JPH03217995A - Handwritten character recognizer - Google Patents

Abstract

PURPOSE:To improve the recognizing speed of handwritten characters by defining previously the number of remaining strokes granted to the branches of a tree structure and comparing the defined number of strokes with the number of uninspected strokes of an input character. CONSTITUTION:A range of the number of remaining strokes granted to a branch is described to the relevant branch of a recognition dictionary 18. When a handwritten character is recognized, the range of the number of remaining strokes is compared with the number of uninspected strokes of the inputted handwritten character. Then a recognizing circuit 17 carries on the scan of the relevant branch and recognizes the inputted handwritten character only when the number of uninspected strokes of the handwritten character is kept within the range of the number of remaining strokes.

Description

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

[Summary of the Invention] The present invention provides a handwritten character recognition device that uses a tree-structured dictionary that does not perform major classification based on the number of input strokes, by defining the number of remaining strokes allowed on the branches of the tree structure and identifying the remaining strokes of input characters. The recognition speed is improved by comparing the number of verification strokes.

[Prior Art] In general, online recognition of handwritten characters involves the following two methods.
Processed in stages. That is, the first stage: various special @ (feature data) are extracted from the handwriting of handwritten characters input from a tablet or the like.

Second stage: Compare and test the extracted feature data with the features (verification code) of each character described in the recognition dictionary to detect characters with a sufficiently high degree of matching.

When performing this second stage processing, if the recognition dictionary is made into a tree structure, the capacity of the recognition dictionary can be reduced and the comparison test can be performed at high speed.

Figure 4 shows a simplified recognition dictionary with such a tree structure, and (3l) to (42) are the verification techniques (and the order of scanning of comparative tests when there is no cancellation of scanning of comparative tests). ), (51) to (56) are branches, and 01 to C7 are character codes of recognition results. Then, the character code Cn(n
=1 to 7) are terminal branches (34) to (36), (38), and (40) to (42) of techniques (31) to (42).
One for each is defined.

When a handwritten character is input and its feature data is extracted: First, in technique (31), a verification code is extracted from the recognition dictionary, and the feature data is verified in accordance with this verification code. Then, according to the test result of this branch (3l), scanning proceeds from the branch (51) to, for example, the branch (32), and similarly from the branch (52) to the branch (3l).
Branch to 3).

However, in this technique (33), if the degree of discrepancy between the verification code and the handwriting data is large, this technique (33)
) will be rejected.

Then, the scan is performed at the branch (52) between the branches (33) and (36).
) and move on to technique (36). In this technique (36), if the degree of match between the verification code and the feature data is high, this technique (36) is the last branch, so it is defined in this branch (36). The character code C, which is the character code C, is output as the recognition result of the input character.

Furthermore, in technique (38), if the degree of matching between the verification code and the handwriting data is high, character code C4 defined in technique (38) is output as the second candidate for the recognition result of the input character. The third candidate and subsequent candidates may be output in the same way.

[Problem to be solved by the invention]

However, in the tree-structured recognition dictionary described above,
Since unnecessary branches in which the number of strokes of the input character and the number of strokes of the verification character do not match are also verified, the recognition speed decreases.

This invention attempts to solve these problems.

[Means to solve the problem]

Therefore, in the present invention, the range of the number of remaining strokes allowed for the branch of the recognition dictionary is written, and when the handwritten character is recognized, the range of the number of remaining strokes is written in the branch of the recognition dictionary. The input handwritten character is recognized by comparing it with the number of unverified strokes, and only when the number of unverified strokes of the input handwritten character is within the range of the remaining number of strokes, scanning is continued on that branch. It is.

[Effect]

Unnecessary skill tests are no longer performed, improving recognition speed.

〔Example〕

In Fig. 1, when a character is written on the tablet (11), the output data of this tablet (11) is supplied to the pre-processing circuit (12) to remove noise, quantize it (reducing the polygonal line), etc. Preprocessing is performed, and the preprocessed data is supplied to the input buffer (l3) and stored.

