JP2996823B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognizing device for cutting out characters written in predetermined places from image data read by an image scanner or image data transmitted by a facsimile machine, and performing character recognition processing.

[0002]

2. Description of the Related Art Conventionally, an OCR (optical character reader) has been known as a device for inputting necessary data to a computer. In OCR, a recognition target such as a character, a symbol, or a mark written in a predetermined position of a form is recognized by character recognition processing or position detection processing, and is converted into predetermined data that can be input to a computer. In recent years, in addition to directly inputting a form into an OCR, a system that uses a facsimile machine or the like to cut out a recognition target from image data of a form transmitted from a remote place and perform recognition processing has also been used. Was.

[0003] In a general character recognition device, a character information portion is detected from image data, cut out, and then character recognition and word recognition are performed to identify a recognition target. . In this case, if even one character cannot be specified, word recognition becomes impossible.

On the other hand, for example, in a postal code reader,
Like the postal code and the address, two types of character information that complement each other are used to identify the recognition target. If the postal code cannot be read, the postal code is used by using the address recognition result. A method has also been proposed in which the reading accuracy of the postal code is increased by specifying (Japanese Patent Laid-Open No. 2-242389).

[0005]

However, in the above-mentioned conventional system, it is a condition that one of the two types of character information can be reliably recognized. Therefore, 2
When both types of character information cannot be read, there is a problem that the recognition also fails.

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem. Even when any of a plurality of types of character strings that complement each other when specifying a recognition target cannot be read, character string recognition is performed. It is an object of the present invention to provide a character recognition device that can make character recognition possible and can significantly improve a character string recognition rate.

[0007]

SUMMARY OF THE INVENTION A character recognition apparatus according to the present invention performs character recognition processing on a plurality of types of character string images which are mutually complemented when specifying a recognition target, and performs recognition with a certain degree of certainty or more. Character recognition processing means for outputting an incomplete recognition result of a character string including the extracted character and information on its position, and each type of sentence that can be recognized
Compensation by storing character strings in association with other types of character strings
A dictionary and the same type of character string including the recognized character at a recognized position based on the incomplete recognition result of each type of character string obtained by each of the character recognition processing means are included in the supplementary dictionary.
And a supplementary processing means for mutually extracting character string candidates of other types of character strings corresponding to these from the supplementary dictionary and supplying the extracted character string candidates to the respective character recognition processing means. Is characterized in that the recognition processing is repeated until the complete recognition result of the character string is determined while narrowing the range of the recognition target by the character string candidate supplied from the compensation processing means.

Further, another character recognition device according to the present invention comprises:
A plurality of character recognition processing means for respectively performing character recognition processing on a plurality of types of character string images which are mutually complemented when specifying a recognition target; and an incomplete character string of each type obtained by each of the character recognition processing means. And a compensation processing table for obtaining a complete recognition result of each type of character string from the combination information of the recognition results.

[0009]

According to the present invention, when a plurality of types of character string images which are mutually complemented when a recognition target is specified are recognized by a plurality of character recognition processing means, complete recognition is performed by the character recognition processing means. Even when the result is not obtained, some character string candidates for other types of character strings are mutually extracted from the incomplete recognition result. Then, each character recognition processing means narrows down the character strings recognized by itself by mutually referring to these character string candidates obtained from the incomplete recognition results of the other character recognition processing means. Finally, a complete recognition result can be obtained. Therefore, according to the present invention, even when only incomplete recognition results can be obtained from individual character recognition processing means, a complete recognition result is finally obtained due to the mutual complementarity of a plurality of types of character strings. It becomes possible.

According to another character recognition apparatus of the present invention, a complementary processing table for obtaining a complete recognition result of each type of character string at once from combination information of incomplete recognition results regarding a plurality of types of character strings. , The recognition result can be obtained at high speed.

Here, the plurality of character recognition processing means does not necessarily need to be physically provided in plurality.
When the recognition processing of each type of character string is executed in order, this means that each character string recognition processing step corresponds to a character recognition processing unit.

