JPH07239916A - Printed character recognition device - Google Patents

Abstract

PURPOSE:To obtain a high recognition rate of printed characters regardless of the kinds of character font. CONSTITUTION:A dictionary number corresponding to a font used by each customer is registered in a disk storage device 3 of a work station 1. Upon the receipt of a printed matter from a customer, the document is read by an image reader 7 and the read image is displayed on the work station 1. Then, a part of the displayed document image is recognized tentatively by using a dictionary whose number corresponds to the customer, and when the result of correct recognition indicates 85% or over, the image is recognized regularly by using the dictionary. When the result of correct recognition indicates 85% or below or the dictionary number corresponding to the customer has not been registered yet, the regular recognition is implemented by using a total dictionary including the entire fonts.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a print character recognition device, and more particularly to a print character recognition device that can handle a plurality of types of character fonts.

[0002]

2. Description of the Related Art For example, when a transaction document from a large number of business partners is automatically recognized by using a character recognition device (OCR), the character font used for printing the transaction document usually differs depending on the business partner. Is. Therefore, a document that contains a lot of noise, such as a document received from a facsimile, may have a character font and OC that are used in the document.
If the character dictionary corresponding to the R dictionary is different, it is difficult to obtain a high recognition rate.

[0003]

As described above, in the OCR for recognizing a printed document, if the character font of the document does not correspond to that of the OCR, there is a case in which a sufficient recognition rate cannot be obtained and the document cannot be used.

Therefore, a main object of the present invention is to obtain a high recognition rate in character recognition of a printed document, regardless of the type of font of the printed document.

A secondary object of the present invention is to obtain a highly accurate recognition result in the shortest possible time when performing character recognition on a document printed with various character fonts.

[0006]

SUMMARY OF THE INVENTION A print character recognition apparatus according to the present invention includes a dictionary storage unit that stores a plurality of recognition dictionaries corresponding to a plurality of character fonts, and each printing source that provides a recognition target document. , A dictionary table in which dictionary numbers corresponding to the character fonts used are registered, and by referring to this dictionary table, dictionary means for selecting the dictionary number of the recognition dictionary to be used for the document to be recognized, and selecting It is characterized by further comprising a recognition means for performing character recognition of the recognition target document by using the dictionary in the dictionary storage means corresponding to the generated dictionary number.

[0007]

According to the present invention, various recognition dictionaries corresponding to various character fonts are prepared in advance. When the document to be recognized is received, first, the dictionary number corresponding to the character font used by the printing source is searched from the dictionary table, and the document to be recognized is recognized using the recognition dictionary corresponding to the dictionary number. Character recognition is performed. Therefore, even if the character font to be used is different for each printing source, an appropriate recognition dictionary is automatically selected according to the printing source, so that character recognition can be performed with a high recognition rate. Moreover, the processing time is shorter than that when all dictionaries are used.

In the preferred embodiment, after a registered dictionary number is once selected from the dictionary table, each of the dictionaries having the selected dictionary number is used to determine a specified partial range in the document to be recognized. Temporary recognition means for tentatively recognizing and obtaining the recognition rate is further provided. Then, if the recognition rate from the temporary recognition means is equal to or higher than a predetermined threshold value, the dictionary selection means finally selects the dictionary number having the highest recognition rate among them, and sets the threshold value. If there is no more than the value, all dictionaries in the dictionary storage means are selected. Further, even when the dictionary number corresponding to the printing source is not registered in the dictionary table, all dictionaries are selected.

In this way, the registered dictionary number in the dictionary table is incorrect, or it is not registered,
Even if the printing source changes the font used, the wrong dictionary is not used, and a high recognition rate can always be obtained.

Further, in addition to the above-mentioned proofreading, the apparatus of the present invention may further include dictionary registration means for selecting the dictionary having the highest correct answer frequency based on the recognition result from the recognition means and registering it in the dictionary table. . Then, even if the printing source changes the character font or a new printing source is added, an appropriate dictionary number is automatically registered in the dictionary table, so that manual registration work can be omitted or reduced.

