JPH06309505A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve the character recognizing factor by means of two types of dictionaries having high precision. CONSTITUTION:The dimension number of character feature value is set by a dimension setting part 1 and then sent to a single character end detecting part 2 and an address generating part 3. An address counter 4 successively sends its count value to the part 2 and an address converting part 5. The part 5 converts the received count value into an address and sends this address to an unknown character memory 6. The part 3 generates an address based on the dimension number set by the part 1 and the count value supplied via the part 2 and sends the address to a dictionary memory 7. The feature value equivalent to a single unknown character is previously written in the memory 6, and four pieces of data are read out of the memory 6 for each dimension of feature value and in the order of addresses and sent to a correlation with metic part 9. The feature value data equivalent to several thousands of characters are previously stored in the memory 7, and four pieces of data are read out of the memory 7 for each dimension of feature value and in the order of addresses and sent to the part 9, respectively.

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 an unknown character pattern using a dictionary pattern.

[0002]

2. Description of the Related Art A conventional character recognition device extracts a feature amount from an unknown character, compares it with a feature amount stored in advance as dictionary data for each character, calculates a similarity degree by a correlation operation, and obtains the highest similarity degree. The large character is used as the recognition result. FIG. 6 is a block diagram showing the configuration of a conventional character recognition device. In the figure, an address signal is periodically sent from the address counter 4 to the unknown character memory 6 and the dictionary memory 7. When the address signal is input to the unknown character memory 7, the characteristic amount of the unknown character stored in the corresponding address is read and sent to the correlation calculator 9. In the dictionary memory 7, when the address signal is input, the features from the first character to the last character are read out in order and sent to the correlation calculator 9.

The correlation calculator 9 compares the transmitted characteristic quantities and obtains the degree of similarity by the correlation calculation.
Send to the sorting unit 10. The sorting unit 10 sorts the obtained similarities each time, and when the sorting of the similarity of all the characters of the dictionary is completed, the dictionary character having the highest similarity is output as a recognition result. The dictionary used here stores M features each of which is a feature amount consisting of N bits, and when the total number of characters in the dictionary is S, the total number of data in the dictionary is N. × M × S bits. The number of feature quantities per character is generally called the number of dimensions.

[0004]

However, since the conventional character recognition device is provided with only one of the dictionary for normal quality characters and the dictionary for low quality characters, the quality level of unknown characters is There is a problem that the recognition rate decreases when the quality level of the dictionary does not correspond. For example, when using a dictionary for normal quality characters, it is easy to make an error when recognizing low quality characters with faintness or crushing. Also,
Even if an attempt is made to use a dictionary having a large number of character feature amount constituent data in order to improve the recognition rate, there is a problem that a highly accurate dictionary cannot be used because the hardware configuration of the correlation calculation unit cannot be easily changed. . The present invention has been made to solve the above problems, and an object of the present invention is to enable the use of a highly accurate dictionary to improve the recognition accuracy and to improve the efficiency of the processing time. The object is to provide a character recognition device that can do this.

[0005]

In order to achieve the above object, a first aspect of the present invention is to extract a feature amount from an unknown character pattern and store the feature amount in advance as a dictionary for each character. In a character recognition device that outputs a character having the highest similarity obtained by comparison as a recognition result, a means for designating the number of data per dimension of the feature quantity to be compared for each character and a feature quantity for each character. It is characterized in that it is provided with a dictionary configured in the same manner as the number of generated data.

A second aspect of the invention is to recognize the character having the highest degree of similarity obtained by extracting a feature amount from an unknown character pattern and comparing it with a feature amount stored in advance as a dictionary for each character. In the character recognition device for outputting as, a plurality of dictionaries storing feature quantities for character groups having different quality levels are provided. Third
According to the second invention, in the second invention, there is provided means for designating a dictionary to be used from a plurality of dictionaries.

