JPH08287188A - Character string recognition device - Google Patents

Abstract

PURPOSE: To provide a character string recognition device which can automatically recognize a character string, entered in free format with a character touch and entanglement, fast with high precision. CONSTITUTION: This device has an image input part 1, a format decision part 2, a character cutting part 3, a normalization part 4, a feature extraction part 5, a collation part 6, a knowledge processing part 8, a character generation part 10, a 1st means, and a 2nd means. At this time, the 1st means can be equipped with the normalization part 4, feature extraction part 6, and collation part 6. Further, the 2nd means can be equipped with the knowledge processing part 8. Recognition candidate characters 18 obtained by collating feature patterns 17 of an input character with reference patterns 19 by the collation part 6 are matched by the knowledge processing part 8 against knowledge dictionary data 20, and a character category 21 which is estimated to be desirable for an inconsistent part is outputted. A character generation part 10 generates a character image 23 corresponding to the character category 21 and the input character image 15 is matched again against the generated character image 23.

Description

Detailed Description of the Invention

[0001]

[Industrial field of use] Regarding the method for automatically recognizing character strings written on forms and postal items, there are no particular restrictions on the entry of character frames, etc.・
The present invention relates to a method for automatically recognizing a character string such as a name written in free format (indefinite pitch).

[0002]

2. Description of the Related Art In a character string recognizing device for reading a form which is limited in writing such as a character frame, it is relatively easy to cut out characters because adjacent characters rarely come into contact with each other. On the other hand, in a free-form character string recognition device that recognizes addresses, names, etc. written on forms and mails that have no restrictions on the entry of character frames, etc. Due to the combination, it is difficult to correctly estimate the area of one character, and it is often difficult to cut out the character. Therefore, in a character string recognition device intended for free format, it is necessary to deal with the difficulty of character segmentation due to contact and intricacy of these characters.

There are the following two conventional free-form character string recognition methods.

The first method is a method of obtaining a plurality of combinations of cutouts of a character string and selecting an appropriate combination of cutouts by matching with the knowledge information. This method is "character segmentation and recognition from handwritten character strings using language information" (Murase et al., IEICE Transactions, 86 /
9, Vol. J69-D, No. 9, pp1292-1
301, September 1986), and "Handwritten character string recognition using character concatenation information" (Kobayashi et al., IEICE Technical Report, PRU-91-67, pp39-46, 19).
October 1991).

Specifically, a basic block obtained by cutting a character string into excessive divisions is obtained, and a plurality of combinations of character cutting are obtained as a combination of basic blocks. Next, individual character recognition is performed on the integrated block (combination of basic blocks) corresponding to each combination of character cutouts. Finally, in the knowledge processing unit, the recognition candidate character that is the output of the individual character recognition corresponding to each integrated block is matched with the knowledge information regarding the character to be read, and the candidate with the highest matching is set as the reading result.

In the second method, the process of cutting out a character from a character string is performed only for a single set of candidates, but for a character string in which there is a possibility that the characters may come into contact with each other or the characters may be intricate, the contacting or This is a method in which feature data of complicated character patterns is registered as a recognition dictionary and collectively read in word units of two or more characters.

[0007]

In the first method, a correct cutout result may not always be included in the step of obtaining a basic block obtained by cutting out a character string in an excessively divided manner.

For example, when "large" and "tsuka" in the character string "Otsuka" are in contact with each other, it is possible to determine that the character string consists of two characters only after recognizing the character. It becomes possible to cut out as two characters. But,
In the first method, since the basic block is obtained from the character string before recognizing the character, it is difficult to obtain the compulsory cut-out point when the character strings are in contact with each other. Therefore, it is not always possible to extract a combination of “large” and “mound” as a basic block. In this case, the correct cutout result is not included.

As described above, in the first method, the probability of including the correct clipping result can be increased by increasing the number of basic blocks, but it is difficult to completely set the probability of including the correct clipping result to 1. is there. Further, when the number of basic blocks is increased, the number of integrated blocks, which is a combination of basic blocks, increases exponentially. Therefore, performing individual character recognition on all integrated blocks and further performing knowledge processing on the combination of each integrated block has a problem that the processing amount becomes extremely large.

