JPS63238680A - Matching method and dictionary retrieving method for symbol string - Google Patents

Abstract

PURPOSE:To simplify matching and flow control of dictionary retrieval by not only inserting a special symbol but also adding control information to and on symbol strings registered in a register and a dictionary. CONSTITUTION:When it is checked whether a first symbol string A in a register 11 coincides with a preliminarily registered second symbol string B or not, the special symbol is inserted to a second symbol string B and procedures of a correspondence relations between symbols to be compared are changed in accordance with this symbol. Not only the special symbol is inserted but also control information is added to and on each dictionary symbol string registered in the dictionary, and matching between dictionary symbol strings and an unknown symbol string is controlled in accordance with this control information and procedures of correspondence relations between symbols to be compared are controlled by the special symbol, and the flow of dictionary retrieval after matching between dictionary symbol strings and the unknown symbol string in the case of coincidence between them is controlled in accordance with control information added to the dictionary symbol string. Thus, matching and control flow of dictionary retrieval are simplified.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a symbol string matching method and a dictionary search method for searching for a symbol string that matches an unknown symbol string from a dictionary in which symbol strings are registered. More specifically, the present invention relates to a symbol string matching method and dictionary search method suitable for matching a feature vector of an input pattern with a feature vector of a recognition dictionary and searching a recognition dictionary in a pattern recognition device.

[Prior art]

Devices that recognize patterns such as characters are equipped with a dictionary in which feature vectors (feature sequences) of various patterns are registered, and the dictionary is searched for a feature vector that matches the feature vector (unknown feature sequence) extracted from the input pattern. By doing so, a candidate pattern for the input pattern is determined.

By the way, when matching feature vectors together in such a dictionary search, due to the effects of noise, pattern deformation, etc., it is necessary to uniquely determine the correspondence of all elements of the feature vectors and compare the elements to achieve matching. , the recognition rate decreases when a mismatch (establishment, failure) is determined. As a countermeasure to this problem, when matching with the feature vector of the dictionary, some elements of the feature vector of the input pattern may be ignored to determine whether they match or do not match, or the procedure for determining the correspondence of the elements to be compared may be changed. It is valid.

However, even if this matching method is controlled,
If only one feature vector is registered in the dictionary for one type of pattern, it cannot cope with the deformation of the pattern, so in order to be able to sufficiently deal with pattern deformation, it is necessary to register multiple features for one type of pattern. It is necessary to register the vector in the dictionary.

Furthermore, when selecting one from multiple candidates (for example, when determining whether the input pattern is α, β, or neither), it is necessary to register the feature vectors of each of the multiple candidates in the dictionary. There is. In this case, in addition to controlling the matching method, it is necessary to devise ways to appropriately control the flow of dictionary search after a match is made using the feature vector of a certain dictionary.

However, conventionally, the above-mentioned matching method, comparison correspondence relationship, and dictionary search flow have been controlled using state transition diagrams (
For example, IEICE technical research report PRL85-35, automatic dictionary creation method in Kato r stroke structure analysis method J
), the description for the control is extremely complicated, and the control process becomes complicated.

〔the purpose〕

An object of the present invention is to simplify the description of changing the correspondence between elements to be compared in matching symbol strings such as pattern feature vectors with similar symbol strings registered in advance, and to simplify symbol string matching with simplified control. Another object of the present invention is to provide a method for searching a dictionary of symbol strings, such as a pattern recognition device, by easily describing the flow of matching, comparison correspondence, and dictionary search in a variety of ways and appropriately. The purpose of the present invention is to provide a dictionary search method that can be used in a variety of ways.

〔composition〕

In the symbol string matching method of the present invention, when checking whether a first symbol string matches a second symbol string recorded in advance, a special symbol is inserted into the second symbol string. The procedure for comparing correspondence relationships is changed according to this symbol.

Furthermore, in the dictionary search method of the present invention, when a special symbol is inserted into each dictionary symbol string recorded in the dictionary, control information is added to each dictionary symbol string, and the matching method between the dictionary symbol string and the unknown symbol string is changed. In addition to controlling according to the control information added to the dictionary symbol string, the procedure for the correspondence of the symbols to be compared is controlled by a special code inserted in the dictionary symbol string, and the dictionary symbol string and the unknown symbol string are When the matching results in a match, the flow of the subsequent dictionary search is controlled in accordance with the control information added to the dictionary symbol string.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a functional block diagram of a symbol string matching processing device according to an embodiment of the present invention. In FIG. 1, the A register 11 contains a feature vector (hereinafter referred to as Code string A) is stored starting from address 0.

