JP2000348142A - Character recognizing device, method therefor and recording medium for recording program executing the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize quick word collation of high precision at the time of recognizing a character string included in a picture of a signboard or the like. SOLUTION: A picture including character information like an enterprise name, a telephone number, or a personal name is inputted (2), and character pictures are segmented from the inputted picture, and segmented character pictures are recognized to output a recognition result character code string (3 to 5). Position information of the place where the picture was inputted is acquired (6), and the range of information in a word dictionary 1 used for word collation is narrowed down on the basis of acquired position information (7). Narrowed-down information in the word dictionary 1 is used to collate words in the word dictionary 1 with the recognition result character code string (8), and a word of which the number of characters coinciding with the recognition result characters is largest as the word collation result is outputted as the character recognition result (9).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition technology for performing effective word matching for recognizing a company name, a telephone number, a personal name and the like in a scene image input outdoors.

[0002]

2. Description of the Related Art Since a high character recognition accuracy cannot be expected for an outdoor object such as a signboard, it is difficult to uniquely narrow down the words by the conventional word matching method. Therefore, for example, "Araki, Miyamoto, Suzuki, Nakamura,
Sugimura, "Signboard Character Recognition by Integration of Multiple Information", 199
8 IEICE D-12-25 ", Japanese Patent Application No. 10-222.
No. 913 (character recognition device) has been proposed that performs word matching by integrating the character recognition results of multiple character strings. However, when the size of a word dictionary is large,
In some cases, narrowing down becomes difficult.

[0003]

Therefore, in order to improve the effect of the word matching process when using a low-accuracy character recognition result, in order to reduce the number of erroneous matching character strings due to erroneously recognized characters, a word dictionary is required. It is necessary to narrow down.

The present invention has been made in view of the above circumstances, and has as its object the use of a word dictionary including position information and position information on a place where an image has been acquired, thereby realizing low quality. An object of the present invention is to provide a character recognition device that realizes an effective word matching method for character recognition results and obtains correct words with high accuracy.

[0005]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention focuses on the use of location information, and uses a word dictionary linked to the location information to form a word dictionary used for word matching. Narrow the range and improve word matching accuracy.

As means for this, a word dictionary in which company names, telephone numbers, personal names and addresses are stored in pairs, and an image input section for inputting images containing company names and telephone numbers, are provided. A character string cutout unit that cuts out a character string area from the extracted image, a character cutout unit that cuts out a character image for each character string in a character image area for all character strings, and a recognition result character that performs recognition of the cutout character image. A character recognition unit that outputs a code sequence, and position information of a place where an image is input are stored in a PHS (Personal Handyphone System) or GP.
A position information acquisition unit that acquires using the S (Global Positioning System), and a word dictionary that narrows down the position information in the word dictionary based on the acquired position information and selects a part of the word dictionary used for word matching A word matching unit that performs word matching between a selecting unit, a character code string obtained by a character recognizing unit, and a word dictionary obtained by a word dictionary selecting unit, and a result obtained by the word matching unit. And an output unit for outputting a result having the largest number of matching characters.

The operation of the present invention is as follows. According to the present invention, by selecting a word dictionary using position information, it is possible to reduce erroneous matching by comparing less word dictionaries with character recognition results. In addition, even when a company name or an individual name having the same name exists, or when an attempt is made to recognize a telephone number without an area code, a correct answer can be obtained by using the location information.

[0008]

FIG. 1 is a block diagram showing an overall outline of a character recognition device according to an embodiment of the present invention. The character recognition device according to the present embodiment includes a word dictionary 1, an image input unit 2, a character string cutout unit 3, a character cutout unit 4, a character recognition unit 5, a position information acquisition unit 6, a word dictionary selection unit 7, a word collation unit. And an output unit 9.

The word dictionary 1 stores a company name and an address, a telephone number and an address, or a personal name and an address in pairs. As the word dictionary 1, for example, a telephone directory database such as a town page may be used.

The image input unit 2 inputs an image including a company name, a telephone number, and a personal name. This is an image including plate-type character information such as a signboard, a station signboard, and a poster. The character string cutout unit 3 cuts out a character string image from the image input by the image input unit 2. A chunk of characters arranged in a vertical or horizontal line with small variations in character size is cut out as one character string. Here, for example, a character string such as an advertiser name, telephone number, address, etc. written on a signboard for a station sign, and a character string extending over multiple lines such as a poster title and contact information for a poster are all cut out. .