[0011]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 shows the system configuration of an embodiment of a printed character recognition apparatus according to the present invention. The main use of this system is to read images of printed documents sent from various customers and automatically recognize the characters.

In FIG. 1, the workstation 1 is
The computer 11 controls the entire system, selects a recognition dictionary, calibrates and outputs the recognition result.
It comprises a display 13 and a keyboard 15, and further comprises a disk storage device 3 which stores a dictionary table described later.

An image processing unit (IPU) 5 for processing an image of a printed document according to a command from a computer 11 is connected to the workstation 1. An image reader 7 for reading an image of a printed document and a character recognition unit (CRU) 9 for recognizing characters of the printed document based on the read image are connected to the IPU 3.

FIG. 2 shows an example of a dictionary table stored in the disk storage device 3.

As shown in FIG. 2, customer codes for identifying various customers and dictionary numbers registered for each customer code are described in the dictionary table. here,
The dictionary number is an identification number of a plurality of recognition dictionaries stored in the CRU 9. As will be described later, since one recognition dictionary is provided for each character font in CRU 9 for various kinds of character fonts, the dictionary number is also a character font identification number. The registration of the dictionary number for each customer in the dictionary table is performed from the workstation 1 by checking the character font used by each customer in advance.

FIG. 3 shows character types (categories) to be recognized, and exemplifies 56 types of categories such as numbers, alphabetic characters and symbols used in English. FIG. 4 shows the structure of the recognition dictionary for the above 56 categories stored in the memory of the CRU 9.

As shown in FIG. 4, one dictionary is prepared for one character font, and 30 dictionaries D1 to D30 are prepared corresponding to 30 character fonts in total. Although reference vectors for 256 categories can be stored in each dictionary area, in this embodiment, only the reference vectors for the 56 categories shown in FIG. 3 are stored in the areas 1 to 56, and the reference vectors 57 to 57 are stored. Area No. 256 is vacant.

The reason why each dictionary has an area for 256 categories is that the hardware of the CRU 9 has 256 areas.
This is because 256 1-bit processing units are arranged in parallel so that category recognition processing can be performed simultaneously. In CRU9, when the dictionary used for recognition is selected, each bit of the feature vector of the character to be recognized is set to 25 bits.
Simultaneously to the six processing units, the corresponding bits of the reference vector of 256 categories of the selected dictionary are sent in parallel to the 256 processing units. This operation is repeated for all the bits of the feature vector of the recognition target character, and when the last bit ends, 2
Distance values for 56 categories will be obtained at the same time.

FIG. 5 shows the flow of the overall processing performed by the computer 11 of the workstation 1 in this system.

As shown in FIG. 5, the process starts when the image of the print document read by the image reader 7 is received from the image IPU 5 (step S1). First, the received document image is displayed on the display 13 (step S2), and then the dictionary selection process is started (step S2).
3). Details of this dictionary selection processing will be described later.

When the dictionary is selected, the dictionary number and the recognition execution command are transmitted to the IPU 5. Then, the IPU 5 sends the document image and the dictionary number to the CRU 9, and the CRU 9
Recognizes using the document image and the dictionary corresponding to the dictionary number, and returns the recognition result to the computer 11 through the IPU 5.

The computer 11 receives the recognition result (step S5), and then enters the calibration process (step S5).
6). The received recognition result includes 16 candidate character images of 1st to 16th in descending order of distance value for each character of the document. In the initial state of the calibration process,
Of the 16 candidates for each character, only the first candidate image is displayed in association with the original image. The operator compares the candidate image with the original image, and if there is a character in which the candidate image is incorrect, points it out with a keyboard or a mouse.

Upon receiving this indication, the computer 11 displays a list of the remaining 15 candidates for the character, and the operator selects the correct answer character from the list. The calibration process is completed by repeating such operations.

When the proofreading process is completed, a correct document image based on the recognition result after the proofreading is printed out and the document data of the document is stored in the disk storage device 3 and the process is completed.

FIG. 6 shows a detailed processing flow of the dictionary selection described above.