[0012]

Embodiments of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a functional block diagram showing a character recognition device according to one embodiment of the present invention. Image data obtained by reading a predetermined form with an image scanner (not shown) is supplied to the character detection cutout unit 1.
The character detection cutout unit 1 detects and cuts out a character from the projection of the specific area of the image data in the X and Y directions. Among the image data cut out by the character detection cutout unit 1, a first type of character string is supplied to a character recognition unit 2, and a second type of character string is supplied to a character recognition unit 3. The character recognition units 2 and 3 respectively refer to the character recognition dictionaries 4 and 5 and
Recognize each character from the extracted character string image data. The character recognition results from the character recognition units 2 and 3 are supplied to word recognition units 6 and 7, respectively. Word recognition unit 6,
7 executes word recognition processing based on the character recognition result with reference to the word recognition dictionaries 8 and 9, respectively. Character recognition unit 2,
The character recognition dictionary 4, the word recognition unit 6, and the word recognition dictionary 8
The first character recognition processing means 15 constitutes the character recognition unit 3,
The character recognition dictionary 5, the word recognition unit 7, and the word recognition dictionary 9
The second character recognition processing means 16 is constituted.

The word recognition result of the word recognition unit 6 is supplied to a word compensation processing unit 10, and the word recognition result of the word recognition unit 7 is supplied to a word compensation processing unit 11. The word compensation processing unit 10 obtains character string candidates for the first type of character string from the incomplete word recognition result obtained by the word recognizing unit 6 with reference to the supplementation dictionary 12, and from these character string candidates. The specified second type of character string candidate is supplied to the character recognition dictionary 5, the word recognition unit 7, and the word recognition dictionary 9, respectively. Similarly, from the incomplete recognition result obtained by the word recognizing unit 7, the word compensation processing unit 11 refers to the compensation dictionary 12 to find a character string candidate for the second type of character string, and The first type of character string candidate specified from the candidate is supplied to the character recognition dictionary 4, the word recognition unit 6, and the word recognition dictionary 8, respectively.

The character recognizing units 2 and 3 and the word recognizing units 6 and 7 perform recognition processing again based on the supplied first type and second type character string candidates, and narrow down the character string candidates. I will go in.

Now, the first type of character string is a postal code,
Assuming that the type of character string is the name of a ward, city, town or village, this device performs the recognition process as follows. That is, assuming that the result of the first recognition performed by the character recognition unit 2 is "?? 7" and the result of the first recognition performed by the character recognition unit 3 is "???" With the recognition results alone, the word recognition units 6 and 7 cannot immediately obtain a desired recognition result. Therefore, the word compensation processing unit 11
A ward, municipal, and municipal name in which the second character is “river” in a three-character configuration is searched from the supplementary dictionary 12. As the search results, for example, “Tachikawa City”, “Akikawa City”, “Arakawa Ward” and “Shinagawa Ward”
Is obtained, the word compensation processing unit 11 next searches the word supplement dictionary 12 for a first type of character string candidate identified from these character strings, that is, a postal code. Assuming that the search results are “190”, “197”, “116”, and “142”, these are character string candidates of the first type.

The first type character string candidates "190", "197",... Thus obtained are supplied to the word recognition section 6. The word recognizing unit 6 recognizes “?? 7” and the character string candidate “1” supplied from the word replenishment processing unit 11.
From “90”, “197”, “116”, and “142”, it is recognized that the first type of character string is “197”.

Next, the word supplementation processing unit 10 reads the second type character string identified by the postal code “197”, which is the first type character string recognized by the word recognition unit 6, from the complement dictionary 12. A search is performed to obtain character string candidates of “Akigawa City” and “Fussa City”. This second type of character string candidate is supplied to the word recognition unit 7. In the word recognition unit 7, the second type of character string is “Akikawa city” from the recognition result “? Kawa?” And the character string candidates “Akikawa city” and “Fussa city” supplied from the word compensation processing unit 10. Is recognized. In this way, the first type of character string candidate and the second type of character string candidate complement each other with incomplete character recognition results of each type, so that a complete recognition result can be obtained from an incomplete recognition result. Obtainable.