Further, in the second method, it is not realistic to register all combinations of characters that come into contact with each other. In particular, when three or more characters come into contact with each other, the number of character combinations becomes enormous and the number of registrations becomes enormous.

In view of such a point, the present invention is a character string recognition capable of automatically recognizing a character string written in a free format having contact of characters and intricacy of characters with high speed and high accuracy. The purpose is to provide a device.

[0012]

SUMMARY OF THE INVENTION A character string recognition apparatus of the present invention comprises an image input section for inputting an image of a character string to be read which is written in free format, and a character string from the input image. A format determination unit that extracts a region and determines format information indicating the direction of whether the character string is vertical writing or horizontal writing, and a predetermined number of characters from the image of the character string region extracted by the format determination unit In the character cutout unit that cuts out the character pattern with, a normalization unit that performs a normalization process according to the character size on each of the character patterns cut out by the character cutout unit, and an input character obtained by the normalization unit. A feature extraction unit that extracts character features of a normalized pattern and generates a feature pattern of an input character, and a correspondence between the feature pattern of the input character and a reference pattern stored in a recognition dictionary memory. The matching unit that outputs the recognition candidate character based on the result of the matching and the recognition candidate character and the knowledge dictionary data of the character to be read stored in the knowledge dictionary memory are matched to obtain a match. Using the knowledge processing unit that outputs a character category estimated to be desirable for a non-existing portion and the character image data stored in a character image memory that stores one character image in one character category, the knowledge processing is performed. A character generation unit that generates a character image corresponding to the character category output from the unit,
Normalization processing is performed on the character image generated by the character generation unit to obtain a normalized pattern of a generated character, character features of the normalized pattern of the generated character are extracted to obtain a characteristic pattern of the generated character, A first means for collating the characteristic pattern of the generated character with the characteristic pattern of the input character to obtain a recognition candidate character string, and the recognition candidate character string obtained as a result of collation and the knowledge dictionary data are matched. , Second means for outputting the character string having the highest consistency as a character string recognition result.

The character string recognition device of the present invention is the first character string recognition device.
The means may include the normalization unit, the feature extraction unit, and the matching unit.

In the character string recognition device of the present invention, the second means may include the knowledge processing section.

[0015]

According to the present invention having the above-described structure, only the characters that cannot be matched after the input character recognition processing is performed.
From a character category estimated to be desirable, a character image in which a plurality of characters touch each other, a character image in which one character is divided, or a character image in which both are combined is generated and collated with the input character image. It is possible to automatically and quickly recognize a character string written in a free format with contact between characters and intricacy of characters without any need for high speed.

Also, when three or more characters are in contact,
Even when characters are arranged in the horizontal direction, it is possible to generate a character image and match it with the input character image. High-speed writing of a character string written in free format with contact between characters and intricacy of characters. In addition, it is possible to automatically recognize with high accuracy.

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. In the present embodiment, a case will be described in which the recognition process is performed for each character.

The configuration of FIG. 1 will be described. FIG. 1 shows an image input unit 1, a format determination unit 2, a character cutout unit 3, a normalization unit 4, a feature extraction unit 5, a collation unit 6, a recognition dictionary memory 7, and a knowledge processing unit 8. The knowledge dictionary memory 9, the character generation unit 10, and the character image memory 11 are provided.

The image input unit 1 inputs an image of a character string 12 written in a free format on a form or postal matter to be read and performs a process of generating image data. In particular,
The reflected light of the light emitted from the light source is transferred to the CCD (Charge).
An image is formed by an imaging device such as a Coupled Device), and this is scanned and converted into an electric signal. Next, the converted electrical signal is quantized to generate multilevel full-screen image data 13.

The format determination unit 2 obtains a rectangular feature circumscribing the black pixel block in the whole image data 13 generated by the image input unit 1, a histogram feature of the vertical and horizontal axes of the black pixel, and the like. And by the distribution of these features,
The format information 22 indicating the vertical writing / horizontal writing directions of the character string is determined, and the character string area image 14 is generated.

