recognition candidate C13 (seedling) constituting the fifth character string candidate is selected based on the character recognition dictionary D2 (10% missing dictionary). Therefore, the arrangement verification unit 15 estimates that the character recognition candidate C13 is missing 10%, and estimates the character lower end position P1 of the character recognition candidate C13 as shown in FIG.

  Similarly, the character recognition candidate C21 (turf) constituting the fifth character string candidate is selected based on the character recognition dictionary D4 (50% missing dictionary). Therefore, the arrangement verification unit 15 estimates that the character recognition candidate C21 is missing 50%, and estimates the character lower end position P2 of the character recognition candidate C21 as shown in FIG.

  Similarly, the character recognition candidate C31 (large) constituting the fifth character string candidate is selected based on the character recognition dictionary D4 (50% missing dictionary). Therefore, the arrangement verification unit 15 estimates that the character recognition candidate C31 is missing 50%, and estimates the character lower end position P3 of the character recognition candidate C31 as shown in FIG.

  Similarly, the character recognition candidate C41 (Buro) that constitutes the fifth character string candidate is selected based on the character recognition dictionary D1 (deficient dictionary). Therefore, the alignment verification unit 15 estimates that the character recognition candidate C41 is not missing, and estimates the character lower end position P4 of the character recognition candidate C41 as shown in FIG.

  As shown in FIG. 6, only the character lower end position P1 of the character recognition candidate C13 is any of the character lower end position P2 of the character recognition candidate C21, the character lower end position P3 of the character recognition candidate C31, or the character lower end position P4 of the character recognition candidate C41. Is shifted upwards. For this reason, the arrangement verification unit 15 determines that the fifth character string candidate is not an appropriate character string candidate, and rejects the fifth character string candidate. As described above, the arrangement verifying unit 15 verifies the arrangement of the character string candidates and rejects inappropriate character string candidates. In other words, the arrangement verification unit 15 verifies a combination of character recognition candidates constituting each character string candidate and rejects an inappropriate character string candidate.

  Subsequently, the character string verification unit 16 verifies whether or not each character string candidate that has not been rejected by the alignment verification unit 15 is an appropriate character string, based on the character string data (address data) stored in the character string database 17. (S6). In other words, the character string verification unit 16 verifies a combination of character recognition candidates that constitute each character string candidate based on the character string data (address data) stored in the character string database 17, and inappropriate character string candidates. Will be rejected. For example, if the first character string candidate (Taro Koshiba) does not exist in the character string data, the character string verification unit 16 determines that the first character string candidate is not an appropriate character string candidate, Reject string candidates for. In this way, the character string verification unit 16 verifies whether or not each character string candidate is an appropriate character string, and rejects an inappropriate character string candidate.

  Subsequently, the recognition unit 13 outputs a character string recognition result corresponding to an appropriate character string candidate based on the character arrangement verification result and the character string verification result (S7). That is, the recognition unit 13 outputs a character string recognition result corresponding to an appropriate character string candidate that has not been rejected by the character string verification by the character string verification unit 15 and has not been rejected by the character string verification by the character string verification unit 16. To do. In other words, the recognizing unit 13 verifies the combination verification result of the character recognition candidates constituting each character string candidate by the arrangement verification unit 15 and the verification of the combination of character recognition candidates constituting each character string candidate by the character string verification unit 16. Based on the result, a character string recognition result corresponding to an appropriate character string candidate is output.

  For example, when the recognition unit 13 selects a plurality of appropriate character string candidates, the average similarity of each character recognition candidate string corresponding to the selected plurality of appropriate character string candidates is calculated, and the maximum average similarity is calculated. The character string recognition result corresponding to the character string candidate (the optimum character string candidate) is output. For example, when the recognition unit 13 selects a fourth character string candidate and a sixth character string candidate as appropriate character string candidates, each character recognition candidate string (character recognition) corresponding to the fourth character string candidate is selected. The average similarity ((900 + 900 + 850 + 950) / 4 = 900) of the candidates C12, C21, C32, C41) and the character recognition candidate strings corresponding to the sixth character string candidate (character recognition candidates C13, C21, C32, C41) Are compared with the average similarity ((800 + 900 + 850 + 950) / 4 = 875), and a character string recognition result corresponding to the fourth character string candidate (optimum character string candidate) is output.