In the B register 12, a feature vector (hereinafter referred to as code string B) as a second symbol string, which has been recorded in advance, is stored at address 1 and thereafter. This code string B
In this embodiment, one or more special codes are inserted into 0 and used to change the order in which the codes in code string A are compared, as will be described later. shall be. Also, address 0 of B register 12
stores a code indicating whether the comparison is partial correspondence or total correspondence. In this embodiment, in the case of one part correspondence, the code of address 0 is 1, and in the case of whole O1 correspondence, the code of address 0 is 1.
shall be.

The read address of register 11 is A address counter 1
4, and the read address of the B register 12 is indicated by the B address counter 15.

Matching of codes read from both registers 11.12,
A comparison circuit 16 compares the discrepancies, and the comparison result is input to the control circuit 17. The control circuit 17 contains the code of address O of the B register 12 (0 or 1) and the special code read from the B register 12 (i.e. 0).
is also input.

In the control circuit 17, the flag 13 is set to "1" or "0" according to the special code read from the B register 12 and the comparison result of the comparison circuit 16, and the address counter 14.15 is set by referring to the state of the flag 13. Control update behavior.

An outline of the operation of this embodiment is as follows. As an initial state, the A address counter 14 is set to O, and the B address counter 1 is set to
5 as 1, the code string A of the A register 11 is read from address 0, and the code string B of the B register 12 is read from address 1, and the corresponding codes are compared by the comparison circuit 16. In addition.

If a special code, that is, O, is inserted at the first address 1 of code string B when a mismatch is detected, A
Set the address counter 14 to +1.

Code string A is compared starting with the code at address 1.

While the special code, that is, 0, is not read from the B register 12, each time a match is detected in the comparison circuit 16, the address counters 14 and 15 are each incremented by 1 and the register 11 is read out.
．． The codes are sequentially read out from 12 and compared by a comparison circuit 16. When the comparison circuit 16 detects a mismatch, the code at address 1 of code string B is checked. If the code is not O, it is assumed that code string A and code string B are not matched and the process ends; however, if the code is O, the A address counter 14 is set to 1 and the B address counter 14 is set to 1.
5 is set to 1, and the code starting from address 1 of code string A is compared and restarted from the beginning. When a mismatch is detected again, the A address counter 14 is set to 2 and B is set to 1.
Address counter 15 is 1), address 2 of code string A
The following code will be compared. This is repeated every time a mismatch is detected, that is, this is to deal with misalignment patterns and deformed patterns.

The match between code string A and code string B progresses, the B address counter 15 becomes i (i>1), and the B register 12 becomes 0.
Suppose that is read out. That is, address i of code string B
A special code has been inserted. At this time, the B address counter 15 is incremented by 1, and the address of code string B (i+1) is
The code is read out from register 12. Then, the A address counter 14 is sequentially ++ until the same code as the code at address (i+1) of code string B is read out.
1, and read the codes one after another from the A register 11.
During this time, updating of the B address counter 15 is stopped. When the two codes match, the address counters 14 and 15 are each incremented by 1 again, and the next code is read from the registers 11 and 12. Thereby, unnecessary codes in code string A can be excluded from comparison with code string B.

When the final code of code string B is reached, B register 1
Check address O of 2. If the code at address 0 is 0 (partial correspondence), it is assumed that code string A and code string B are matched even if the code of code string A has not reached the end. When the address O of the B register 12 is 1 (general correspondence), it is assumed that matching is achieved only when the code string A also reaches the final code, and matching is not achieved otherwise.

Next, a specific example of the matching method will be explained.

Fig. 2 shows an example of code string A, and Fig. 3 shows an example of code string B.
Code string B in FIG. 3, which shows an example, shows an example of partial correspondence (address 0=0).

First, code string A in Figure 2 and code string B in Figure 3 (A)
The matching process with .

