JPH0262684A - On-line handwritten character recognizing device - Google Patents

Abstract

PURPOSE:To miniaturize the size of a feature dictionary and to recognize a character at a high speed by storing a compare test code in an appropriate character recognizing system in the feature dictionary for each part of one character and recognizing one character in plural character recognizing systems. CONSTITUTION:A compare test register 7 to store the intermediate information of a compare test, namely, the information to indicate which part of the character is currently compare-tested, the number of the inconsistencies between a character in a dictionary 8 and the character to be recognized, namely, a rejection number, etc., from compare test circuits 6 and 12, a buffer 9 for a compare tested result to store the data of the compare tested result from the compare test circuits 6 and 12, and an input character judging circuit 10 to receive the data of the compare tested result from the buffer 9 and judge what the input character is are formed. Here, the compare test code in the character recognizing system appropriate for each part of the character is stored in the feature dictionary 8.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to an online handwritten character recognition device.

[Summary of the invention]

This invention aims to miniaturize the feature dictionary and speed up character recognition without reducing character recognition accuracy by performing continuous character recognition processing for each part of a character in an online handwritten character recognition device. It is something that

[Conventional technology]

When iM-weaving handwritten characters online, the characteristic data of the characters input from the input means for handwritten characters is compared with the characteristic data of each character stored in the characteristic dictionary, and the stored characters are The character that is most similar to @ is determined to be the input character.

In this case, there are several methods for performing a comparison test between the input characters and the characters registered in the feature dictionary, one of which is called the basic stroke method.

This basic stroke method is as follows: 1. As shown in Fig. 4A, the input handwriting of the continuous line from the point where the pen is down to the point where it is amp is inputted by the point P O on the handwriting. ,Pt,・
..., Pn and its time series information are approximated by a polygonal line (B in the figure).

(2) "Stroke recognition" is performed by comparing the basic shape of a stroke defined in advance, that is, the "basic stroke type" as shown in FIG. 5, with the polygonal line in item 1.

Based on the results of sections iii and ii, the input stroke is converted to the code number of the basic stroke type that has a similar bulge (in the case of the polygonal line in FIG. 4B, the coat number is "6").

iv, for all strokes of one character, i =
Step iii is repeated, and the obtained sequence of code numbers of strokes is used as characteristic data of the input character.

■In the feature dictionary, all characters are represented by feature data that is a sequence of code numbers (comparison test codes) for basic scripts and loaf types, and this is used for a comparison test with the feature data obtained in section iv. Then, the most similar character is separated by r+1 from the input character.

This method is based on the premise that one stroke corresponds to one stroke of a character.

According to this method, by arranging the code numbers of the basic stroke types in the feature dictionary in the stroke class for each character, it is possible to compare this with the order of the code numbers of the basic stroke types of the input characters. Since the characters input can be identified, the dictionary can be made smaller and the time required for comparison, that is, the recognition time, can be shortened.

Also, there is a sentence ′-? that is different from the basic stroke method mentioned above. As a recognition method, there is a linear approximation method.

This is as follows: 1. Similar to the basic stroke method, the input handwriting (see Fig. 4) is approximated by a polygonal line (B in the same figure).

11. Obtain the length of this polyline, and
The direction is quantized, for example, in 16 steps as shown in FIG.

Based on the results of sections iii and ii, each straight line of the polygonal line is replaced with the code number of the closest direction (1 in FIG. 4B) (for polygonal lines, rOJ, rlOJ).

iv, for all strokes of one character, i =
Step iii is repeated, and the obtained arrangement of coat numbers in the linear direction and the length of each tL line are used as characteristic data of the input character.

■In the feature dictionary, multiple characters are represented in advance by data that excludes the code number of the straight line direction and the length of the straight line, that is, feature data that is a sequence of comparison test codes. Compare and test the obtained characteristic data,
Green also determines similar characters as input characters.

This is the method.

In the case of this linear approximation method, the feature dictionary is larger and the recognition speed is slower than in the basic stroke method, but it is advantageous in the case of characters that are easy to input because multiple strokes are made in one stroke. In other words, for example, in the character ``Ito'', the lHs of ``Gen'' is likely to be written as 1 s] loaf, and in that case, recognition accuracy will drop significantly in the basic stroke 0 - evening style, but linear approximation If this method is used, the recognition accuracy will not decrease much.