Then, in the stroke feature extraction circuit (14), one stroke of the input character (
For each stroke (one stroke), the characteristics of that stroke are extracted as feature data, and this feature data is stored in the stroke feature buffer (15), as well as in the untested buffer (16).
) is incremented by "l".

In this way, characteristic data is generated for each stroke of one character input from the tablet (11), and this data is sequentially stored in the buffer (15), and the value in the buffer (16) is incremented.

When the input of one character is completed, the recognition circuit (l7)
At , the feature data stored in the puffer (15) is compared and tested with the test code in the recognition dictionary (18). Moreover, at this time, the value of the unverified buffer (16) is decremented by "1" every time the verification for one stroke is performed.

In this case, the recognition dictionary (18) has a partial structure dictionary for "Ki" and "Ki", for example. "Ki" has a partial structure dictionary for "Ki".

The remaining number of strokes is 2 to 4.

It has a partial structure dictionary for "father".

"Ken" has a partial structure dictionary of "Tree".

The remaining number of strokes is 5 to 8 strokes.

It has a partial structure dictionary for "^".

It has a "large" substructure dictionary.

When shown in a tree structure, it becomes a "tree" as shown in Figure 2.
Technique to test (21) → Branch (22) → Remaining number of strokes is 2
~ Technique to verify that it is a 4-stroke (23) → ``Like the technique to verify father j (24), techniques (21) to (24) continue, and the letter ``J'' is added to the branch (24). Code Cm is defined.

Also, branch (22) → branch that verifies that the number of remaining strokes is 5 to 8 (25) → branch that verifies “^” (26) →
Like the branch (27) that tests "large", branches (22) to (
27) continues, and the character code Ck for "ken" is defined in technique (27).

While scanning the dictionary (18), if a branch occurs or there is an instruction to test the remaining number of strokes such as instructions (23) or (25), the value of the buffer (16) is compared with the range of the number of remaining strokes described in the recognition dictionary (18), and when it is within the range, scanning proceeds to the next branch, and when it is outside the range, scanning proceeds to the next branch (adjacent branch). .

For example, when "sheets" is input, scanning is performed using technique {2
1}, proceeds to branch (23) via branch (23), and in this branch (23), the value of buffer (16) and branch (2
The range of remaining strokes "2 to 4 strokes" described in 3) is compared, and in this case, it is within the range, so scanning is performed on the branch (23
), progresses to technique (24), passes the test of technique (24), and extracts the character code Cm for "piece".

Also, when “inspection” is input, the scanning is performed on the branch (21
), proceed to branch (23) through branch (23), and in this technique (23), the value of buffer (l6) and branch (23)
) is compared with the range of remaining strokes "2 to 4 strokes", and in this case, it is outside the range, so scanning is performed at branch (23).
Proceed from branch (22) to technique (25), and this branch (
In 25), the value of buffer (16) and the edge (25)
The range of the number of remaining strokes described in ``5 to 8 strokes'' is compared, and in this case, it is within the range, so the scanning proceeds from branch (25) to techniques (26) and (27), and then to ``test''. character code C
k is taken out.

In this way, when the verification is performed in order and the character code of the recognition result is obtained, it is output from the recognition circuit (17) and stored in the output buffer (19).

FIG. 3 shows an example of a processing routine when character recognition as described above is performed by software processing.

That is, when a one-character sentence is input from the tablet (11), the verification process starts from step (101), and in step (102), the total number of strokes of the input character is stored in the unverified buffer (16). be done.

Next, in step (111), it is checked whether the command described in the dictionary (18) is a test for the range of remaining strokes, and if this is the test, the process proceeds to step (112) and this In step (112), it is tested whether the value of the buffer (16) is within the range of the number of remaining strokes described in the dictionary (18),
When it is within the range, the process returns from step (112) to step (111). These processes correspond, for example, to the scanning proceeding from branch (22) through branch (23) to branch (24).