The character cutout unit 4 cuts out character images one by one for each of the character string images cut out by the character string cutout unit 3, and calculates the character size.
The character cutout unit 4 further cuts out a character image in units of one character from the rectangular area of the character string image cut out based on the contrast with the background or the like.
Next, the character recognizing unit 5 recognizes each of the cut-out character images for all the character strings, and outputs a recognition result character code string.

As a method used in the character string extracting section 3, the character extracting section 4, and the character recognizing section 5, for example, the density contrast between the character line and its background is high, and the circumscribed rectangle of the character is often close to a square. A method of extracting and recognizing characters from scene images with noisy conditions and drastic changes in conditions by using the universal properties of characters has been proposed (Jun Otani, Akio Shio, "Character Patterns from Scene Images" Extraction and recognition ”, IEICE Transactions D Vol.J71-D
No.6 pp.1037-1047).

The character string cutout unit 3 may be omitted, and characters may be cut out directly from the scene image input by the image input unit 2. This includes, for example, methods using edge features (Kuwano, Kurakake, Odaka, “Ticker character extraction method for video data search”, IEICE Technical Report PMRU96,
93-103, pp.39-46) can be applied.

The position information acquiring section 6 acquires the position information of the point where the image is input by using PHS or GPS. Since the position information includes an error, the maximum value of the error is also detected. In addition, latitude and longitude information is acquired as position information.

Next, the word dictionary selecting section 7 selects a portion to be used for collation in the word dictionary 1 using the position information acquired by the position information acquiring section 6 and the range of the error. As a result, the number of records used for word matching can be narrowed down to tens or hundreds of records from many records.

The word dictionary 1 is selected, for example, by converting address information in the word dictionary 1 into latitude / longitude information and using the latitude / longitude information detected by the position information acquisition unit 6 to obtain a latitude / longitude including an error. For example, there is a method of selecting a dictionary in a range falling within the range. For example, the Internet version of the town page database is provided with latitude and longitude information (Kenichi Shima, Katsumi Takahashi, Nobuyuki Miura, "Building an Internet Multimedia Phonebook", Online Proceedings Japan World Wide Web Conference
'95 (1995)), by using it as the word dictionary 1, it is possible to narrow down the word dictionary information using the latitude and longitude information.

The word matching section 8 performs word matching using the character code string obtained by the character recognition section 5 and the word dictionary obtained by the word dictionary selection section 7 and counts the number of matching characters. For example, when the recognition target is a signboard of one character string, word matching with a company name in a word dictionary is performed. In addition, in the case of a recognition target having a plurality of character strings, word matching is performed between the plurality of character strings and a plurality of records. Find the combination Alternatively, a recognition result of a character string to be subjected to word matching using format analysis or the like may be selected, and the number of matching characters may be calculated.

Here, as a method of selecting a character string to be subjected to word collation, "Araki, Miyamoto, Suzuki, Nakamura, Sugimura," Signboard Character Recognition by Integrating Multiple Information ", 1998 IEICE Society Conference D-12-25. ], A method utilizing the features of a signboard can be used. As word matching methods, for example, "Tadashi Kitamura, Kiyoshi Nakabayashi, Akitaka Omitsu, Osamu Nakamura," Handwritten character string processing method using word knowledge ",
NTTR & D, 39, 3, pp 429-436 (1990) ”can be used to allow a shift due to a character extraction error.

The output unit 9 outputs a record having the largest number of matching characters counted by the word matching unit 8 as a word matching result. For example, by outputting a company name and inputting it to a search engine such as the Internet, information about the company can be obtained. Also, by giving the telephone number and the address information under the AND condition, it is possible to further narrow down the information to be acquired.

The operation and operation of the thus configured character recognition will be described. FIG. 2 is a flowchart showing the operation of the character recognition device shown in FIG.

First, in step 20, a character image in which a company name, a telephone number, an address, a personal name and the like are written is read.
In step 21, a character string is cut out. Step 22
Then, a character image of each character is extracted from the extracted character string image, and character recognition is performed in step 23.