Literature: [Nikkei Electronics 4, December 5, 1983 issue] [Issues that the invention aims to solve Now, as mentioned above, when character recognition is performed using only the basic stroke method, the dictionary can be made smaller, but the recognition accuracy On the other hand, when character recognition is performed using only the linear approximation method, there is a problem that the recognition accuracy is high but the dictionary becomes large.

Therefore, two feature description systems, the basic stroke method and the linear approximation method, are used in one character recognition device, and the appropriate feature data description method is selected for each character, thereby reducing the size of the dictionary. It is also considered to improve the weave rate.

However, in this case, which feature description system is used to recognize the input character is determined by the external finger point, the number of strokes of the input character, or each group in the dictionary, and a comparison test is performed using a single character in the dictionary. During this time, only one feature description system was used.

For this reason, it was not possible to select an appropriate feature description system for each part of a character, so a lot of effort was required to create a dictionary.

This invention attempts to solve the above problems.

(Means for solving the problem) By the way, if we think about the parts of a single character, we find that there are parts that can be sufficiently recognized using the basic stroke method, and parts that are insufficient for the basic stroke method, and parts that can be recognized using the linear approximation method. In many cases, it is mixed with parts where it would be more reasonable to do so.

Therefore, in this invention, even if it is a single character,
Focusing on the fact that there is a consistent character recognition method for each part, we store the comparison test code of the appropriate character recognition method for each part of a character in a feature dictionary, and recognize one character. This is done using multiple character recognition methods.

[Effect]

Appropriate character recognition methods are used for each part of a single character, making it possible to reduce the size of the dictionary and speed up character recognition without reducing recognition accuracy.

〔Example〕

In Fig. 1, (1) is a tablet for inputting handwritten characters, and (2) is a preprocessing circuit that removes noise components from input handwriting input from tablet (1), transforms it into digits, and approximates it to a polygonal line. . And (3) is a polygonal line buffer that stores data converted into polygonal lines by the preprocessing circuit (2).

(4) is a conversion circuit that converts the polyline data in the polyline buffer (3) into a basic stroke type code number as shown in Figure 5, and (5) stores the code number from the conversion circuit (4). buffer, (6) is the feature dictionary (
This is a basic stroke type comparison test circuit which reads the comparison test code from 8) and compares and tests it with the basic stroke type code number from the buffer (5).

(10) is a conversion circuit that converts the polygonal line data in the polygonal line buffer (3) into data indicating the code number of the straight line direction and the length of the straight line as shown in FIG. 6; (11)
is a buffer 7 that stores the code number from the conversion circuit (10), and (12) is a linear approximation that reads a comparison test code from the feature dictionary (8) and compares and tests it with the code number in the straight line direction from the buffer (11). This is a comparison test circuit of the method.

And (7) is the information from the comparison test circuits (6) and (12) during the comparison test, that is, the information about which part of the character is currently being compared, and the information about the character i) (81 characters and A comparison test register that stores the number of characters that do not match with the characters to be recognized, that is, the number of failures, etc. (9) is a character that is the result of the comparison test from the comparison test circuits (6) and (12). This is an input character judgment circuit that judges the input character.

Here, the feature dictionary (8) stores comparison test codes for character recognition methods suitable for each part of a character.

In other words, for example, as shown in Figure 2, in the case of the character ``Saku'', the ``Ha'' part of this character is compared using the basic stroke method, so the basic stroke method that indicates 1 ppJ is used. In other words, for example, in Figure 5, the neo code number arrangement rlJ, r4J
, r4J, and r6J are stored as feature data. Next, in order to perform the comparison test using the linear approximation method, the “Gen” part that follows “Ha” is code L, which indicates the linear approximation method.
is memorized, followed by the sequence of linear approximation comparison test codes indicating "Gen", that is, the sequence of code numbers shown in Figure 6: rlOJ, ri4J, riot, rOJ
.