The character slicing unit 3 estimates one character region of the character string region image 14 generated by the format determining unit 2 by obtaining circumscribing rectangular features and histogram features of vertical and horizontal axes of black pixels. The input character image 15 is cut out. However, a character string written in a free format such as an address / name on a form or post without a character frame 12
In the character cutout process, the one character area cannot always be estimated accurately. A common example is
There is a case where an area of two or more characters is erroneously estimated as a one-character area, and an area smaller than one character is erroneously estimated as a one-character area. Input character image 1 in this embodiment
5 is defined as an image of an area that is estimated and cut out as a one-character area. Therefore, in the input character image 15, one character area may be accurately cut out, or an error may be cut out.

The normalization unit 4 performs a normalization process for aligning the character image sizes of the input character image 15 cut out by the character cutting unit 3 and the generated character image 23 described later.
As a method of normalization processing for making the image sizes uniform, there are a linear normalization method of enlarging or reducing in the vertical direction or the horizontal direction at an equal ratio, a non-linear normalization method of changing the ratio of enlarging or reducing depending on the region, and the like. In this embodiment, any normalization method can be applied.

The feature extraction unit 5 extracts the character features of the normalized input character image 16 and the normalized generated character image 24 generated by the normalization unit 4. Character features to be extracted include directional features of character edges, character stroke density features, and structural analysis features. In this embodiment, any feature extraction method can be applied.

The collating unit 6 is a feature pattern 1 of the input character image.
7 and the reference pattern 19 registered in the recognition dictionary memory 7
By checking whether or not the distance value, the degree of similarity, or the characteristic component belonging to the reference pattern 19 exists in the characteristic pattern 17 of the input character image, and the characteristic belonging to the characteristic pattern 17 of the input character image. The degree of matching is determined by checking whether or not the component exists in the reference pattern 19, and the recognition candidate character 18 is output. This process is defined as the first collation process.

The recognition dictionary memory 7 receives the learned character image data as input, performs a normalization process by the normalization method used by the normalization unit 4, and extracts a feature pattern extracted by the feature extraction method used by the feature extraction unit 5. The category code pair to which the learning character image data belongs is registered as a reference pattern 19.

The knowledge processing unit 8 uses the recognition candidate character 18 obtained by the collation unit 6 as input data, based on the position of the input character image 15 in the character string area image 14,
The recognition candidate characters 18 are arranged to generate a recognition candidate character string.
A knowledge database to be read is registered in the knowledge dictionary memory 9 in advance. As the knowledge database, there are various kinds of knowledge such as an address, a name, a company name, a product name, a part name, and the like. In the present embodiment, the case of using the address knowledge database will be described. Knowledge processing unit 8
In (1), the recognition candidate character string is searched based on the knowledge data 20 registered in the knowledge database to be read.

Character extraction by the character extraction unit 3, normalization by the normalization unit 4, feature extraction by the feature extraction unit 5, collation by the collation unit 6, and knowledge processing unit 8
As a result of performing the knowledge processing in, the knowledge processing unit 8 outputs a character category 21 considered to be desirable for a portion that is not consistent with the knowledge data 20 to be read.

The character generation unit 10 corresponds to the character category 21 output from the knowledge processing unit 8 based on the character image stored in the character image memory 11 and the format information 22 output from the format determination unit 2. The generated character image is generated and output as the generated character image 23.

The generated character image 23 output from the character generation unit 10 is input to the normalization unit 4, and a normalized generated character image 24 is obtained. Next, the characteristic extraction unit 5 executes the characteristic extraction of the normalized generated character image 24 to obtain the characteristic pattern 25 of the generated character image.

Next, the collation unit 6 performs collation processing again. The characteristic pattern 17 of the input character image and the characteristic pattern 25 of the generated character image are associated with each other, and the distance value, the degree of similarity, or the characteristic component belonging to the characteristic pattern 25 of the generated character image is the characteristic pattern 17 of the input character image. Is checked, whether the characteristic component belonging to the characteristic pattern 17 of the input character image exists in the characteristic pattern 25 of the generated character image, the degree of matching is determined, and the recognition candidate character is determined. 18 is output. This process is defined as the second collation process.

