JPS6019550B2 - Shape recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides a figure recognition device, in particular, a dictionary unit storing standard feature information corresponding to standard figures, which has a high-speed access memory built into a data processing unit and a low-speed access memory outside the data processing unit. In a figure recognition device that is configured separately in The present invention relates to a graphic recognition device that stores information on the high-speed access memory to reduce the time required for recognition processing.

In a figure recognition device, for example, a character recognition device, the feature information extracted from the character to be recognized is compared one by one with a plurality of standard feature information stored in a dictionary, and the standard feature information corresponding to the hit standard feature information is The recognition target character is recognized as corresponding to the character category. For this purpose, a plurality of pieces of standard feature information are stored in the dictionary section, and the standard feature information includes, for example, 20 for number recognition, 3 for fleas, and 3 for alphabetic character recognition.
It requires 0 nashi and 8 night jus, and 100 nashi and 15 tochi Hitomi for katakana recognition. The number of problems increases further when trying to deal with the various writing habits that occur during handwriting. Storing all such standard feature information on a high-speed access memory is undesirable because the memory becomes huge.

For this purpose, standard feature information with a high hit rate is generally stored on a high-speed access memory, and standard feature information that deals with special writing habits is stored, for example, on a floppy disk. However, due to writing habits at the destination where character recognition devices are shipped, standard feature information that is generally thought to have a high hit rate at the manufacturer may not necessarily have a high hit rate at the destination. arise.

The present invention aims to solve the above points,
Considering that the standard feature information mentioned above is created in such a way that ``if a hit is found using more general standard feature information, there is no need to compare it with standard feature information prepared to deal with modified characters.'' However, the purpose is to check the hit rate at the shipping destination during daily work and to store it in the high-distance access memory.

Therefore, the figure recognition device of the present invention includes a dictionary section that stores feature information extracted in advance for each of a plurality of standard figures, and a feature extraction section that extracts features about the recognition target figure. The comparison processing unit includes a comparison processing unit that checks a matching state between the feature information given by the user and the standard feature information given from the dictionary unit, and a determination processing unit that determines a category based on the matching processing result in the comparison processing unit, and the comparison processing unit The figure recognition device that recognizes the target figure based on the matching processing results in the figure recognition device includes at least a high-speed access memory and a low-speed access memory as the dictionary unit, and standard feature information stored in the dictionary unit. A count information storage unit is provided for counting hits for the standard feature information in response to the above, and the content of the count information storage unit for the standard feature information is updated in response to the determination by the determination processing unit; The present invention is characterized in that standard feature information having a high number of hits is stored in the high-speed access memory based on the contents of the count information storage section. This will be explained below with reference to the drawings. The figure shows the configuration of an embodiment of the present invention.

In the figure, numeral 1 is a figure recognition device, 2 is a data processing section, 3 is a video signal generation section, 4 is a preprocessing section, 5 is a feature extraction section, 6
7 represents a dictionary section, 7 a high-speed access memory forming the dictionary section, 8 a low-speed access memory forming the dictionary section, 9 a comparison processing section, and 10 a determination processing section. Also 11
is the feature information extracted from the recognition target figure, 12-0,1
2-1,...・・・．・・・．・represents standard feature information. The figure to be recognized is scanned by the video signal generator 3 and converted into an electrical signal.

The electrical signal obtained by the video signal generating section 3 is guided to a pre-processing section 4 to undergo processing such as noise removal, as is well known, and then guided to a feature extracting section 5. The feature extraction unit 5 extracts various predetermined features of the recognition target figure, such as a loop, a cap, a cup, etc., and creates the feature information 11. The characteristic information 11 is generally composed of N-bit information, and for example, if a loop exists, #1 in the diagram
If logic "1" is set in the bit and it does not exist, logic "1" is set in the bit.
0" will be displayed. The feature information 11 extracted in this way is led to the comparison processing section 9.

On the other hand, standard feature information 12-0, 12 is sent one by one from the dictionary section 6.
11, . That is, first, the standard feature information 12-0 is extracted and compared with the feature information 11, and if a correct match cannot be made, then the standard feature information 12-1 is extracted and compared. While such comparisons are being made, if a correct match is made as described below, the subsequent proposal processing is stopped and the standard figure category corresponding to the standard feature information that has been matched is set. The category to which the recognition target figure belongs is determined. In the illustrated case, the illustrated determination processing unit 10 receives the result of the comparison processing unit 9 comparing the bit correspondences, and determines the correct matching. Each standard feature information 12-0,1
2-1, . . . are provided with a feature information data section D, a mask section M, a category name storage section, and a count value storage section as shown in the figure.