As shown in FIG. 6, first, the operator uses a mouse or a keyboard to specify a range (for example, the first sentence) for temporary recognition from the displayed document image (step S301). Then, the operator inputs the correct character string of the temporary recognition range from the keyboard (step S302). Further, the operator inputs the customer code from the keyboard (step S30
3), and input an execution command (step S30)
4).

Upon receiving these input information, the computer 11 first searches the dictionary table for the dictionary number corresponding to the customer code (step S305) and checks whether or not the dictionary number is registered (step S30).
6). As a result, if the dictionary number is not registered in the column corresponding to the customer code, the dictionary number "0" is selected (step S307).

On the other hand, as a result of the check in step S306, if one or more types of dictionary numbers are registered in the corresponding column, the recognition execution command of the designated range is sent to the IPU 5 together with each dictionary number. As a result, the CRU 9 recognizes the designated range using the dictionary corresponding to the dictionary number, and the recognition result is returned to the computer 11.

When the computer 11 receives the recognition result (step S309), it compares the recognition result with the previously inputted correct answer character string to calculate the recognition rate (correct answer rate) (step S310).

After performing the temporary recognition processing for this designated range for all the dictionaries registered in the corresponding column (step S311), the computer 11 determines that the highest recognition rate among the recognition rates of the dictionaries is predetermined. It is checked whether the threshold value (for example, 85%) is exceeded (step S312), and if it is exceeded, the dictionary number with the highest recognition rate is selected (step S313). If it is not exceeded, the dictionary number "0" is selected. Yes (step S307).

Here, the dictionary number "0" is 30 as described above.
This means that recognition is performed using all of the individual recognition dictionaries.

FIG. 7 shows the flow of the recognition processing (recognition processing of each character of the original image) performed by the CRU 9 which has received the notification of the dictionary number after the dictionary number is determined as described above. In this process, when the matching with the dictionary is completed, the distance values are sorted from the smallest distance and 16 candidate characters are output. The sorting method is performed in two stages by utilizing the 256-category parallel processing function of CRU9. FIG. 8 shows a state of this two-step sorting.

As shown in FIG. 7, when the dictionary number "0" is received, the recognition target character is first recognized using each of the 30 dictionaries (step S401). Next, vertical sorting is performed using the recognition result (step S40).
2).

As shown in FIG. 8, this vertical sorting is the one with the smallest distance (distance 1) for each category.
To the longest distance (distance 30). As a result, a candidate character group C1 of 56 categories with a distance of 1, a candidate character group C2 of 56 categories of a distance of 2, ..., A candidate character group C30 of 56 categories with a distance of 30 are obtained.

Next, a character group C1 having a distance of 1 is selected from the candidate character groups C1 to C30 (step S403),
The character group C1 is horizontally sorted (step S405).

In this horizontal sorting, the candidate characters are rearranged in ascending order of the distance value in the candidate character group C30 at the distance 1 as shown in FIG. After this horizontal sort, CRU9
Selects the top 16 candidate characters from the one with the smallest distance value and returns them to the workstation 1 as the recognition result (step S406).

On the other hand, when the dictionary number other than "0" is received, the CRU 9 recognizes using the dictionary corresponding to the dictionary number (step S404). As a result, since the distance value is obtained for each of the 56 categories, the horizontal direction sorting is performed using the distance value (step S40).
5), and the top 16 candidates are selected and output (step S406).

By adopting the above-described two-step sorting method, when a dictionary number other than "0" is designated, 2
The advantage is that only the stage sort is required.
In the case of the dictionary number “0”, the same category is sorted in the first step, so that the same category does not appear more than once in the final candidate character, so that many categories can be output as candidate characters. Therefore, it is advantageous for the subsequent calibration process.

As described above, in the present embodiment, the dictionary number corresponding to the font used is registered in advance for each customer, and recognition is performed using the dictionary corresponding to that number. Even if printing is performed using, the recognition can be performed at high speed and accurately. Also, when the font used by the customer is changed in the middle or a document from a customer who is not registered is received, it is recognized using all dictionaries, and as described above, the same category is included in multiple candidate characters. Since it is not included, a relatively high recognition rate can be obtained and the subsequent proofreading is easy.