(Second Embodiment)
Next, an example of a character string recognition process according to the second embodiment will be described with reference to FIG. In the second embodiment, an example of an English character string recognition process will be described.

  First, the reading unit 12 reads an image of a sorting object in the middle of the conveyance path (S1). FIG. 8 is a diagram illustrating an example of the sorting target object according to the second embodiment. As shown in FIG. 8, for example, the sorting object is a sealed letter with a window, and a part of the character string “TARG” constituting the destination is hidden at the edge of the window.

  Subsequently, the recognition unit 13 extracts character candidates C1, C2, C3, and C4 as shown in FIG. 9 from the image of the classified object shown in FIG. 8 (S2). A plurality of character-like image regions surrounded by dotted rectangles shown in FIG. 9 are character candidates C1, C2, C3, and C4.

  Note that the recognition unit 13 can also extract a plurality of character candidates from an image region that seems to be a single character. For example, as shown in FIG. 9, the recognition unit 13 can also extract a plurality of character candidates C4a and C4b from an image region that seems to be a single character.

  Subsequently, the recognition unit 13 selects one or more character recognition candidates corresponding to the character candidates C1, C2, C3, and C4 based on the plurality of character recognition dictionaries D1, D2,..., DN (S3). Note that the recognition unit 13 selects one or more character recognition candidates that satisfy a condition of a predetermined similarity or higher for each character candidate. For example, when a predetermined character candidate completely matches the character recognition dictionary DM (1 ≦ M ≦ N), the predetermined character candidate has a similarity 1000 with respect to the character recognition dictionary DM. The recognition unit 13 selects one or more character recognition candidates that satisfy a condition of a similarity of 700 or more for each character candidate based on the plurality of character recognition dictionaries D1, D2,.

  For example, as shown in FIG. 11, the recognition unit 13 selects character recognition candidates C11, C12, and C13 corresponding to the character candidate C1, selects a character recognition candidate C21 corresponding to the character candidate C2, and sets it as the character candidate C3. Corresponding character recognition candidates C31, C32 and C33 are selected, and character recognition candidates C41 and C42 corresponding to the character candidate C4 are selected.

  That is, the character recognition candidate C11 corresponds to “T” having a similarity of 950 (first place) with respect to the character recognition dictionary D1 (missing dictionary), and the character recognition candidate C12 is the character recognition dictionary D4 (50% missing dictionary). ) Corresponds to “I” having a similarity of 900 (second place), and the character recognition candidate C13 is set to “T” having a similarity of 850 (third place) with respect to the character recognition dictionary D2 (10% missing dictionary). Correspond.

  The character recognition candidate C21 corresponds to “A” having a similarity of 900 (first place) with respect to the character recognition dictionary D3 (30% missing dictionary).

  The character recognition candidate C31 corresponds to “B” having a similarity of 900 (first place) with respect to the character recognition dictionary D3 (30% missing dictionary), and the character recognition candidate C32 corresponds to the character recognition dictionary D3 (30% missing). The character recognition candidate C33 corresponds to “D” having a similarity of 700 (third place) with respect to the character recognition dictionary D1 (no missing dictionary). Correspond.

  The character recognition candidate C41 corresponds to “G” having a similarity of 850 (first place) with respect to the character recognition dictionary D3 (30% missing dictionary), and the character recognition candidate C42 corresponds to the character recognition dictionary D2 (10% missing). Corresponds to “E” having a similarity of 700 (second place) with respect to (dictionary).

  Note that one or more character recognition candidates (character recognition candidates C11, C12, C13) corresponding to the character candidate C1 are referred to as a first character recognition candidate group G1, and one or more character recognition candidates (characters) corresponding to the character candidate C2. Recognition candidate C21) is referred to as a second character recognition candidate group G2, and one or more character recognition candidates (character recognition candidates C31, C32, C33) corresponding to the character candidate C3 are referred to as a third character recognition candidate group G3. One or more character recognition candidates (character recognition candidates C41 and C42) corresponding to the character candidate C4 are referred to as a fourth character recognition candidate group G4.

  Subsequently, the character string verification unit 16 selects one character recognition candidate from each character recognition candidate group G1, G2, G3, and G4, and generates one or more character string candidates (S4).