114'' and the length of the straight line. Then, in order to perform the comparison test again using the basic stroke method for the small part that follows "Gen" after Hime, code S is memorized, and then the code S for the comparison test using the basic stroke method that indicates "small" is memorized. Feature data that is a sequence is stored. In addition, in the case of a character for which the first part of the character is to be processed by the linear approximation method, a code L indicating the linear approximation method is stored first. Also, the codes S and L are, for example, “
0'' or rlJ.

Next, the operation of an embodiment of the present invention will be described with reference to FIGS. 1 to 1B. In the above description, an example will be explained in which the first character to be compared and tested is the character "Saku" shown in FIG. 2.

First, characters input to the tab left (1) are supplied to a preprocessing circuit (2) to form a polygonal line, and data that eliminates the polygonal line are supplied to a polygonal line buffer (3).

Then, step (10) of the flowchart shown in FIG.
0), the comparison test register (7) is initialized. And from step (100) to step (101)
In this step (101), the stroke method comparison test circuit (6) calculates the first ratio f of the first character registered in the dictionary (8) to be compared and tested.
9. The verification code will be loaded.

Next, the process proceeds from step (101'') to step (102), and in this step (102), it is determined whether the read comparison test code is a code indicating a linear approximation method.In this case, the first comparison The verification code includes a code indicating ``1-1'' which is the first stroke of ``rope'', for example, the code number rl of the basic stroke type shown in Figure 5.
J is stored, and it is determined that this is not a code indicating the linear approximation method, and a comparison test using the basic stroke method is performed from step (102) to step (103).
Proceed to. And, in this step (103),
The data of the first stroke stored in the polygonal line buffer (3) is supplied to the stroke method comparison and verification circuit (6) via the conversion circuit (4) and buffer (5), and in this comparison and verification circuit (6), , is compared with the code number "1" read from the dictionary (8).

Then, from step (103) to step (10
Proceeding to step 4), in this step (104) it is determined whether the dictionary contains the next comparison test code. And in this example, the following comparison test code:
Since the code number "4" indicating "↓" which is the second stroke of "search" is stored, it is determined that the next comparison test code exists, and from step (104) to step (101
), this code number "4" is the comparison test circuit (6)
read by. Then, in the same manner as described above, steps (102) to (104) are executed, and the next comparison test code is read. In addition, step (10
In 3), if the total number of failed comparison test results up to that point reaches a predetermined value, for example, 3 or more, that character will be rejected even if the comparison test for one character is in the middle. It is determined that the test has passed, and the process proceeds to step (111), which will be described later.

In this way, if step (103) is not determined to be a failure, steps (101) to (104) are continued until the feature data D1 of the basic stroke method is completed, that is, the code L indicating the linear approximation method is read. It is repeated until.

Then, this code L is read into the comparison test circuit (6), and in step (102), when it is determined that the code indicates the linear approximation method, the data indicating the comparison test results up to that point are read into the comparison test circuit (6). ) to the comparison test register (
7). Then, the process proceeds from step (102) to step (105).

Then, in step (105), the linear approximation comparison test circuit (12) reads the next comparison test code after the code L from the dictionary (8), and also reads the contents of the register (7). Then, the process proceeds from step (105) to step (106), and in step (106), it is determined whether the read comparison test code is code S indicating the basic stroke method.

Then, when it is determined in step (106) that the read comparison test code is not code S, this step (
106), the process proceeds to step (107).

Then, the linear approximation type comparison test circuit (12) refers to the contents of the recognition register (7) and transfers the data of the next line to be compared and tested from the line buffer (3) to the conversion circuit (10), Capture via buffer (11). Then, this imported data and the comparison test code read from the dictionary (8) are compared and tested by the comparison test circuit (12). In this step (107), if the total number of failures in the comparison results up to that point is greater than or equal to a predetermined value, the process proceeds from step (107> to step (111), which will be described later). If so, the process proceeds from step (107) to step (108), and in this step (108) it is determined whether there is the next comparison test code.Thus, the feature data D2 of the linear approximation method is completed. Step (105) until the code S indicating the basic stroke method is read.
. . . (10B) are repeated.