The collation unit 6 also has means for comparing the result of the first collation processing with the result of the second collation processing. This is because the degree of matching between the characteristic pattern 17 of the input character image and the best matching reference pattern 19 in the recognition dictionary memory 7 and the characteristic pattern 17 of the input character image and the generated character in the first matching process. This is a function of comparing the degree of matching with the characteristic pattern 25 of the image and outputting the recognition candidate character 18 having a good degree of matching. As a concrete comparison method of the comparison processing result, the average value of the first distance value and the second value
A method of comparing the average value of the distance values for the first time and outputting the recognition candidate character 18 having the smaller average value of the distance values, or an average value of the first similarity and an average of the second similarity. The recognition candidate character 18 having the larger average similarity value is compared with the value.
There is a method of outputting.

The knowledge processing unit 8 uses the recognition candidate character 18 having a good degree of matching, which is obtained by comparing the result of the first matching process and the result of the second matching process in the matching unit 6, as input data. Then, the recognition candidate character string is searched again based on the knowledge data 20 registered in the knowledge database to be read, and the result most matched with the knowledge data 20 is output as the character string recognition result 26.

FIG. 2 is a diagram for explaining the processing procedure in one embodiment of the present invention, and shows the case where the character string "Otsuka Nitta, Kawagoe City" has been input. An address database is used as the knowledge to be read.
In the address database, as address data, "Sawashina, Kawagoe City", "Amanuma Nitta, Kawagoe City", "Otsuka Nitta, Kawagoe City"
Suppose there is. In addition, in the recognition dictionary memory 7,
"River", "Yoshi", "City", "Sand", "New", "Rice", "Heaven", "Swamp"
It is assumed that there are 13 types of reference patterns 19 “large”, “mound”, “loan”, “right”, and “small”.

The operation in the configuration of FIG. 1 will be described with reference to FIG. In FIG. 2, since two characters “Large” and “Tsuka” in the input character string are in contact with each other, it is difficult to separate the two characters “Large” and “Tsuka” in the character cutout unit 3, The input character image 15 is cut out as one character “Otsuka”. Therefore, the input character image 15 generated by the character cutout unit 3 has six character blocks 31 to 31.
It becomes 6.

The character recognizing unit including the normalizing unit 4, the feature extracting unit 5, and the collating unit 6 recognizes the six character blocks 31 to 36. In the matching unit 6, the feature pattern 17 of the input character image and the 13 of the recognition dictionary memory 7 are input.
The collation processing is performed on the reference pattern 19 of the type, and the recognition candidate character 18 of “Romashinden, Kawagoe City” is obtained. However, the address of Kawagoe City in the address database has place names such as "Sunashinden", "Amanuma Shinden", "Otsuka Shinden", etc., but not "Lomminden". Therefore, according to the knowledge data 20 of the knowledge processing unit 8, the character category 21 considered to be desirable as the character of the character block 34 is
It is estimated to be either "sand", "Amanuma", or "Otsuka".

The character generator 10 generates a character image corresponding to the character code. The generated character image 37 in which “heaven” and “swamp” are vertically arranged and the generated character image 38 in which “large” and “tsuka” are vertically arranged are obtained from the information that the format determination unit 2 indicates vertical writing. . For the one-character image of "sand",
Since it has already been verified in the first verification step, it is not necessary to generate it again for verification.

Next, in the normalization section 4, the generated character images 37 and 38 are normalized, and the normalized generated character image 3
9, 40 are obtained. In the feature extraction unit 5, the normalized generated character images 39 and 40 are characteristic-extracted, and the characteristic pattern 25 of the generated character image is obtained.

After that, in the collation unit 6, the second feature between the feature pattern obtained by normalizing the character block 34 and extracting the feature and the normalized feature pattern of the generated character images 39 and 40 is generated.
The second matching process is performed. Further, in the matching unit 6,
By comparing the result of the first collation process with the result of the second collation process, a result having a good degree of matching is obtained as the recognition candidate character 18.

In FIG. 2, the matching between the generated character image 40 obtained by normalizing "Otsuka" which is the result of the second collation processing and the character block 34 is the reference pattern which is the result of the first collation processing. When it is determined that the matching is better than the matching with “Loan”, “Otsuka Nitta, Kawagoe City” is obtained as the recognition candidate character 18.

In the knowledge processing section 8, since the address data "Otsuka Nitta" exists in the address database, "Kawagoe City Otsuka Nitta" is obtained as the character string recognition result 26.