A fourth aspect of the present invention is, in the second aspect, a means for specifying the quality level of an unknown character based on the quality level of the dictionary for which the greatest degree of similarity is obtained, and a character for which character recognition is continuously performed. A method of comparing the quality levels specified for each character in the column, correcting the minority characters whose quality level is different from other majority characters to the quality level of other characters, and a dictionary in which the corrected characters correspond to the new quality level. And means for performing the comparison process again using.

[0008]

According to the first aspect of the invention, the number of data per dimension of the feature amount to be compared for each character is designated, and the dictionary is constructed with the same number of data as the designated number of data per dimension. There is. As a result, it is possible to change the number of feature quantity data to be compared, and it is possible to use a highly accurate dictionary having a large number of data per dimension. Second
In the invention of claim 1, since a plurality of dictionaries that store the feature amount for each character group having different quality levels are provided, the feature of the unknown character is compared for each of the plurality of dictionaries, and the largest similarity among them is obtained. Is adopted. As a result, even if the quality level of the unknown character changes, high-accuracy recognition is always performed.

In the third invention, it is possible to specify the dictionary to be used from among the plural dictionaries provided. As a result, the optimum dictionary can be designated according to the quality of the unknown character to be recognized, and the processing time can be shortened without lowering the recognition rate. In the fourth invention, based on the quality level of the dictionary for which the largest similarity is obtained,
The quality level of the unknown character is identified. Then, the quality level specified for each character of the character string in which character recognition is continuously performed is compared, and the minority character having a quality level different from that of other majority characters is corrected to the quality level of other characters, and
The corrected characters are compared again with the dictionary corresponding to the new quality level. As a result, the recognition rate is further improved.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a first embodiment according to the first invention. In this embodiment, a dimension number setting unit 1, a one-character end detection unit 2, an address generation unit 3, an address conversion unit 5, and a recognition processing end detection unit 8 are added to the conventional example shown in FIG. The dimension setting unit 1 sets the dimension number of the character feature amount in advance, and sends the set value to the one-character end detection unit 2 and the address generation unit 3.

As shown in FIG. 2, the address counter 4 sequentially counts up and sends the count value to the one-character end detection unit 2 and the address conversion unit 5. Based on the input count value, the address conversion unit 5 converts the count value into the address of the unknown character memory 6 and sends it to the unknown character memory 6. The address generator 3 has a set number of dimensions,
As shown in FIG. 2, an address of the dictionary memory 7 is generated based on the count value input via the one-character end detection unit 2 and sent to the dictionary memory 7 and the recognition process end detection unit 8.

In the unknown character memory 6, the characteristic amount extracted from the unknown image data of one character is written in advance,
Four pieces of data per one-dimensional feature amount are sequentially read according to the input address and sent to the correlation calculation unit 9. Data indicating the characteristics of several thousand characters is written in the dictionary memory 7, and four pieces of data per one dimension of the characteristic amount are sequentially read according to the input address and sent to the correlation calculation unit 9.

In FIG. 2, four pieces of data "1-1", "1-2", "1-" read from the dictionary memory 7 are used.
"3" and "1-4" correspond to "1" which is the first feature amount of the unknown character. Further, the data “2-1” to “2-4” of the dictionary memory 7 that are subsequently read correspond to the data “2” that is the second feature amount of the unknown character memory 6. For example, the value "11" is
This means that it can be represented as the sum of four data, "3", "3", "3", "2".

The correlation calculation unit 9 compares the sent characteristic amount data and executes the correlation calculation, obtains the similarity for each character of the dictionary, and sends it to the sorting unit 10. Sorting section 1
With 0, sorting is executed in order from the largest similarity value input for each character in the dictionary. When the similarity is sorted for all the characters in the dictionary memory 7, the character having the highest similarity among them is output as the recognition result.

In the illustrated example, the number of constituent data per one dimension of the feature amount is 4, but the dimension number setting unit 1
It can be set to any number from N to N. As described above, by changing the number of data items per one dimension of the feature amount, it is possible to change the number of data items per one dimension of the feature amount in the dictionary memory 7 as well. It is also possible to upload. Ie 1
It is possible to use a highly accurate dictionary with a large number of data per dimension, and the recognition accuracy is improved accordingly.