Further, in step (111), if the instruction is not for testing the range of remaining strokes, the process proceeds to step ('1
11), the process proceeds to step (121), and in this step (121), it is checked whether the command described in the dictionary (l8) is for testing the stroke shape. (12
1) to step (122), and this step (1)
In step 22), the stroke shape is verified and checked to see if it passes.

Then, when the result is passed, the process proceeds from step (122) to step (123), and this step (123)
In step (122), the remaining number of strokes stored in the unverified buffer (16) is decremented by the number of strokes to be verified in step (122), and then the process proceeds to step (111).
Return to These processes correspond, for example, to the scanning passing through the branch (24).

Further, in step (121), when the stroke shape is not verified, the process is performed in step (12l,).
Proceed to step (131) from this step (131).
), another test is performed to check whether it passes, and if it passes, the process proceeds from step (131) to step (132), and this step (132)
At step (132), it is checked whether there is a next test, and if there is, the process moves from step (132) to step (1
Return to 11).

Further, in step (132), if there is no next test, the process continues from step (132) to step (1).
Proceeding to step 33), this step (133) is executed when there are no more tests on the current branch, so the obtained candidate character code is stored in the output buffer (l9), and the process then returns to step Proceed to (141).

Further, in step (112), the buffer (16)
When the value of is not within the range of the number of remaining strokes described in the dictionary (18), the process proceeds from step (112) to step (141). Furthermore, step (122),
In (131), even if the test fails, the process continues from step (122) and (131) to step (141).
Proceed to.

Then, in step (141), it is checked whether there is a neighboring branch (node), and if there is, the process proceeds from step (141) to step (142), and in this step (142), the scanning is performed on the neighboring branch. At the same time, the remaining number of strokes in the unverified buffer (16) is also returned, and then the process returns to step (111).

In addition, in step (141), if there is no adjacent branch, the process proceeds from step (141) to step C151.
), and in this step (151), the character code of the output buffer (19) is supplied to a display means (not shown) to display the corresponding character, and in step (152)
ends this routine.

Thus, according to the present invention, handwritten characters can be recognized using a tree-structured recognition dictionary, but in this case,
In particular, according to this invention, the range of the number of remaining strokes allowed for that branch is written on the branch of the tree structure, and the range of the number of remaining strokes allowed for that branch is compared with the unverified number of strokes of the input character. No scanning or verification will be performed. Therefore, the recognition speed can be increased, and interference can be reduced, so that the recognition rate can be improved.

In addition, in the above description, the middle of the branch, for example, the branch (26) and (
27), it is also possible to write the range of the number of remaining strokes allowed and perform the verification. Further, a field for determining whether or not to test the number of remaining strokes can be provided as a recognition test command.

〔Effect of the invention〕

According to this invention, the range of the number of remaining strokes allowed for the branch of the tree structure is written in advance and compared with the unverified number of strokes of the input character, so that the branches that do not correspond to the number of remaining strokes are scanned. and examinations will no longer be conducted. Therefore, the recognition speed can be increased, and interference can be reduced, so that the recognition rate can be improved.

[Brief explanation of drawings]

FIG. 1 is a functional configuration diagram of an example of this invention, FIG. 2 is a diagram showing a dictionary structure thereof, FIG. 3 is a flowchart showing an example of this invention, and FIG. 4 is a diagram for explaining the same. (1l) is a tablet, (l2) is a preprocessing circuit, (13
) is the input buffer, (14) is the stroke feature extraction circuit, (l5) is the stroke feature buffer, (16) is the unverified buffer, (17) is the recognition circuit, (18) is the recognition dictionary, and (19) is the output buffer. It is. Agent Hidemori Matsukuma Figure 2

Claims (1)

[Claims] In a recognition device that recognizes handwritten characters input using a tree-structured recognition dictionary, each branch of the recognition dictionary describes a range of the number of remaining strokes allowed for that branch, When recognizing, the range of the remaining strokes is verified with the unverified stroke count of the input handwritten character, and only when the unverified stroke count of the input handwritten character is within the range of the remaining strokes, the A handwritten character recognition device that recognizes the input handwritten characters by continuously scanning them.