FIG. 3 is a diagram for explaining the processing procedure in one embodiment of the present invention, and shows the case where the character string "Kawagoe-shi Sunaden" is entered. As the knowledge to be read, an address database is used as in FIG. In the address database, it is assumed that there are “Sawashina, Kawagoe-shi”, “Amanuma Nitta, Kawagoe-shi”, and “Otsuka Nitta, Kawagoe-shi” as address data. In addition, in the recognition dictionary memory 7, "river", "existence", "city", "sand", "new", "field".
It is assumed that there are 13 types of reference patterns 19 of “heaven”, “swamp”, “large”, “mound”, “loam”, “right”, and “small”.

FIG. 4 is a diagram showing data in the character image memory in the configuration of FIG. As shown in FIG. 4, the character image memory 11 stores character categories 80 and character image data 83 in association with each other. It is also possible to add the attribute information 81 and 82 to each of the character image data 83 to 87. FIG. 4 shows a case in which attribute information 81 and 82 indicating whether the character image data 83 to 87 are separated vertically or horizontally are added. For example, "stand" is vertically aligned with "mu" and "ro".
Although it is separated into two partial patterns of and, but not in the horizontal direction, (vertical, horizontal) = (1,0). However,
The "swamp" is divided into two partial patterns, "shi" and "call" in the horizontal direction, but not in the vertical direction, so (vertical,
Horizontal) = (0, 1). Also, this attribute information 81, 8
2, it is also possible to add a vertical separation position, a horizontal separation position, etc., but as shown in FIG. 4, the partial pattern is simply set so that the separation position 86 is simply set to the half position of one character. The case where they are arranged will be described.

The operation in the configuration of FIG. 1 will be described with reference to FIG. In FIG. 3, one character “sand” in the input character string is separated into two character blocks, and the two character blocks exist apart from each other. It is difficult to integrate them as one character, and the input character image 15 is cut out as two characters of “right” and “small”. Therefore, the input character image 15 generated by the character cutout unit 3 has seven character blocks 51 to 57.
Becomes

The character recognizing unit including the normalizing unit 4, the feature extracting unit 5, and the collating unit 6 recognizes the seven character blocks 51 to 57. The collation unit 6 performs collation processing on the normalized characteristic patterns of the input character images 66 and 67 and the 13 types of reference patterns 19 of the recognition dictionary memory 7. In Figure 3, the reference pattern 6 for "Large" and "Tsuka"
4 shows the collation process of the input character images 66 and 67 with the normalized input character images 66 and 67. As a result of the matching process, the reference pattern that is most matched with the normalized input character image 66 is “right”, the reference pattern that is most matched with the normalized input character image 67 is “small”, and “Kawagoe City Right Koshin The recognition candidate character 18 "Ta" is obtained. However, the Kawagoe city address in the address database has place names such as "Sunashinden", "Amanuma Nitta", and "Otsuka Nitta", but not "Right Koshinden". Therefore, according to the knowledge data 20 of the knowledge processing unit 8, the character category 21 that is considered to be desirable as the characters of the character blocks 54 and 55 is estimated to be any of “sand”, “Amanuma”, and “Otsuka”.

The character generator 10 generates a character image corresponding to the character code. "Large""Tsuka" and "Ten""Swamp"
The two-character image of No. need not be collated again because it has already been collated in the first collation stage. "sand"
Since the single character image of No. is not collated, the generated character image “sand” 59 is horizontally divided into two generated character images “stone” and “small” based on the information that the format determination unit 2 is in horizontal writing. To be

Next, the normalization unit 4 normalizes the two divided generated character images to obtain normalized generated character images 60 and 61. In the feature extraction unit 5, the normalized generated character images 60 and 61 are feature-extracted, and the characteristic pattern 25 of the generated character image is obtained.

After that, in the collation unit 6, the second collation processing is performed between the characteristic pattern obtained by normalizing the character blocks 54 and 55 and extracting the characteristic, and the normalized characteristic pattern of the generated character images 60 and 61. Done. In the matching unit 6, the result of the first matching process and the result of the second matching process are compared, and a good matching result is obtained as the recognition candidate character 18.