Next, the second according to the second, third and fourth inventions
An example will be described. FIG. 3 is a block diagram showing the configuration of the second embodiment. In the figure, 11 and 12 are unknown character memories, in which characteristic quantities of unknown character patterns having the same contents are stored. Reference numeral 13 denotes a correlation calculation / sorting unit, which compares the feature amount read from the unknown character memory 11 with the feature amount read from the normal quality character dictionary memory 14 to execute the correlation operation to execute the correlation operation.
The degree of similarity is obtained for each character of, and the degree of similarity obtained for each character of the dictionary is sorted in descending order.

Similarly, 15 is a correlation calculation / sorting unit, which compares the characteristic amount read from the unknown character memory 12 with the characteristic amount read from the low quality character dictionary memory 16 to execute the correlation operation and execute the dictionary operation 16 The degree of similarity is obtained for each character of, and the degree of similarity obtained for each character of the dictionary is sorted in descending order. 17
Is a processor, the correlation calculation / sorting unit 13,
Finally, the unknown character is recognized based on the character having the highest degree of similarity obtained in step 15.

FIG. 4 is a flowchart showing the operation of the second embodiment, and the operation will be described below based on this flowchart. First, when the character reading is started, it is determined whether or not the character quality is set (S1). If it is set, it is determined whether or not the setting content is of low quality (S
2). If the quality is not low quality but normal quality, the normal quality character dictionary is selected (S3). Next, using the selected normal quality character dictionary, recognition processing is performed on the unknown character and the recognition result is output (S4, 5).

When the quality is low in S2, a dictionary for low quality characters is selected (S6), the recognition process is performed using the dictionary for low quality characters, and the recognition result is output (S7,
8). If no character quality is set in S1, parallel execution of the normal quality character dictionary and the low quality character dictionary is selected (S9). Next, the recognition process is executed for each dictionary, and the similarities obtained are compared (S10,
11). Of the obtained similarities, the higher similarity is set as the correct answer, and the character having the similarity is once set as the reading of the target unknown character (S12).

Here, it is determined whether or not another character in the character string to which the unknown character which has been regarded as the correct answer belongs is recognized by the low quality character dictionary (S13). If it is not low quality but normal quality, it is determined whether or not the recognition result as the correct answer in S12 is normal quality (S14). When the quality is not the normal quality but the low quality, the recognition result determined as the correct answer in S12 is switched to the character having the similarity obtained in the normal quality character dictionary to be the recognition result (S15). If the quality is normal, the recognition result that is correct in S12 is output as it is (S16).

In S13, if it is recognized by the low quality character dictionary, it is determined whether or not the recognition result which is the correct answer in S12 is of low quality (S17). When the quality is not the low quality but the normal quality, the recognition result which is the correct answer in S12 is switched to the recognition result by switching to the character having the similarity degree obtained in the low quality character dictionary (S18). When the quality is low, the recognition result which is correct in S12 is output as it is (S19). The above-mentioned processes are executed by the processor 17 shown in FIG.

FIG. 5 is a flow chart showing the recognition processing in FIG. 4, and the recognition processing will be described below based on this flow chart. When the recognition process is started, first, the address for the dictionary memory is initialized and the address for the unknown character memory is initialized (S21, S2).
2). Next, the number of dimensions M is set (S23). Next, based on the designated address, the feature amount data is read from the dictionary memory and the feature amount data is read from the unknown character memory (S24, 25).

The data read here are compared with each other to perform a correlation operation (S26). Further, the address is updated, the data of the next dimension is read out and the correlation calculation is repeated until the set M dimension is completed (S27, 28).
In this way, the similarities obtained by the correlation calculation for one character in the dictionary are sequentially stored, and the similarities are sorted (S29, 30). By repeating the storage and sorting of these similarities for all the characters in the dictionary, the character with the highest similarity becomes the recognition result (S31).