FIG. 4 shows the transition of addresses, codes, and flags in this case. First, set the flag to 1 (initial condition), read and compare code 4 at address O of code string A and code 4 at address 1 of code string B (step 1) 1 If the two codes match, the code string The code at address 2 of code string B is read out (step 2). Since the code at address 2 in code string B is 0, code 2 at the next address 3 is read out and the flag is set to O (step 3). Then, code 1 at address 1 of code string A is read out (step 4), but this code does not match code 2, so it is ignored.
Next, read code 2 at address 2 and compare it with code 2 at address 3 in code string B (step 5).0 Since both codes match, read the code at address 4 in code string B and set the flag to 1. Return (Step 6). Since the code at address 4 of code string B is O, code 4 at the primary address 5 is read and the flag is set to 0 again (step 7). Then, code 1 at address 3 of code string A
(Step 8), but since this code does not match Code 4, it is ignored, and Code 4 of the primary address 4 is read and compared with Code 4 of Address 5 of code string B to find a match (Step 9). , because code string B ends at address 5.

Checking the code at address 0 In this example, the code at address O is 0, indicating a partial correspondence, so even if code string A does not reach the final address 7, code string A matches code string B. Suppose that

Next, code string A in Fig. 2 and code string B in Fig. 3 (b)
The matching process with .

The transition of addresses, codes, and flags in this case is shown in FIG. 5. In FIG. 5, steps up to step 5 are the same as in FIG. 4. In step 6, code 3 at address 4 of code string B is read and the flag is set to 1 (because a match was found in step 5). In the next step 7, code 1 at address 3 of code string A is read out, and code 1 at address 4 of code string B is read out.
Compare with code 3. Since the two do not match, the matching process is terminated at this point, and code string A becomes code string B.
It is determined that it does not correspond to

Next, code string A in Fig. 2 and code string B in Fig. 3 (c)
The matching process with .

This is an example in which a special code, ie O, is also inserted at the beginning of code string B (address 1). It is searched repeatedly with different changes. Address, code in this case.

FIG. 6 shows the transition of the flags.

In FIG. 6, the flag is initialized to 1 in step 1, and address 0 of code string A and address 1 of code string B are set to 1.
However, since the code at address 1 of code string B is 0, in the next step 2, code 4 at address 2 of code string B is read out, and the comparison starts from there. The matching process progresses, and in step 9 code 1 at the final address 5 of code string B is read out, and in the next step 10 code 3 at address 5 of code string A is read out. Code I
Because the JIF code is 3, it would normally be determined that code string A and code string B do not correspond, but since address 1 of code string B is code = 0, the address of code string B is determined in step 12. Code 4 of address 1 of code string A is read out again in step 13, and the matching process is repeated. In this way, (4, 4, 1) of code string B becomes address 4 (=4), address 6 (=4), and address 7 of code string A.
(=1), and as a result, it is determined that the correspondence between code string A and code string B is established.

FIG. 7 shows an example of code string B in the case of global correspondence (address 0=1). As the code string A to be compared with this, the one shown in FIG. 2 is used.

In the case of global correspondence, when all code strings after address 1 or 2 of code string B exist in code string A, it is determined whether the address of code string A has reached the end, and if it has reached the end, the correspondence is established. Suppose we do this, and it will not be established unless it reaches the end. The role of special code 0 in code string B is the same as in the case of partial correspondence.

In the case of matching Figure 2 and Figure 7 (A), code string B
When all the codes (4, 1) exist in code string A, the address of code string A is 1 and has not reached the end. Therefore, the correspondence between code string A in FIG. 2 and code string B in FIG. 7(a) does not hold.

In the case of matching between Figure 2 and Figure 7 (b), all codes (4, 2, 4) of code string B exist in code string A, but the address of code string A at this time is 4. Therefore, it is assumed that the correspondence relationship does not hold.

In the case of matching between Figure 2 and Figure 7 (c), when the codes (3, 4, 1) of code string B are all present in code string A, the address of code string A becomes the final address 7. , a correspondence relationship is established.

Next, let us consider an example of the integrated application of Jt to pattern matching of this embodiment. Let codes 1 to 4 be the movement direction of the tracking point of the pattern contour in stroke feature extraction as shown in Figure 8 (a).
3" is previously written as code string B as shown in FIG. In this case, the transformed code string A of the number "3" shown in FIG. 8(C) is extracted as shown in FIG. By performing matching processing with B, a correspondence relationship can be established.

