JPH076203A - Character recognition device - Google Patents

Abstract

PURPOSE:To provide the character recognition device in which characters are precisely recognized and the processing speed is improved with respect to the character recognition device recognizing characters according to a predetermined character string. CONSTITUTION:Image data read by a document read section 1 are segmented by a character recognition preprocessing section 2 as a character string with a predetermined character number, and a character recognition section 3 applies character recognition processing to the character string. A parameter control section 5 implements setting processing of a parameter used to set a proper character segmented number based on data outputted from the character recognition section 3 and an operator control section 4, the proper parameter value is stored in a memory and then characters having a highest character number in the recognition rate of correctly recognized characters based on the learned data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing printed characters and handwritten characters, and more particularly to a character recognition device for recognizing a character according to a predetermined character string.

[0002]

2. Description of the Related Art Today, a character recognition system, such as an OCR (optical character reader), has been developed for recognizing a target character from image data using a computer. This character recognition system reads characters on a form by a photoelectric conversion device such as a CCD, cuts out a predetermined character string from the read image data, and performs character recognition on the cut out character string.

Conventionally, the number of characters constituting this character string is set by performing the recognition process on the form to be subjected to the character recognition process while changing the number of characters, and determining the number of characters that allows the most accurate character recognition. For example, with respect to the character recognition of a form used for a predetermined purpose, the number of characters to be cut out is sequentially changed to 3 characters, 4 characters, and so on, and the character recognition is performed. The number of characters is decided. The number of characters thus determined is fixedly set as the number of characters of the character string to be cut out from the image data, and thereafter, the character string having this number of characters is sequentially read from the image data to perform the character recognition process.

[0004]

In the conventional character recognition device, recognition processing is performed on a character string having the number of characters fixedly set as described above, but when the environment at the time of setting changes,
For example, although print characters were initially described in the form, when handwritten characters are described later, or when the format of the form to be used is changed, the character recognition cannot be performed correctly.

Further, in the conventional setting of the number of characters constituting a character string, the number of characters constituting the form is set without considering the change of so-called fields constituting the form. That is, for example, in the case of a bank transfer form as a form, the transfer form has various fields such as a name field, an address field, and an amount field. The number of characters is set without considering. For example,
Generally, the letters in the name column are written at wide intervals, and conversely, the address column is written closely. Further, the amount column is usually formed in a grid shape, and a character is written in each grid. Therefore, in the conventional setting method in which the number of characters to be cut out is set as one field without considering such a difference in the configuration of each field, accurate character recognition cannot be performed.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a character recognizing device capable of appropriately setting the number of character cutouts in consideration of fields of a form, performing accurate character recognition, and improving the processing speed. .

[0007]

As one configuration example of a character recognition device of the present invention, an image data conversion means for converting written characters into image data, and a character cutout for cutting out predetermined characters from the image data Means and
Parameter setting / control means for constantly setting the parameter value for determining the number of characters to be cut out by the character cutting means, and the parameter setting / control means for always holding the set parameter value at an appropriate value based on information obtained from a form, and the parameter setting / control means And a character recognition means for recognizing a character string cut out in accordance with the number of character cutouts determined based on the parameter value stored in the storage means. There is.

Further, the parameter setting / control means is
It is composed of means for learning the correlation of the optimum parameter values of the entry fields and means for estimating the proper parameter values from the correlation of the optimum parameter values of each entry field.

[0009]

According to the present invention, when the characters of the form are read as image data by the image data converting means and the read image data is cut out as a character string having a predetermined number of characters, the parameter setting / control means makes a field. Different parameter proper values are set for each, and the parameter values of each field are stored in the storage means in the form of learning data as learning data, so that the learning data can be used later for accurate character recognition processing, and the processing speed can be improved. Is to improve.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a system block diagram of the character recognition device of this embodiment. This system is composed of a form reading unit 1, a character recognition preprocessing unit 2, a character recognition unit 3, an operator operation unit 4, and a parameter control unit 5. The form reading unit 1 is, for example, a device that reads image data written on paper, and a photoelectric conversion unit (character reading unit) (not shown) that reads characters on the paper and photoelectric conversion is performed on the paper on which the characters are written. It is configured by a transport unit (not shown) that transports to another unit. The photoelectric conversion unit is composed of a CCD sensor or a MOS type image sensor, photoelectrically converts the characters on the paper sent to the conveying unit and reads them, and outputs them as digital data to the character recognition preprocessing unit 2.

The character recognition preprocessing unit 2 carries out preprocessing such as noise removal and the above character cutout from the image data read by the form reading unit 1. For example, when the read character data has noise such as so-called “stain” or “chip”, this is removed (noise removal), and a predetermined number of character strings to be described later in the read character data are cut out and cut out. The character is output to the character recognition unit 3.

The character recognition unit 3 has a word dictionary, recognizes the characters cut out by the character recognition preprocessing unit 2, searches for similar candidate characters, and performs character recognition processing. The operator operation unit 4 is composed of a display and a keyboard, and performs, for example, display processing of characters recognized by the character recognition unit 3 and input processing of operation signals such as correction of erroneously recognized characters instructed by the operator.