In FIG. 3, generated character images 60 and 6 obtained by normalizing "stone" and "small" which are the results of the second collation processing.
The matching between 1 and the normalized input character images 66 and 67 is the reference pattern “right” which is the result of the first matching process.
It is determined that the matching is better than the matching with “small”, and “Kawagoe-shi Sunaden” is obtained as the recognition candidate character 18.

In the knowledge processing unit 8, since the address data "Sanshinden" exists in the address database, "Kawagoe-shi Sanshinden" is obtained as the character string recognition result 26.

[0051]

As described above, according to the present invention, after the input character recognition processing is performed, a character image is generated from a character category that is estimated to be desirable only for a character that does not match, and an input character image is generated. By the collation, there is an effect that a character string written in a free format including contact of characters and intricacy of characters can be automatically recognized at high speed and with high accuracy without performing unnecessary processing.

When three or more characters are in contact,
Even when characters are arranged in the horizontal direction, it is possible to generate a character image and collate it with the input character image. This has the effect that automatic recognition can be performed with high accuracy.

[Brief description of drawings]

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

FIG. 2 is a diagram illustrating a processing procedure in an embodiment of the present invention.

FIG. 3 is a diagram illustrating a processing procedure in an embodiment of the present invention.

FIG. 4 is a diagram showing data in a character image memory in the configuration of FIG.

[Explanation of symbols]

1 image input unit 2 format determination unit 3 character cutout unit 4 normalization unit 5 feature extraction unit 6 collation unit 7 recognition dictionary memory 8 knowledge processing unit 9 knowledge dictionary memory 10 character generation unit 11 character image memory 12 written in free format Character string 13 Full surface image data 14 Character string area image 15 Input character image 16 Normalized input character image 17 Characteristic pattern of input character image 18 Recognition candidate character 19 Reference pattern 20 Knowledge data 21 Character category 22 Format information 23 Generated character image 24 Normalized generated character image 25 Characteristic pattern of generated character image 26 Character string recognition result 31, 32, 33, 34, 35, 36 Character block 37, 38 Generated character image 39, 40 Normalized generated character image 51, 52, 53, 54, 55, 56, 57 Character block 58, 62, 63 Knowledge processing result sentence Character string 59 Generated character image 60, 61 Normalized generated character image 64, 65 Reference pattern 66, 67 Normalized input character image 80 Character category 81, 82 Attribute information 83, 84, 85, 87 Character image data 86 Separation position

Claims (3)

[Claims] 1. An image input unit for inputting an image of a character string to be read, which is written in a free format, and a character string area is extracted from the input image to determine whether the character string is vertically written. A format determination unit that determines the format information indicating the direction of horizontal writing, a character cutout unit that cuts out a character pattern in units of a predetermined number of characters from the image of the character string area extracted by the format determination unit, and the character cutout. A normalization unit that performs a normalization process according to the character size on each of the character patterns cut out by the unit, and a character feature of the normalization pattern of the input character obtained by the normalization unit is extracted to extract the input characters. A feature extraction unit that generates a feature pattern is associated with the feature pattern of the input character and a reference pattern stored in the recognition dictionary memory, and recognition is performed based on the association result. A matching unit that outputs a complementary character, and the recognition candidate character and the knowledge dictionary data of the character to be read stored in the knowledge dictionary memory are matched, and the character category estimated to be desirable for the part that does not match is determined. Using the knowledge processing unit to output and the character image data stored in the character image memory that stores one character image in one character category, the character processing unit corresponds to the character category output from the knowledge processing unit. A character generation unit that generates a character image, normalizes the character image generated by the character generation unit to obtain a normalized pattern of the generated character, and extracts character characteristics of the normalized pattern of the generated character. A first means for obtaining a recognition candidate character string by collating the characteristic pattern of the generated character with the characteristic pattern of the input character, and the result of collation. And the recognition candidate character strings obtained by taking the matching between the knowledge dictionary data, and a second means for outputting the most consistent string as a character string recognition result, the character string recognition unit. 2. The character string recognition device according to claim 1, wherein the first means includes the normalization unit, the feature extraction unit, and the matching unit. 3. The character string recognition device according to claim 1, wherein the second means includes the knowledge processing section.