The processing flow of the control circuit 17 in FIG. 1 is summarized as shown in FIGS. 9 and 10. Note that Figure 10 is
The details of the determination step 116 in the figure are shown.

In FIG. 9, 101 and 102 are initial setting steps. Steps 103 to 105 read the code at address 2 when address 1 of code string B is code O. If code O in code string B does not appear, 1
The steps 06→107→108→109→110→106 are looped to match the corresponding codes of code string A and code string B. Step 111 is a step in which when the code O inserted in the code string B appears, the address of the code string B is incremented by one. Steps 112-11
4 is a step for re-initializing code string A and repeating the matching process again if address 1 of code string B is code 0 when a mismatch is detected in step 107.

201 in FIG. 10 is a step for determining whether it is a partial correspondence or a whole correspondence, and 202 is a step for determining whether the code string A has reached the final address.

Note that Figure 1 merely shows a functional block diagram, and in reality, for example, a personal computer or the like is used to
The processing shown in the figure is performed by software.

FIG. 11 shows a functional block diagram of a dictionary search device according to another embodiment of the present invention. In FIG. 11, dictionary memory 110
For example, a symbol string such as a feature vector of a pattern is stored in advance as a dictionary symbol string (hereinafter referred to as a dictionary code string) in 0. Unknown symbol strings such as feature vectors of input patterns (
An unknown code string (hereinafter referred to as an unknown code string) is input to an input register 1102, and a dictionary search result is output to an output register 1104.

The search control unit 1106 accesses the dictionary memory 100 and controls the entire dictionary search process, and includes a register group 1108 in connection with such control. This register group 1108 includes C0DE, AN, BN, JAD
R, CN, MODE.

It consists of FLG and RLT registers.

Reference numeral 1110 is a matching processing unit that performs matching processing between a dictionary code string and an unknown code string.

The unknown code string and dictionary code string are stored in the search control unit 1106.
A register 1112 and B register 1114 by
are set respectively. ADR1 counter 1116 is A
The ADR2 counter 1118 is an address counter that specifies the read address of the register 1112, and the ADR2 counter 1118 is an address counter that specifies the read address of the B register 1114.

The comparison circuit 1120 compares the codes output from the A and B registers to see if they match or do not match, and inputs the comparison result to the matching control unit 1122. This matching control section 1122
is for controlling the operation of the matching processing section 1110, and is for controlling the CNT counter 1 related to the control of the matching procedure.
124, and a specific register (flag) in the register group 1108 can be accessed. The output code of the B register 1114 is also input to the matching control section 1122. ADRI counter 1116 and ADR2
Counter 1118 is controlled by matching control section 1122.

FIG. 12 shows the overall flow of dictionary search processing.

Step 1203 is a processing step by the matching processing unit 1110, and the remaining steps are performed by the search control unit 11.
This is step 06. Details of the dictionary search processing operation will be described later.

Figure 13 illustrates the structure of a dictionary. As shown in Figure 6, the constituent units of a dictionary are a variable-length dictionary code string, a character code (1 byte), a mode identifier (1 byte), and a jump address (
(1 byte) and the number of code strings (1 byte).

The number of code strings indicates the number of bytes of the dictionary code string.

The mode identifier is information for specifying a matching method (partial correspondence, whole correspondence) and a method for controlling the subsequent flow of dictionary search when a match is found. The jump address specifies the starting address of the dictionary constituent unit to which the dictionary code strings to which the jump is specified by the mode identifier are to be proceeded to when matching (established) occurs. Further, a special code for controlling the procedure of the correspondence of symbols to be compared in matching is inserted into the dictionary code string as necessary.

Such a mode identifier, jump address, and special code are added to each dictionary code string as control information.

In addition, at the end of the dictionary, the code rFF is used as a terminator.
J is placed.

A specific example of a dictionary with such a structure is shown in FIG. 14, where ■ to ■ are dictionary constituent units, respectively. The character code O is a reject code. Code 0 in each dictionary code string
is a special code as part of the control information, which is inserted for the same purpose as the special code in the first embodiment.

Note that the codes other than the special code in the dictionary code string are specifically the stroke direction code of the character pattern as described in connection with FIG.

Here, the relationship between identifiers (generally expressed as rbl, b2J) and modes will be explained.