  For example, the character string verification unit 16 selects the character recognition candidate C11 from the first character recognition candidate group G1, selects the character recognition candidate C21 from the second character recognition candidate group G2, and then selects the third character recognition candidate group. A character recognition candidate C31 is selected from G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a first character string candidate (TABG) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C12 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and then selects a third character recognition candidate. A character recognition candidate C31 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a second character string candidate (IABG) is generated.

  Similarly, the character string verification unit 16 selects the character recognition candidate C11 from the first character recognition candidate group G1, selects the character recognition candidate C21 from the second character recognition candidate group G2, and then selects the third character recognition candidate. A character recognition candidate C32 is selected from the group G3, a character recognition candidate C41 is selected from the fourth character recognition candidate group G4, and a third character string candidate (TARG) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C12 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and then selects a third character recognition candidate. A character recognition candidate C32 is selected from the group G3, a character recognition candidate C42 is selected from the fourth character recognition candidate group G4, and a fourth character string candidate (IARE) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C13 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and selects a third character recognition candidate. A character recognition candidate C32 is selected from the group G3, a character recognition candidate C42 is selected from the fourth character recognition candidate group G4, and a fifth character string candidate (TADE) is generated.

  Similarly, the character string verification unit 16 selects a character recognition candidate C13 from the first character recognition candidate group G1, selects a character recognition candidate C21 from the second character recognition candidate group G2, and selects a third character recognition candidate. A character recognition candidate C33 is selected from the group G3, a character recognition candidate C42 is selected from the fourth character recognition candidate group G4, and a sixth character string candidate (TARE) is generated.

  Subsequently, the alignment verifying unit 15 verifies whether or not the lower ends (lower sides) are arranged in an orderly manner when the missing character string candidates are restored (S5). For example, the character recognition candidate C11 (T) that constitutes the fifth character string candidate is selected based on the character recognition dictionary D1 (missing dictionary). Therefore, the alignment verification unit 15 estimates that the character recognition candidate C11 is not missing, and estimates the character lower end position P1 of the character recognition candidate C11 as shown in FIG.

  Similarly, the character recognition candidate C21 (A) constituting the fifth character string candidate is selected based on the character recognition dictionary D3 (30% missing dictionary). Therefore, the arrangement verification unit 15 estimates that the character recognition candidate C21 is 30% missing, and estimates the character lower end position P2 of the character recognition candidate C21 as shown in FIG.

  Similarly, the character recognition candidate C33 (D) that constitutes the fifth character string candidate is selected based on the character recognition dictionary D1 (the missing dictionary). Therefore, the arrangement verification unit 15 estimates that the character recognition candidate C33 is not missing, and estimates the character lower end position P3 of the character recognition candidate C33 as shown in FIG.

  Similarly, the character recognition candidate C42 (E) constituting the fifth character string candidate is selected based on the character recognition dictionary D2 (10% missing dictionary). Therefore, the arrangement verification unit 15 estimates that the character recognition candidate C41 is missing 10%, and estimates the character lower end position P4 of the character recognition candidate C42 as shown in FIG.

  As shown in FIG. 10, the character lower end position P1 of the character recognition candidate C11, the character lower end position P2 of the character recognition candidate C21, the character lower end position P3 of the character recognition candidate C33, and the character lower end position P4 of the character recognition candidate C42 are in the character string direction. Are not aligned. For this reason, the arrangement verification unit 15 determines that the fifth character string candidate is not an appropriate character string candidate, and rejects the fifth character string candidate. As described above, the arrangement verifying unit 15 verifies the arrangement of the character string candidates and rejects inappropriate character string candidates. In other words, the arrangement verification unit 15 verifies a combination of character recognition candidates constituting each character string candidate and rejects an inappropriate character string candidate.

  Subsequently, the character string verification unit 16 verifies whether or not each character string candidate that has not been rejected by the alignment verification unit 15 is an appropriate character string, based on the character string data (address data) stored in the character string database 17. (S6). In other words, the character string verification unit 16 verifies a combination of character recognition candidates that constitute each character string candidate based on the character string data (address data) stored in the character string database 17, and inappropriate character string candidates. Will be rejected. For example, if the fourth character string candidate (IARE) does not exist in the character string data, the character string verification unit 16 determines that the fourth character string candidate is not an appropriate character string candidate, and the fourth character string Reject the candidate. In this way, the character string verification unit 16 verifies whether or not each character string candidate is an appropriate character string, and rejects an inappropriate character string candidate.