FIG. 3 shows an image of extracting a character string from an input image. The image input unit 2 inputs an image captured and captured such that one signboard is inserted for a signboard and one poster is inserted for a poster, as in a camera image 11 shown in FIG. 3, for example. The character string cutout unit 3 cuts out a rectangular area, which is a lump of characters, as a character string from such a camera image 11 by contrast with the background or the like. As a result, two character strings 12-1 and 12-2 are cut out from the camera image 11 in FIG.

The character extracting section 4 includes two character strings 12-
Each of the characters 1 and 12-2 is separated into one character unit, and the separated character image 13 is passed to the character recognition unit 5. The character recognition unit 5 performs character recognition by pattern matching or the like. As a result, the recognition result 14-
1, 14-2 are obtained.

In the above processing, the processing of the character string cutout unit 3 is omitted, and as shown in FIG.
Alternatively, the character image 15 may be directly extracted from 1 and may be recognized by the character recognition unit 5.

In step 24 of FIG. 2, the position information of the point where the character image is read is obtained. In step 25, the word dictionary 1 is selected based on the position information obtained in the previous step 24. Here, selecting the word dictionary 1 means selecting a record to be used in the word dictionary 1. When there are a plurality of word dictionaries 1, selecting an appropriate word dictionary 1 based on the position information is also included.

For example, a company name as shown in FIG.
When there is the word dictionary 1 having the telephone number, the address, and the latitude / longitude information, the word dictionary selection unit 7 performs the position measurement from the latitude / longitude information acquired at the time of capturing the image using the PHS or GPS by the position information acquisition unit 6. Are narrowed down to records having latitude / longitude information within a range included in the error, and a narrowed dictionary 31 is obtained.

If the record of the word dictionary 1 does not have the latitude and longitude information, as shown in FIG. 4B, an address information database (DB) having correspondence information between the address and the latitude and longitude information
32, the latitude / longitude information acquired by the position information acquisition unit 6 is converted into address information, and the word dictionary 1 is narrowed down using the converted address information.

Thus, for example, the entire town page database contains 25,000 records in the case of Yokosuka City, for example.
It is possible to reduce the size of a dictionary for performing word matching on 00 or several tens of cases, thereby achieving high speed and high accuracy.

In step 26, the selected word dictionary 1
Word matching is performed using (the narrowed-down dictionary 31) and the character code string obtained in step 23, and the number of matching characters as the word matching result for each record is calculated.
Finally, in step 27, a word matching result having a large number of matching characters is output as the maximum likelihood candidate, and the process ends.

FIG. 5 is a diagram for explaining an example of this word collation. For example, M _1: "Araki _{dental" M 2: "TEL: 12-3456} " M 3: " Opening hours: Monday to Saturday at AM10 o'clock ~PM15" three lines of string M1~M3 is a sign that was written It is assumed that as a result of recognizing the character string image cut out from the image by the character recognizing unit 5, the following C1 to C3 are obtained as recognition result character strings.

C ₁ : “Aramoto dentistry” C ₂ : “TFI: 17-8456” C ₃ : “Government time: Sunday-Saturday AM0 to PM15: 00” On the other hand, the position of the place where the image of this signboard was taken The dictionary 31 narrowed down based on the information, as shown in FIG.
Assuming that the records are R _{1 to} R ₄ , the word matching unit 8
Collates the recognition result string C ₁ -C _3, company names in each record in the dictionary 31, a telephone number, all the combinations of the addresses.

That is, the number of matching characters is calculated for each of the records R _{1 to} R ₄ selected by the word dictionary selecting section 7 as follows. Company name and the C _1, phone number and C _2, matches the address and C _3, to calculate the sum of the matched characters. Company name and the C _1, phone number and C _3, matches the address and C _2, to calculate the sum of the matched characters. The company name is compared with C ₂ , the telephone number with C ₁ , and the address with C _3, and the total number of matching characters is calculated. ......

As described above, word matching is performed for all combinations, and the word matching result of the combination having the largest total number of matching characters is defined as the word matching result of the record. Expressing this by the formula, word collating result record R ₁ is as shown in FIG. 5 (C).