For dictionary code strings whose identifier is set to "00", such as the dictionary constituent unit in (■), "partial correspondence" matching is performed, and if the result is a match, the search is terminated and the dictionary is exited), or a mismatch. In this case, the process proceeds to matching processing with the dictionary code string of the dictionary constituent unit (■) located immediately after.

For dictionary code strings whose identifiers are set to "01", such as the dictionary constituent unit (2), "partial correspondence" matching is performed. If the matching result is a match, the process proceeds (jumps) to the matching process of the dictionary code string of the dictionary constituent unit (■) specified by the jump destination address, but if the result is a mismatch, the immediately following dictionary constituent unit Proceed to the matching process of the dictionary code string (■).

For dictionary code strings whose identifier is set to "10", such as the dictionary constituent unit in ■, matching is "generally compatible".
and exits the dictionary if the result is a match. If the matching results do not match, the following dictionary constituent unit (
Proceed to the matching process with the dictionary code string in (2).

For dictionary code strings whose identifiers are set to "11", such as the dictionary constituent unit in ■, matching is performed using the [whole correspondence], and if the result is a match, the dictionary constituent unit specified by the jump address is The process proceeds (jumps) to the matching process with the dictionary code string of (■), and if there is a mismatch, the process proceeds to the matching process of the dictionary code string of the immediately following dictionary constituent unit (■).

FIG. 16 is a flowchart of the operation of the matching processing section 1110, and FIG. 17 is a flowchart of the operation of the matching processing section 1110, and FIG.
6 is a detailed flowchart.

The matching processing operation will be explained with reference to these flowcharts.

When the matching control unit 1122 is activated by the search control unit 1106, the matching control unit 1122 sets the CNT counter 1124 as an initial setting.
is set to -1 (step 1301).

Step 1302 is an initial setting step for the actual matching process after step 1303.
The value of 124 is set, the ADR2 counter 1118 is set to O, and the FLG register (flag) is set to 1.

Next, the matching control unit 1122 checks whether the FLG register is in the reset state (set value is O) (step 130).
3) If it is in the reset state, jump to step 1306. If the FLG register is in the set state, it is checked whether the output code of the B register 1114 (the code of the dictionary code string read from the address specified by the ADR2 counter 1118) is O, that is, a special code (step 1304), and the special code is If not, step 1306
Jump to. If it is a special code, if you reset the FLG register to 0, the ADR2 counter 11 will also be reset.
18 +1 step 1305), step 1303
Return to

In step 1306, ADRI counter 1116
is incremented by 1, the code is read from the next address of the A register 1112, and the comparison circuit 1120 checks whether there is a match (step 1307).

If a match is found, the ADH2 counter 1118 is incremented by 1, the code is read from the next address of the B register 1114, and the FLG register is set (step 130).
8), and checks whether the value of the ADR2 counter 1118 is less than or equal to the value of the BN register (step 1309).

Note that the number of code strings in the dictionary code string is preset in the BN register by the search control unit 1106.

If ADR2 counter value≦BN register value.

It is checked whether the output code of the B register 1114 is a special code, that is, O (step 1310), and if it is not a special code, the process returns to step 1306 and the ADRI counter 11
16 is +1 and the code of the next address in the A register 1112 is read. In step 1310, B register 1
If the output code of 114 is determined to be 0, the ADR2 counter 1118 is incremented by 1, the code of the next address of the B register 1144 is read out, and the FLG register is reset (step 1311), and the process returns to step 1306.

If the determination condition in step 1309 is not met, the process advances to step 1316. The details of this step 1316 are explained in the 17th section.
This will be explained using figures. First, check whether the value of the b1 part of the MODE register is 0 (step 1401).
Identifier information is preset in the MODE register by the search control unit 1106. If b1 = 0, that is, if the dictionary code string specifies "partial correspondence" matching, the matching control unit 1122 determines the matching result as a match, and sets the RLT register as a flag of the result to 1. (step 1405).

In the case of b1 to 0 (the corresponding dictionary code string is specified for "general matching" matching), further check whether the value of the ADRI counter 1116 matches the value of the AN register (
Step l 402). In addition.

The number of codes of the unknown code string is preset in the AN register by the search control unit 1106.

If the comparison in step 1402 results in a match, that is, if the matching has progressed to the last code of the unknown code string, the matching result is determined to be a mismatch, and 1 is set in the RLT register (step 1404). If not, the matching result is determined to be a mismatch, and 0 is set in the RLT register (step 1403).