The feature information data section D has N bits corresponding to the above-mentioned extracted feature information 11, and for example, the logic "1" or "0" set in the #1 bit corresponds to that of the above-mentioned extracted feature information 11. made to have the same meaning. Of course #2
The same applies to bits, #3 bits, . . . #N bits. The mask section M has N mask bits corresponding to each bit of the feature information data section D. For example, when the mask bit corresponding to the #0 bit is set to logic "1", this indicates that the contents of the #0 bit of the data section D are essential for the recognition process. Further, when the logic "0" is set, it indicates that the contents of the #0 bit of the data section D are ignored for recognition processing. The contents of the category name storage section indicate the category of the standard figure corresponding to the standard feature information. Then, if the above extracted feature information 11
When the corresponding standard feature information matches correctly, that is, there is a hit, the contents of the category name storage section are used as the answer, that is, the recognition result of the corresponding figure to be recognized. Furthermore, the contents of the count value storage section are incremented by 11 when the standard feature information is hit. At this time, there is a limit to the bit width of the count value storage section. Therefore, as described above, the contents of the count value storage section of each standard feature information 12-0, 12-1, ...... are increased to 11, and that of any one standard feature information is the upper limit of the count. When the value is reached and a carry occurs, the contents of the count value storage section of all standard feature information including the own standard characteristic information are decreased by 1. Therefore, each standard feature information 12-0, 12-1, . . . at a certain point in time when recognition processing is actually performed.
The contents of the count value storage section . . . represent the number of times the standard feature information has been correctly hit or the hit rate. Each standard feature information 12-0, 12-1, .・・・．
・・・． Since ... is prepared as described above, now 1
Standard feature information 12 has (D,,D2,...DN) (M,,M2,...MN) as data part D and mask part M. Then, the extracted feature information 11 is (F,,F2,...
...FN), the comparison processing unit 9
is MATCH=F, Yu (D.・M,) + F2 Yu (D2・M
2) Perform a comparison of 10...10FN fields (DN/MN).

Note that ■ represents exclusive OR logic, * represents AND logic, and 10 represents OR logic. When MATCH is "0" as a result of the above comparison, it is assumed that the determination processing unit 1 has been correctly matched. Then, the contents of the category name storage section shown in the matched standard feature information are output as recognition results, and in the case of the present invention, the contents of the count value storage section shown in the standard feature information are fully output. Do 1. Also, if a shift occurs at this time, the contents of the count value storage section in all standard feature information are all decremented by 1 as described above.
When MATCH is "1" as a result of the above comparison, the determination processing unit determines that there is no matching by one river, and reads out the next standard feature information as described above.

When a manufacturer ships a figure recognition device, standard feature information that is generally more likely to be hit is stored in the high-speed access memory 7 so that matching can be achieved at an earlier point in time, while the standard feature information that is generally more likely to be hit is stored in the high-speed access memory 7. Standard feature information that can be handled, that is, standard feature information that is likely to have a low hit rate, is stored in the low-speed access memory 8.

That is, the standard feature information 12-1, 12-1, .・・・・・・・・・
The comparison process is performed in the order of
The standard feature information is read out from the memory 8 and comparison processing is performed. However, due to special characteristics such as the writing habits of users at shipping destinations and the infrequent use of alphabetic characters, standard feature information that generally has a high hit rate for manufacturers may have a high hit rate at shipping destinations. do not have.

In order to solve this problem, the present invention has a count value storage section as described above, which is updated based on the result of recognition processing at the shipping destination. That is, for example, at night, the high-speed access memory 7 built in the data processing unit 2
The contents are saved on a low-speed access memory 8 such as a floppy disk. Then, in the morning, etc., the data is loaded from the floppy disk onto the high-speed access memory 7. At the time of this loading, the contents of the count value storage section for each of the standard feature information are checked, and the standard feature information is stored in the order of the count value, that is, the standard feature information with the highest hit rate, for example, at a younger address.
By repeating this process at the time of loading, the order of arrangement corresponds to the special circumstances of the shipping destination, and recognition processing can be performed in less time in response to the special circumstances of the shipping destination. .

As explained above, according to the present invention, it is possible to obtain an arrangement in the dictionary according to the circumstances of the shipping destination, and it is possible to reduce the average recognition processing time.

[Brief explanation of the drawing]

The figure shows the configuration of an embodiment of the present invention. In the figure, 1 is a figure recognition device, 2 is a data processing unit, 5 is a feature extraction unit, 6 is a dictionary unit, 7 is a high-speed access memory, 8 is a low-speed access memory, 9 is a comparison processing unit, and 11 is an extracted unit. Characteristic information 12-0, 12-1, . . . represents standard characteristic information.

Claims (1)

[Scope of Claims] 1. A dictionary unit that stores feature information extracted in advance for each of a plurality of standard figures, and a feature extraction unit that extracts features for the recognition target figure, and a feature extraction unit that extracts features from the feature information extracted by the feature extraction unit. a comparison processing section that checks the matching state of the feature information given by the dictionary section and the standard feature information given from the dictionary section; and a judgment processing section that determines a category based on the matching processing result in the comparison processing section; The figure recognition device that recognizes the target figure based on the matching processing result includes at least a high-speed access memory and a low-speed access memory as the dictionary unit, and corresponds to the standard feature information stored in the dictionary unit. and includes a count information storage section for counting the number of hits for the standard feature information, and updates the content of the count information storage section for the standard feature information in response to the determination by the determination processing section, and updates the count information for the standard feature information. Based on the contents of the storage section, the standard feature information with a high number of hits is accessed using the above-mentioned high-speed access method.
A figure recognition device characterized in that it is configured to be stored on a memory. 2. A graphical representation according to claim 1, wherein the high-speed access memory is a memory built into a data processing section of the graphical recognition device, and the low-speed access memory is an external storage file device. recognition device. 3 the contents of said fast access memory are stored by loading the contents of said slow access memory;
At the time of loading, the contents of the count information storage section are referred to, and a plurality of standard feature information whose contents of the count information storage section indicate a high value are loaded onto the high-speed access memory. A figure recognition device according to claim 1 or 2. 4. Claims 1 to 3, wherein the high-speed access memory stores the loaded standard feature information in descending order of the value indicated by the content of the count information storage section. The figure recognition device according to any one of the above.