In order to search the character string candidates at high speed in the word supplement processing units 10 and 11, an index file may be used for the supplement dictionary 12. That is, as shown in FIG. 2, only one character of the second type of character string is readable.
The index files 211, 212, 213, and
…, 21n are created and these index files 2
11 to 21n store indices and pointers to the corresponding character strings stored in the master database file 22. This makes it possible to immediately search for the first type of character string candidate from the second type of incomplete recognition result. A similar file configuration may be used when searching for a second type of character string candidate from a first type of recognition result.

In order to further increase the search speed, for example, as shown in FIG. 3, the first and second types of incomplete recognition results are input, and the first and second types specified by the combination information are input. It is sufficient to create a compensation processing table 31 that outputs a complete recognition result of the above.

In order to further increase the recognition rate, the character recognizing units 2 and 3 and the word recognizing units 6 and 7 perform the recognizing process again, including candidates that have not reached the predetermined matching level and are considered suspicious. It is possible. For example, the recognition result of each character of the character string in the character recognition unit 2 is shown in Table 1 below.
Suppose you are asked as follows.

[0021]

[Table 1]

In this case, the threshold output as the recognition result is
If it is 0.75, the recognition result output from the character recognition unit 2 is “?? 7” as described above. Here, for example, “190” as a character string candidate from the word
When "197", "116", and "142" are supplied,
For each string, calculate the sum of the matching levels. For example, the total of “190” is 1.05, and the total of “197” is 1.63. Then, the one with the highest numerical value is set as the recognition result. Thereby, the recognition rate can be further improved. Further, as a result of such recognition processing, when a tendency is observed that the matching level for a specific character is always low, the character recognition dictionaries 4 and 5 and the word recognition dictionaries 7 and 8 are updated so as to increase the matching level. You may do so.

In the above embodiment, the first type character string candidates "190", "197",... Are first extracted from the recognition result "? Kawa?" Of the second type character string. A second type of character string candidate may be extracted from the type character string recognition result “?? 7”. Which one to do first?
What is necessary is just to set suitably according to the number of candidates etc.

Further, for example, a postal code, an address, a reading,
When performing recognition processing using more types of character strings that complement each other, such as the area code of a telephone number, for example, as shown in FIG.
3. A system of character recognition processing means 481 to 48n comprising a word recognition section 44 and a word recognition dictionary 45 is provided for each type of character string, and more character string candidates fed back from the word compensation processing section 46 are used. The re-recognition process may be executed. As a result, more accurate recognition processing can be performed based on more information.

[0025]

As described above, according to the present invention, even when only incomplete recognition results can be obtained from individual character recognition processing means, a plurality of character string candidates obtained from these incomplete information can be obtained. The recognition process of each character string is complemented by each other, so that when identifying the recognition target, even if any of multiple types of character strings that complement each other cannot be read, character strings can be recognized. , And the character string recognition rate can be significantly improved.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a character recognition device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration example of a supplement dictionary in the device.

FIG. 3 is a functional block diagram showing another configuration example of the compensation processing unit in the apparatus.

FIG. 4 is a functional block diagram of a character recognition device according to another embodiment of the present invention.

[Explanation of symbols]

1, 41: character detection cutout section, 2, 3, 42: character recognition section, 4, 5, 43 ... character recognition dictionary, 6, 7, 44 ... word recognition section, 8, 9, 45 ... word recognition dictionary, 10,11,4
6 ... word supplement processing unit, 12, 47 ... word supplement dictionary, 15
.., First character recognition processing means, 16... Second character recognition processing means, 481 to 48n.

Claims (1)

(57) [Claims] 1. A string containing the characters identify when an image of a plurality of types of character strings mutually compensate each other recognized each character recognition process at a predetermined value or more certainty of the recognition target and the information of the position A plurality of character recognition processing means for outputting incomplete recognition results of each character string, and converting each type of character string that can be recognized into another type of character string
And a supplementary dictionary stored in association with the same type including the recognized character at a recognized position based on an incomplete recognition result of each type of character string obtained by each of the character recognition processing units. Referring to the character string in the compensation dictionary the character string candidates of other types of character strings corresponding to these
Compensation processing means for mutually extracting from the compensation dictionary and supplying each of the character recognition processing means to each other, wherein each of the character recognition processing means determines a range of a recognition target by a character string candidate supplied from the compensation processing means. A character recognizing device that repeats recognition processing while narrowing down until a complete recognition result of the character string is determined.