The explanation will be returned to FIG. 16. If it is determined in step 1307 that the codes do not match, it is checked whether the FLG register is set (step 1312). If it is in the reset state, return to step 1306 and register A register 1.
The code at the next address of 112 is read out, and if it is in the set state, it is checked whether the value of the b1 part of the MODE register is O (step 1313); if bl~0, that is, the dictionary code string is ”, the matching is determined to be a mismatch, and the RLT
Set the register to 0 (step 1315), bl=
If o, the CENT counter 1124 is incremented by 1 (step 1314), and the process returns to step 302 to perform matching again. In this case, the matching process will be performed from the code at the address next to the first address of the A register 1112 in the previous matching.

Note that steps 1315, 1403, 1404゜1405
If the process proceeds to , the matching control unit 1122 notifies the search control unit 1106 that the matching process has ended.

The matching process described above will be explained using a specific code string.

First, the matching operation of "one-part correspondence" will be explained.

Under the control of the search control unit 1106, the code string shown in FIG. 2 is set as an unknown code string in the A register 1112 with the address relationship shown in the figure, and the code string shown in FIG. Assume that the code string is set in the B register 1114 according to the address relationship shown in the figure. FIG. 19 shows the transition of addresses, codes, and flags in the matching process in this case.

M in the register group 1108 by the search control unit 1106
The ODE register is set to bl=o.

When the matching control unit 1122 is activated, the FL
Set the G register to 1 and read the code at address O of the B register 1114, but this code is special code 0.
Therefore, the code at address O of the A register 1112 is read (step 1 in FIG. 19). A, B register 1
The codes read from 112 and 1114 are both 4 and match, so the ADR2 counter 1118 is increased by +1.
Then, the code at address 1 of B register 1114 is read out (step 2).

This code is a special code O, so when the matching control unit 1122 resets the FLG register to O, it also adds 1 to the ADR2 counter 1118 and sets the B register 1.
The code at address 2 of 114 is read out (step 3).

This code is 2, which does not match, and the FLG register is also in the reset state, so set the ADRI counter 1116 to +LL,
The code at address 1 of A register 1112 is read (step 4). This is also a mismatch and FLG=O, so the ADR1 counter 1116 is +1 and the A register 1 is
The code at address 2 of 112 is read out (step 5).

This code is 2, which matches the code in the dictionary code string, so when the FLG register is set, the next code is read from the ADH2 counter 1118 and the ILB register 1114 (step 6).

Since this code is O, the FLG register is reset, the ADR2 counter 1118 is increased by 1, and the next code is read (step 7).

Since this code is not O, the ADRI counter 1116
is incremented by 1 and the code at address 3 of the A register 1112 is read out (step 8). This code is 1, which does not match the output code 4 of the B register 1114, and FLG=O
Therefore, the ADR1 counter 1116 is set to +LL and the code at address 4 of the A register 1112 is read (step 9).

Since this code matches with 4, set the FLG register. At this time, ADR2 counter 111
Since 8 has increased to the address corresponding to the final code of the dictionary code string, the matching control unit 1124 determines that a match has been achieved and sets 1 in the RLT register as a flag indicating the determination result of the matching result.

In this way, matching is executed while controlling the procedure of the correspondence of the codes to be compared by the control code O, but in this N, the matching method is "one-part correspondence", so
The code at address 1.3 of the unknown code string is ignored, and the code at address 0, 2.4 is ignored at address 0 of the dictionary code string.
, 2.4, it is determined that matching has been achieved (correspondence has been established).

In the case of matching (one-part correspondence) between the unknown code string in FIG. 2 and the dictionary code string in FIG. 18(b).

By the same operation as in the above example, the transition of addresses, codes, and flags becomes as shown in FIG.
Proceeding to 315, the result is a mismatch.

Next, matching ("partial correspondence") between the unknown code string in FIG. 2 and the dictionary code string in FIG. 18(c) will be explained. In this case, the transitions of addresses, codes, and flags are as shown in FIG.

First, the code at address O in the dictionary code string is read (step 1), but since this code is 0, that is, a special code, the process proceeds to step 1305 in FIG. 16, so when the FLG register is reset. Shimoni ADR
2 counter 1118 is incremented by 1, the code at address 1 of the dictionary code string is read out, and comparison with the unknown code string starts from here (step 2).