Then, in step (105), code S is read into the comparison test circuit (12), and in step (106), when it is determined that the read code is code S, the recognition results etc. up to that point are displayed. The data is compared to the test circuit (1
2) to the register (7). Then, the process proceeds from step (106) to step (101).

Then, the stroke method comparison test circuit (6) is rejected!
(Continues from 81 with the next comparison verification code of code S, and also reads the contents of register (7), and proceeds from step (101') to step (102) to step (103). , comparison test circuit (61
In this step (103), register (7)
With reference to the contents of ) is compared with the comparison test code read from ).

If the character is not determined to be rejected in step (103), steps (101) to (104) are repeatedly executed until the basic stroke method feature data D3 is completed.

Then, it is determined that the feature data D3 has been completed in step (
If the judgment is made in step 104), the process proceeds from step (104) to step (109), and the comparison test results up to that point are compared to
It is supplied from the test circuit (6) to a comparative test result buffer <9) and stored. Then, next step (109
) to step (111), and this step (11
In step 1), it is determined whether or not there is a character to be compared next in the dictionary, and if it is determined that there is, step (1) is performed.
The process proceeds from step (11) to step (100), where a comparison test is performed in the same manner as described above.

Now, in the above explanation, we have shown an example of the character ``Saku'' where the last part of the character was registered using the basic stroke method, but if the last part of the character is registered using the linear approximation method, then , When the end of the last part of the character is determined in step (10B), the process proceeds from step (10B) to step (110), and the comparison test results up to that point are sent from the comparison test circuit (12). buffer (9
) and stored. Then, step (
Step (110) proceeds to step (111), and in this step (111) it is determined whether or not there is a character to be compared next.If it is determined that there is no character, step (11
1 to step (112). Then, in this step (112), the comparison test results up to that point are supplied from the buffer (7) to the input character judgment circuit (13). In this input character judgment circuit (13),
Among the comparison test results, the character with the least number of failures is determined to be the input character.

As described above, according to this embodiment, a code indicating an appropriate character recognition method, that is, the basic stroke method or the linear approximation method, is assigned to each part of one character using the characteristic acid IF.
(81), and according to this, the comparison test is performed for each part of the character in a method suitable for that part, so for example, if the number of bits of the comparison test code is 8 pits, each character It is possible to set 255 types of codes for each recognition method, which allows for recognition of input characters without increasing the number of bits of the comparison test code compared to when only one type of character recognition method is used. It is possible to reduce the size of the dictionary and speed up character recognition without reducing accuracy by 11%.

In addition, in the above example, the basic stroke method and
Although two character recognition methods, a linear approximation method and a linear approximation method, have been adopted, it is not limited to these two methods, and other character recognition methods can also be used.

〔Effect of the invention〕

As described above, according to the present invention, a comparison test code of a character recognition method that is consistent with each part of a character is stored in the feature dictionary, and a comparison test is performed for each part of the character according to the code. The present invention has the advantage that the dictionary size can be reduced and the character recognition speed can be increased without deteriorating the character recognition accuracy.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG. 2 is a block diagram of an embodiment of the present invention.
Figure 3 shows an example of data to be registered in the dictionary, Figure 3 shows a flowchart, Figure 4 shows a polygonal line approximation diagram, Figure 5 shows basic stroke types, and Figure 6 shows straight line direction codes. It is. (1) is a tablet for inputting handwritten characters, (2) is a preprocessing circuit, (3) is a polyline buffer, (6) is a basic stroke comparison test circuit, (8) is a feature dictionary, (12)
is a comparison test circuit using a linear approximation method. Koshirae rta story (2) Joy 4 diagram straight J turtle direction code Figure 6 base L or draw 99 if code Figure 5

Claims (1)

[Scope of Claims] A means for inputting handwritten characters; a feature dictionary storing, for each part of a character, a comparison verification code using a character recognition method suitable for that part among a plurality of character recognition methods; A comparison test circuit is provided which compares and examines the input character and the character registered in the feature dictionary according to the comparison test code registered in the dictionary, An online handwritten character recognition device that performs a comparison test using a character recognition method according to a comparison test code of a feature dictionary.