  Subsequently, the recognition unit 13 outputs a character string recognition result corresponding to an appropriate character string candidate based on the character arrangement verification result and the character string verification result (S7). That is, the recognition unit 13 outputs a character string recognition result corresponding to an appropriate character string candidate that has not been rejected by the character string verification by the character string verification unit 15 and has not been rejected by the character string verification by the character string verification unit 16. To do. In other words, the recognizing unit 13 verifies the combination verification result of the character recognition candidates constituting each character string candidate by the arrangement verification unit 15 and the verification of the combination of character recognition candidates constituting each character string candidate by the character string verification unit 16. Based on the result, a character string recognition result corresponding to an appropriate character string candidate is output.

  When the recognition unit 13 selects a plurality of appropriate character string candidates, the recognition unit 13 calculates the average similarity of each character recognition candidate sequence corresponding to the selected plurality of appropriate character string candidates. The character string recognition result corresponding to the character string candidate having the maximum average similarity (optimum character string candidate) is output. For example, when the recognition unit 13 selects a first character string candidate and a second character string candidate as appropriate character string candidates, each character recognition candidate string (character recognition) corresponding to the first character string candidate is selected. The average similarity ((950 + 900 + 900 + 850) / 4 = 900) of the candidates C11, C21, C31, C41) and the character recognition candidate strings (character recognition candidates C12, C21, C31, C41) corresponding to the second character string candidates Are compared with the average similarity ((900 + 900 + 900 + 850) /4=887.5), and a character string recognition result corresponding to the first character string candidate (optimum character string candidate) is output.

  The present invention is not limited to the character string recognition process described in the first and second embodiments. For example, a method for selecting an optimal character string candidate from a plurality of appropriate character string candidates is not limited to the method described in the first and second embodiments.

  In the first and second embodiments, the alignment verification unit 15 rejects inappropriate character string candidates, the character string verification unit 16 rejects inappropriate character string candidates, and the recognition unit 13 In the above description, the optimum character string candidate is selected from the remaining one or more appropriate character string candidates.

  However, an optimum character string candidate can be selected as follows. For example, the arrangement verification unit 15 gives a character arrangement evaluation value to each character string candidate based on the character arrangement verification result, and the character string verification unit 16 applies to each character string candidate based on the character string verification result. The character string evaluation value is given, and the recognition unit 13 comprehensively selects the optimum character string candidate from each character string candidate based on the character string evaluation value of each character string candidate and the character string evaluation value of each character string candidate. Character string candidates can also be selected.

  In addition, the recognition unit 13 selects an optimum character string candidate from each character string candidate based on the verification of one of the character arrangement verification by the arrangement verification unit 15 and the character string verification by the character string verification unit 16. You may make it do. In other words, the classification system may not require the verification of both the character arrangement verification by the arrangement verification unit 15 and the character string verification by the character string verification unit 16 as an essential configuration.

  The recognition unit 13 may select an optimum character string candidate from each character string candidate based on the similarity total value of each character recognition candidate string corresponding to each character string candidate. The recognition unit 13 may select an optimum character string candidate from each character string candidate based on at least one of the similarity total value, character arrangement verification, and character string verification.

  In the first and second embodiments, it has been described that character string verification (S5) and character string verification (S6) are executed after generation of character string candidates (S4). However, the character string recognition process may be executed as follows. For example, the character string candidates may be generated so that the character arrangement verification condition and the character string verification condition are satisfied.

  In the first and second embodiments, the character string recognition process in the case where the lower end of the character is hidden has been described. However, the character recognition process in the case where one or more of the upper end, the lower end, the right end, and the left end of the character are hidden can be realized as follows.

  For example, the character recognition dictionary database 14 has ((N × 4) -3) character recognition dictionaries D1, D21, D22, D23, D24, D31, D32, D33, D34,..., DN1, DN2, DN3, DN4 A (N: natural number) character recognition dictionary is stored.

  The character recognition dictionary D1 is a character recognition dictionary that is generated from a plurality of characters that have no missing characters. The character recognition dictionary D21 (10% missing dictionary) is a character recognition dictionary generated from a plurality of characters lacking the upper end of 1% to 20% characters. The character recognition dictionary D22 (10% missing dictionary) is a character recognition dictionary generated from a plurality of characters lacking the lower end of 1% to 20% characters. The character recognition dictionary D23 (10% missing dictionary) is a character recognition dictionary generated from a plurality of characters lacking the right end of 1% to 20% characters. The character recognition dictionary D24 (10% missing dictionary) is a character recognition dictionary generated from a plurality of characters with the left end of 1% to 20% characters missing.