As the matching process progresses, the code comparison results in a mismatch in step 10, and since the FLG register is in the set state at that time, the process advances to step 1313 in FIG. In this case, bl=o (one-part correspondence), so step 13
The process returns to step 1302 via step 14. Therefore, the second matching starts from step 11 in FIG. Therefore, the code at address O in the unknown code string is excluded from matching, and matching is performed for the code at address 1 and onwards.

When the matching process progresses and a match is achieved up to the last code of the unknown code string in step 7, step 13 in FIG.
The process proceeds to step 1309 via step 08, and from this step N, the process proceeds to step 1316, where it is finally determined that there is a match.

Note that when matching is performed for the third time, the codes at addresses 0 and 1 of the unknown code string are ignored, and the codes from address 2 onwards are targeted.

Next, the "overall matching" matching process will be explained.

In "overall matching" matching, when all the dictionary code strings exist in the unknown code string, it is necessary to judge whether the matching has progressed to the final code of the unknown code string (step 1402 in Fig. 17), and if it has progressed to the end. It is determined that the matching matches, but if the matching has not progressed to the end, it is determined that there is a mismatch. The role of the special code O in the dictionary code string is the same as in the case of "one-part correspondence".

For example, when matching the unknown code string shown in Figure 2 with the dictionary code string shown in Figure 22 (A), the code at address 1 of the unknown code string matches all of the dictionary code strings, and the unknown code string Since matching does not proceed to the final code, steps 1402 to 140 in FIG.
The process proceeds to step 3 and is finally determined to be a mismatch.

The same applies to the case of matching the unknown code string in FIG. 2 and the dictionary code string in FIG. 22(b), and it is determined that they do not match.

In the case of matching the unknown code string in Fig. 2 with the dictionary code string in Fig. 22 (d), the final address 7 of the unknown code string
Since all of the codes (3, 4° 1) in the dictionary code string are matched, the matching is determined to be a match.

Next, the dictionary search operation when the contents of the dictionary are as shown in FIG. 14 and an unknown code string as shown in FIG. 15 is input via the input register 1102 will be explained with reference to FIG. do.

When the search control unit 1106 sets the input unknown code string in the A register 1112, it also sets the number of code strings in the AN register in the register group 1108 (step 1200). Note that the search control unit 1106 has a pointer for dictionary access, which is set to address 0 at this point.

Next, the search control unit 1106 checks whether 1 byte of the address specified by the pointer is the terminator rFFJ. Initially, one byte at address 0, that is, byte 1 of the character code of the dictionary constituent unit ■, is examined, and since it is not a terminator, the process advances to step 1202.

When proceeding to step 1202 for the first time, the search control unit 1106 reads out the dictionary constituent unit and uses its character code (
α) in the C0DE register, the identifier in the MODE register, the jump address in the JADR register, and the number of code strings in B
The dictionary code string is stored in the N register, and the dictionary code string is stored in the B register 1114.
Set to .

Then, the pointer is updated to the first address (8) of the next dictionary configuration rp. Note that since the dictionary code string has a variable length, its length is recognized by the number of code strings, and the pointer is updated. After such control, the search control unit 1
106 activates the matching control section 1122.

A under the control of the matching control unit 1122.

B. Matching processing as described with reference to FIGS. 16 and 17 is performed for the code strings in registers 1112 and 1114 (step 1203). When the final matching result determination result is obtained, a completion notification is sent to the search control unit 1106.

When the search control unit 1106 receives the completion notification, it checks the value of the RLT register, which is the flag of the matching result (
Step 1204). At first, ■ ^? , and the writing code string, and since they do not match, RLT=O. Therefore, the search control unit 1106 performs step 1206.
The process proceeds to step 1201, and refers to the 1 byte (dictionary constituent unit (first byte of ■) of the address indicated by the pointer), and performs the determination in step 1201.

In this case, since it is not a terminator, the dictionary constituent unit ■ is output one after another and the dictionary code string is set in the B register 1114,
The other bytes are set in the corresponding registers of the register group 1108, and the matching control unit 1122 is activated (step 1202).