The parameter control unit 5 is a device for setting the number of characters to be cut out when the above-mentioned character recognition preprocessing unit 2 cuts out a character, and based on a parameter value described later, a character recognition target sheet is selected. On the other hand, the data of the cut-out character number having the highest character recognition rate is output to the character recognition preprocessing unit 2. The parameter value set by the parameter control unit 5 is set based on the operation signal input from the operator operation unit 4 and the data output from the character recognition preprocessing unit 2 and the character recognition unit 3.

The processing operation of the character recognition apparatus having the above-mentioned configuration will be described below. First, as described above, the form reading unit 1 photoelectrically converts characters on a sheet (not shown) to read them as image data (digital data). Noise is removed from the image data by the character recognition preprocessing unit 2 as described above, and a character string of the designated number of characters is cut out from the image data according to the character cutout number data output from the parameter control unit 5, and the character recognition is performed. Output to the part 3. The character recognition unit 3 performs character recognition processing for each input character string.

Then, the parameter controller 5
Parameter value setting and changing processing is performed based on the flowchart shown in FIG. That is, the parameter control unit 5
Inputs the data output from the character recognition preprocessing unit 2,
The process is executed based on this data.

First, when the parameter control unit 5 receives the form recognition data (form recognition result) from the character recognition preprocessing unit 2, it determines whether there is learning data of the form based on this data (step (hereinafter referred to as ST)). 1). This determination is based on the form recognition data read from the character recognition preprocessing unit 2 and whether or not there is learning data for the entry field and the parameter value. Like the above example, it seems that the conditions for writing characters such as a name entry field, an address entry field, an amount entry field, etc. are different, as in the above example, such as a bank transfer form. The learning data is stored in a memory (not shown) for a form that has already been learned. Therefore, by searching this learning data, it can be determined whether or not the corresponding parameter value is stored. Here, when the learning data is already stored (ST1 is YES
(Yes)) and processing (ST9 to ST11).
Incidentally, this processing (ST9 to ST11) will be described later on.

Therefore, first, when there is no learning data (ST1 is NO), the following process is performed for each field (entry field) for one form. First, the difference between the arbitrary character cutout number and the correct answer cutout number at the time of initial setting is calculated to determine the parameter proper value search direction (S
T2). In this processing, character recognition processing is performed by setting an arbitrary number of character cutouts at the time of initial setting to, for example, "6", and when it is determined that the correct number of character cutouts after correction processing by the operator is, for example, "4", the difference "2" To calculate
Based on this result, the parameter control unit 5 determines the parameter search direction. This search direction is determined in the direction in which the parameter value increases if "the number of correct cut characters is less than the number of character cuts", and in the direction that the parameter value is smaller if "the number of correct cut characters> the number of cut characters". To do. That is, the characters written in a specific entry field for a certain form have a certain tendency, and the search direction for the appropriate parameter value is determined based on the tendency. Therefore, in the above-mentioned example, the number of correct cut-out characters is “4” and the number of cut-out characters is “6”, so that “the number of correct-cut cut-out characters is smaller than the number of cut-out characters”, and the parameter value is searched in the direction of increasing. still,
The width of the normal parameter value is set in the range of "126" to "516".

Next, in the direction determined by the above processing,
A preset constant parameter value (for example, 30,
40) is changed (ST3). For example, in the case of the above example in which the parameter value is moved in the larger direction, the predetermined value (eg, “30”) is increased with respect to the current parameter value (eg, “256”). On the contrary, when moving in the direction in which the parameter value is smaller, the predetermined value (eg, “30”) is decreased with respect to the current parameter value (eg, “256”).

The character recognition preprocessing unit 2 performs a character cutting process based on the parameter values set in this way (S).
T4). Then, the difference between the number of characters cut out by the above-described processing and the number of correct character cutouts output from the operator operation unit 4 is calculated again (ST5). And
The absolute value of the difference calculated in this way is compared with the difference obtained by the same process (ST2) immediately before, and it is determined whether the difference in the current character cutout number is larger than the difference in the previous character cutout number. (ST6). Here, when the current character cutout number difference is smaller than the previous character cutout number difference (ST6 is NO), the parameter value is sequentially moved in the proper direction, and the difference between the correct answer cutout character number and the character cutout number is becoming smaller. And judge
Further, the process returns to the process (ST3) and a new parameter value is replaced.

On the other hand, when the difference between the current character cutout numbers is larger than the previous character cutout number difference (YES in ST6), the process proceeds to the next process (ST7). That is, in this case, the difference between the number of correct characters and the number of extracted characters is usually small, but if the difference is large, it is determined that the characters are not in the same field, and it is determined that the entry field has changed. . Therefore, at this time, the correct answer rate is obtained for each parameter value within the range where the difference between the number of character cutouts measured up to that point and the number of correct character cutouts is the smallest.
The parameter value with the highest correct answer rate is set as the appropriate value of the entry field measured up to that point (ST7). Therefore, by this processing, the parameter appropriate value of one entry field of the form is set. That is, in the case of the above-mentioned example, it means that the appropriate parameter values of one field (for example, the name field) of the bank transfer sheet as a form have been obtained.