  The character recognition dictionary D31 (30% missing dictionary) is a character recognition dictionary generated from a plurality of characters in which the upper end of 21% to 40% characters is missing. The character recognition dictionary D32 (30% missing dictionary) is a character recognition dictionary generated from a plurality of characters with the lower end of 21% to 40% characters missing. The character recognition dictionary D33 (30% missing dictionary) is a character recognition dictionary generated from a plurality of characters from which the right ends of 21% to 40% characters are missing. The character recognition dictionary D34 (30% missing dictionary) is a character recognition dictionary generated from a plurality of characters with the left end of 21% to 40% characters missing.

  The recognizing unit 13 is based on the character recognition dictionaries D1, D21, D22, D23, D24, D31, D32, D33, D34,... Is elected. The arrangement verification unit 15 verifies the arrangement of each character candidate based on the detection result of one or more of the upper end, lower end, right end, and left end of each character candidate.

  As described above, even if it is unknown which direction of the character is hidden, a highly accurate character string recognition process can be realized.

  In the first and second embodiments, the arrangement verification unit 15 estimates the missing degree of each character candidate, verifies the arrangement of the character string candidates, and rejects inappropriate character string candidates. explained. Further, the arrangement verification unit 15 estimates the missing degree of each character candidate, estimates the size of each character candidate, verifies the arrangement of the character string candidates from these estimation results, and rejects inappropriate character string candidates. You can also

  As described above, the classification system according to the first and second embodiments can recognize a character string with high accuracy even when the height of a hidden character line is different from the height of other character lines. In addition, a character string can be recognized with high accuracy even when there is a large amount of missing characters. In addition, the sorting system according to the first and second embodiments can recognize a character string with high accuracy even when the degree of hiding of the character is different for each character. Thereby, the sorting system of the first and second embodiments can sort the sorting object with high accuracy.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1 ... Classification system, 11 ... Conveyance part, 12 ... Reading part, 13 ... Recognition part, 14 ... Character recognition dictionary database, 15 ... Arrangement verification part, 16 ... Character string verification part, 17 ... Character string database, 18 ... Classification part

Claims (9)

Detecting means for detecting each character candidate from the image;
Recognition means for recognizing each character candidate based on a plurality of character recognition dictionaries corresponding to a plurality of different character missing degrees;
A character recognition device.   The recognition means selects one or more character recognition candidates corresponding to each character candidate based on a plurality of character recognition dictionaries, generates a plurality of character string candidates by combining the character recognition candidates, and verifies the combination of character recognition candidates. The character recognition device according to claim 1, wherein an optimum character string candidate is selected based on the result, and a recognition result of each character candidate corresponding to the optimum character string candidate is output. First verification means for verifying a combination of character recognition candidates based on a character string database storing a plurality of character string patterns;
The character recognition apparatus according to claim 2, wherein the recognition unit outputs an optimum character string candidate based on a verification result by the first verification unit. A second verification unit that estimates a degree of character missing of each character candidate and verifies a combination of character recognition candidates based on the estimated missing degree of each character candidate;
The character recognition device according to claim 2 or 3, wherein the recognition means outputs an optimum character string candidate based on a verification result by the second verification means.   The character recognition device according to claim 4, wherein the second verification unit estimates a missing degree of one of the upper end, the lower end, the right end, and the left end of each character candidate.   The recognition means selects one or more character recognition candidates for each character candidate, generates a plurality of character string candidates by combining the character recognition candidates, and selects an optimum character string candidate based on the similarity of each character recognition candidate The character recognition device according to claim 1, wherein:   A sorting apparatus comprising the character recognizing device according to any one of claims 1 to 6, and further comprising a sorting unit that sorts a sorting object based on a recognition result of each character candidate.   7. The image recognition apparatus according to claim 1, further comprising: reading an image from a classification object and receiving the image transmitted from a classification processing unit that classifies the classification object based on classification information. And a classification control device comprising a communication means for transmitting the classification information corresponding to the recognition result of each character candidate to the classification processing unit. Detect each character candidate from the image,
A character recognition method for recognizing each character candidate based on a plurality of character recognition dictionaries corresponding to a plurality of different character missing degrees.

Images