In the second embodiment, as described above, by providing the two types of dictionaries having different character qualities, it is possible to use the dictionary corresponding to the unknown character quality even if the quality thereof changes, and the recognition accuracy is improved. improves. When recognizing a character string, recognition is performed while selecting a dictionary according to the quality of each of the constituent characters, and in the obtained recognition result, a dictionary with a quality different from the quality of other characters in the character string is obtained. If there is a character recognized by, the recognition result for that character may be erroneous, and therefore the recognition is performed again using the dictionary having the same quality as the other characters. Thereby, the recognition accuracy is further improved.

Although the two types of dictionaries are used in the second embodiment, three or more types of dictionaries can be used in the same manner. Further, the first and second embodiments can be realized by combining them with each other, or can be independently applied to the character recognition device.

[0026]

As described above, according to the first invention,
By making it possible to change the number of data for one dimension of the feature amount to be compared for each character, it is possible to use a highly accurate dictionary having a large number of data for one dimension and improve the recognition accuracy. According to the second aspect of the present invention, a plurality of dictionaries storing feature quantities for character groups having different quality levels are provided, and by comparing the characteristics of unknown characters for each dictionary, the quality level of unknown characters can be improved. Even if it changes, high-precision recognition is always performed.

According to the third aspect of the present invention, the dictionary to be used is specified from among the plural dictionaries provided, so that the optimum dictionary is specified according to the quality of the unknown character to be recognized and the recognition rate is lowered. It is possible to shorten the processing time without causing it. According to the fourth invention, when there are characters of different quality levels in a character string in which character recognition is continuously performed, the quality level of the character is corrected to be the same as the others, and then the corresponding dictionary By doing the comparison process again,
The recognition rate can be further improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment according to the first invention.

2 is an explanatory diagram showing data contents read from the memory of FIG. 1. FIG.

FIG. 3 is a block diagram showing a configuration of a second embodiment according to the second to fourth inventions.

FIG. 4 is a flowchart showing the operation of the second embodiment.

FIG. 5 is a flowchart showing in detail the recognition processing in FIG.

FIG. 6 is a block diagram showing a configuration of a conventional example.

[Explanation of symbols]

 1 dimension number setting unit 2 one-character end detection unit 3 address generation unit 4 address counter 5 address conversion unit 6 unknown character memory 7 dictionary memory 8 recognition processing end detection unit 9 correlation calculation unit 10 sorting unit 11, 12 unknown character memory 13 correlation calculation • Sorting unit 14 Dictionary memory for normal quality characters 15 Correlation operation / Sorting unit 16 Dictionary memory for low quality characters 17 Processor

Claims (4)

[Claims] 1. A character whose characteristic value is extracted from an unknown character pattern and which is compared with a characteristic value stored in advance as a dictionary for each character and which has the highest degree of similarity, is output as a recognition result. The recognition device is provided with a means for designating the number of data of the feature amount per one dimension to be compared for each character, and a dictionary configured so that the feature amount of each character is the same as the designated data amount. Characterized character recognition device. 2. A character whose characteristic value is extracted from an unknown character pattern and is compared with a characteristic value stored in advance as a dictionary for each character and which has the highest degree of similarity and is output as a recognition result. A character recognition device comprising a plurality of dictionaries that store feature quantities for character groups having different quality levels. 3. The character recognition device according to claim 2,
A character recognition device comprising means for designating a dictionary to be used from a plurality of dictionaries. 4. The character recognition device according to claim 2, wherein means for specifying the quality level of the unknown character based on the quality level of the dictionary for which the largest similarity is obtained, and characters for which character recognition is continuously performed. A means to compare the quality levels specified for each character in the column, to correct minority characters whose quality level is different from other majority characters to the quality level of other characters, and a dictionary of corrected characters to the new quality level. A character recognition device comprising: means for performing a comparison process again using.