Next, it is judged whether or not the parameter values of all the fields for one form have been set. In the above-mentioned processing, since the parameter values of only one entry field have been set yet, the processing returns to the processing (ST2). , Executes character cutting processing for the next entry field. That is, in the next entry field (for example, address field), the difference between the number of character cutouts and the number of correct answer cutouts is calculated to determine the parameter proper value search direction (ST2), and the parameters are changed by a predetermined constant value. (ST3). Then, character cutting is performed based on the parameter values set in this way (ST4), and the difference is calculated again as described above (ST5).
Compare the differences obtained by the previous same processing (ST
6) When the difference between the current character cutout numbers is smaller than the previous character cutout number difference (ST6 is NO), the process returns to ST3, while when the current character cutout number difference is larger than the previous character cutout number difference (ST6 is YES). ), And the same processing (ST7) as described above is executed.

In this way, the appropriate parameter values are set every time the entry fields change in sequence, and when the appropriate parameter values for all entry fields are set for one sheet, the process proceeds to the process (ST8). In this processing, the above-described processing is performed for all fields in the form, and when this processing is completed, an inter-field correlation equation is obtained and stored as learning data. This correlation formula is calculated using the first field (for example, “name column” in the case of bank transfer paper) and the second field.
Fields (for example, “address field” in the case of the bank transfer form) are to be associated and obtained.
For example, if the parameter of the first field is X and the parameter of the second field is Y, then Y = aX + b (a, b
Is a constant). Correlation equations are similarly obtained for the second and subsequent fields. The correlation equation thus obtained is stored in a memory (not shown) as learning data together with the type of form (ST8).

Then, a proper parameter value is set (S
(T11), the parameter proper value thus set is output to the character recognition preprocessing unit 2 described above. The character recognition preprocessing unit 2 performs character cutting processing based on this parameter value. That is, the image data read by the form reading unit 1 is cut out by a predetermined number of characters based on the parameter value output from the parameter control unit 5. The character recognition unit 3 performs character recognition on the character string cut out in this manner. Further, the result of this character recognition is displayed on the display of the operator operation unit 4, and the parameter control unit 5
The recognition result is notified to.

On the other hand, in the above-mentioned judgment (ST1), when it is judged that the learning data already exists, the above-mentioned processing (ST2-ST8) is executed only for the first field. That is, in this case, when the above-mentioned processing (ST2 to ST8) is performed for the first field and the parameter value of the first field is set, the correlation formula (ST8) obtained as described above is obtained for the second field and thereafter. It is read from the memory, and the parameter proper value of the second field is obtained from the parameter proper value of the first field and the above correlation equation (ST10). Furthermore, for the second and subsequent fields, the correlation parameter is used in the same manner to obtain the appropriate parameter value. That is, when the learning data already exists in the memory,
For the second and subsequent fields, the learning data can be used to know the proper parameter values.

The parameter proper value thus obtained is set in the parameter control unit 5 (ST11),
It is output to the character recognition preprocessing unit 2. Character recognition preprocessing unit 2
Performs the character cut-out processing based on this parameter value as described above, cuts out the image data read by the form reading unit 1 by a predetermined number of characters based on the parameter value, and the character recognition unit 3 performs character recognition.

In this embodiment, the case of the bank transfer form as a form has been described, but it is used for other forms having a plurality of entry fields, for example, manuscript paper having a predetermined mesh or a company. The same applies to paper of various formats.

[0027]

As described in detail above, according to the present invention, even for a form having a plurality of fields, an appropriate parameter value is set for each field, and character cutting processing is performed based on this parameter value. , Character recognition with an extremely high accuracy rate can be performed.

Further, since proper parameter values are stored in the memory as learning data in the form of a correlation equation, it is possible to obtain parameter values later by using this learning data and perform proper character segmentation. The processing can be performed in a short time.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of a character recognition device according to an embodiment.

FIG. 2 is a flowchart illustrating an operation of the character recognition device according to the embodiment.

[Explanation of symbols]

 1 Form reading unit 2 Character recognition preprocessing unit 3 Character recognition unit 4 Operator operation unit 5 Parameter control unit

Claims (3)

[Claims] 1. Image data conversion means for converting the described characters into image data, character cutout means for cutting out predetermined characters from the image data, and setting of a parameter value for determining the number of characters cut out by the character cutout means, And parameter setting / control means for always holding the set parameter value at an appropriate value based on information obtained from a form, storage means for storing the parameter value set by the parameter setting / control means, and the storage A character recognition device, comprising: character recognition means for recognizing a character string cut out according to the number of character cutouts determined based on a parameter value stored in the means. 2. The character recognition device according to claim 1, wherein the parameter setting / control means comprises means for learning the correlation of the optimum parameter values of the entry field. 3. The character recognition device according to claim 1, wherein the parameter setting / control means comprises means for estimating a parameter appropriate value from a correlation of parameter appropriate values in the entry field.