[0034] Max (N ₁₁ and matched characters + N ₁₂ and matched characters + number of matched characters of N ₁₃ and C ₃ of the C ₂ of C _1, N ₁₁ and C
Number of matched characters + N ₁₃ of the _first matched characters + N ₁₂ and C ₃ and C ₂
Number of matched characters, match the number of characters N ₁₁ and matched characters + N ₁₃ and C ₃ of matched characters + N ₁₂ and C ₁ to C _2, ............) = n of
₁ record R ₂ words matching result n _2, word collating result n ₃ records R _3, it is also calculated similarly word collating result n ₄ records R _4.

The final word matching result of the signboard is shown in FIG.
As shown in (D), this is the record of the largest among n ₁ , n ₂ , n ₃ and n ₄ . This is output by the output unit 9 as a word matching result. Only the company name may be output, or the telephone number and the address may be output together.

As is clear from the above description of the embodiment, the present invention uses position information in order to specify a character string of a signboard or the like in which it is generally difficult to obtain a highly accurate character recognition result. , Realizing high-speed and high-precision word matching. As a result, items that cannot be uniquely identified without location information, such as chain stores, can be identified.
Therefore, word matching can be performed with higher accuracy than conventional word matching using a huge dictionary. In addition, not only character recognition, but also the target that the user has seen is specified, so that efficient information retrieval support using the result can be performed.

[0037]

As described above, according to the present invention,
By integrating the character recognition and the position information, it is possible to effectively narrow down the information to be output and displayed as compared with the case where only character recognition is used. Therefore, it is possible to efficiently support the information search by the user using the display information as a clue.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an overall outline of an embodiment of the present invention.

FIG. 2 is a flowchart showing an operation according to the embodiment of the present invention.

FIG. 3 is a diagram illustrating a process for extracting a character string from an input image.

FIG. 4 is a diagram illustrating selection of a word dictionary.

FIG. 5 is a diagram illustrating word matching.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Word dictionary 2 Image input part 3 Character string extraction part 4 Character extraction part 5 Character recognition part 6 Position information acquisition part 7 Word dictionary selection part 8 Word collation part 9 Output part

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Hidego Kuwano 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo F-term within Nippon Telegraph and Telephone Corporation (reference) 5B064 BA01 EA19 5J062 AA01 BB00 CC07 GG02 5K024 AA79 CC11 GG01 GG10 5K036 KK09 9A001 BB03 BB04 FF03 HH20 HH22 HH28 JJ19 JJ25 JJ78 KK37 KK42 LL02

Claims (3)

[Claims] An image input unit for inputting an image including character information such as a company name, a telephone number or a personal name; and a character image cut out from the image input by the image input unit.
A character recognition device including a recognition unit that recognizes the cut-out character image and outputs a recognition result character code string; a position information acquisition unit that acquires position information when an image is input by the image input unit; A word dictionary selection unit for narrowing down word dictionary information used for word matching based on the position information obtained by the position information obtaining unit; and a recognition unit using the word dictionary information narrowed down by the word dictionary selection unit. A word matching unit that performs word matching with the character code string output by the above, and an output unit that outputs, as a character recognition result, a word having the largest number of matching characters among the word matching results obtained by the word matching unit. A character recognition device characterized in that: 2. A step of inputting an image including character information such as a company name, a telephone number or a personal name, cutting out a character image from the input image, recognizing the cut-out character image, and recognizing a character as a recognition result. Outputting a code sequence;
A step of acquiring position information when the image is input; a step of narrowing down information in a word dictionary used for word matching based on the acquired position information; A character characterized by comprising a step of performing word matching between a word in a dictionary and the above-mentioned recognition result character code string, and a step of outputting, as a character recognition result, a word having the largest number of matching characters in the above word matching result. Recognition method. 3. A process of inputting an image including character information such as a company name, a telephone number or a personal name, cutting out a character image from the input image, recognizing the cut-out character image, and recognizing a character as a recognition result. A process of outputting a code sequence, a process of acquiring position information when the image is input, a process of narrowing down information in a word dictionary used for word matching based on the acquired position information, A process of performing word matching between a word in a word dictionary and the above-described recognition result character code string using dictionary information, a process of outputting a word having the largest number of matching characters in the above-described word matching result as a character recognition result, Recording a program for executing a character recognition method, characterized by recording a program for causing a computer to execute the method.