The present invention can be implemented in various modes other than the above embodiments. For example, when recognition is performed with the dictionary number “0”, the dictionary stores the candidate character that is ranked first as a result of the horizontal sorting, and the storage result is totaled for all characters of the manuscript. The dictionary that has the highest score (the frequency of correct answers is high) may be automatically registered in the dictionary table. As a result, the dictionary number is automatically registered in the dictionary table, and manual registration can be omitted or reduced.

Further, according to the operator's discretion, temporary recognition in the dictionary selection processing can be omitted so that the dictionary number designated by the operator can be forcibly used, or if there are a plurality of registered dictionary numbers, the plurality of dictionary numbers can be used. It is also possible to use all of the dictionaries to perform the recognition process similar to the dictionary number "0". Thereby, the trouble of performing the temporary recognition can be omitted depending on the case.

[0043]

As described above, according to the present invention,
Regardless of the type of font used to print the document, it is possible to automatically select the corresponding dictionary and obtain a high recognition rate. Further, since the corresponding dictionary number is registered in advance, the entire processing time becomes short.

[Brief description of drawings]

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

FIG. 2 is a diagram showing an example of a dictionary table of the same embodiment.

FIG. 3 is a diagram showing an example of a recognition category of the same embodiment.

FIG. 4 is a diagram showing a dictionary configuration of the same embodiment.

FIG. 5 is a flowchart showing the overall processing flow of the workstation of the embodiment.

FIG. 6 is a flowchart showing a detailed flow of dictionary selection processing in FIG.

FIG. 7 is a flowchart showing the flow of CRU recognition processing according to the embodiment.

FIG. 8 is a diagram showing a state of sorting in two stages of the recognition processing of FIG. 7.

[Explanation of symbols]

 1 workstation 3 disk storage device 5 image processing unit (IPU) 7 image reader 9 character recognition unit (CRU)

Claims (5)

[Claims] 1. An apparatus for recognizing characters in a printed document from one or more printing sources, and a dictionary storage unit for storing a plurality of recognition dictionaries corresponding to a plurality of character fonts, which is used for each printing source. A dictionary table in which dictionary numbers corresponding to the character fonts are registered, and by referring to the dictionary table, dictionary means for selecting the dictionary number of the recognition dictionary to be used for the document to be recognized, A print character recognition device comprising: a recognition unit that performs character recognition of the recognition target document using a dictionary in the dictionary storage unit corresponding to a dictionary number. 2. The apparatus according to claim 1, wherein when the dictionary selection unit registers a dictionary number corresponding to a printing source of the recognition target document in the dictionary table, the registered dictionary number. Is selected, and if not registered, all the dictionaries in the dictionary accumulating means are selected. 3. The apparatus according to claim 2, wherein when the dictionary selecting unit selects the registered dictionary number, the recognition target document is selected by using each dictionary corresponding to the selected dictionary number. If the dictionary selection means includes a recognition rate that is equal to or higher than a predetermined threshold among the recognition rates from the temporary recognition means, the maximum recognition rate is obtained by temporarily recognizing a designated range in the recognition range. The print character recognition device is characterized in that the dictionary number that has obtained the recognition rate is finally selected, and if there is no dictionary number greater than or equal to the threshold value, all dictionaries in the dictionary storage means are selected. 4. The apparatus according to claim 1, further comprising dictionary registration means for selecting a dictionary having the highest correct answer frequency based on a recognition result from the recognition means and registering the dictionary in the dictionary table. Print character recognition device. 5. A method of performing character recognition of a print document from one or more printing sources, refers to a dictionary table in which a dictionary number corresponding to a character font used for each printing source is registered in advance. By doing so, the dictionary process of selecting the dictionary number of the recognition dictionary to be used for the document to be recognized, and the dictionary corresponding to the selected dictionary number, A printing character recognition method, comprising: a recognition process of selecting a character from the selected recognition dictionary and performing character recognition of the recognition target document using the selected recognition dictionary.