Since a match is obtained by matching the dictionary constituent unit (2), the search control unit 1106 checks the value of the MODE register (step 1208). In this case, the value of the MODE register (:1 identifier) is 01, so the dictionary configuration unit (2)
The jump destination address is set in the pointer, and the determinations in step 1210) and step 1201 are made with reference to the first byte of 1 pie 1 to 2 of the configuration m in address 24.

This byte is not a terminator, so step 12
02, the dictionary code string of the dictionary constituent unit (4) is set in the B register 1114, the other bytes are set in the register group 1108, and then the matching control unit 1122 is activated.

Since this matching of dictionary code strings results in a mismatch, the process proceeds to step 1204. Step 1206, Step 1
Proceeding to step 202, matching of the dictionary code string of the first place (■) in the dictionary configuration tr and the unknown code string is executed. This matching results in a match and the RLT register is set to 1, so the process advances to step 1208. In this case, the value of the MODE register is 00, so the process advances to step 1212. In this step, the search control unit 106 outputs the character code (α) set in the C0DE register to the output register 1.
104 and exits from the dictionary (step 1212).

In other words, the search of the dictionary is completed.

Note that if the matching with the dictionary code string of ■ does not match, the terminator is referenced and as a result step 121
4, a reject code is set in the output register 1104, and the 1F search ends.

In this way, the flow of the dictionary search can be flexibly controlled according to the control information added to the dictionary code string, and the control is simple and the description for the control is also easy.

Note that the matching control section, search control section, etc. shown in FIG. 11 may be realized by hardware, or may be realized by rough hardware.

FIG. 23 shows a functional block diagram of a dictionary search device according to a third embodiment of the present invention. The overall configuration diagram of this dictionary search device is the same as the dictionary search device of the second embodiment shown in FIG. 11, and the overall flow of the dictionary search operation is as shown in FIG. It is exactly the same as the example.

However, the ADRO counter 112 for control is not installed in the matching control unit 1122a in the matching processing unit 11'l Oa.
5. ADR counter 1126, ADRB counter 112
7 is added, and the output code of the A register 1112 is input to the matching control unit 1122a, and there are some differences in the contents of the matching process. In addition, the dictionary memory 1100
Dictionary data structure and input register 11 stored in
There are some differences in the structure of the unknown code string input to 02.

Hereinafter, this embodiment will be described in detail, focusing on the differences from the second embodiment.

FIG. 24 illustrates the structure of the dictionary. As is clear from a comparison between this figure and FIG. 13 related to the second embodiment, in the dictionary according to the present embodiment, the next address of each code in the code string of each dictionary constituent unit has a length associated with the code. Information about the size or amount is added. Specifically, each code is a direction code indicating the direction of the stroke of the pattern as described in connection with FIG. 8, and information on the length of the corresponding stroke is added to each code.

Note that a special code (0) for changing the comparison correspondence is inserted into the code string of the dictionary as necessary, as in the second embodiment, but for this special code, information indicating 0 is inserted. is added. The dictionary structure other than this is the same as in the second embodiment.

FIG. 25 is an example of an unknown code string input to the input register 1102. As shown in FIG. This is a structure in which information about length or quantity (for example, stroke length) is added.

In this embodiment, in matching the dictionary code string and the unknown code string as described above, not only each code (for example, direction code) is compared, but also the length or quantity (for example, We also compare strokes (combined with strokes). In order to also compare such lengths or amounts, the matching control unit 1122a of the matching processing unit 1110a
As mentioned above, the three registers 1125, 112f;
.

1127 is newly equipped.

FIG. 26 is a flowchart of the operation of the matching processing section 1110a. Note that step 2112 is a processing step similar to step 1316 in FIG. 16, and its contents are as shown in FIG. 17.

In the matching process of this embodiment, as in the second embodiment, the codes of the unknown code string and the dictionary code string are compared, and if they match, step 2109 is performed to match the length or length associated with the code being compared at that time. It is checked whether the absolute value d of the difference in quantity is less than or equal to a predetermined value TH (for example, 2), and if the absolute value of the difference is less than or equal to TH, the code is determined to be a match.
A major feature is that when the absolute value of the difference exceeds TH, the comparison of the codes is determined to be a mismatch. That is, the main difference between the first and second embodiments is that matching is performed by comparing codes and comparing the lengths or quantities associated with the codes.

The matching processing operation will be described in detail below along the processing flow shown in FIG. When the matching control unit 1122a is activated by the inspection control unit 1106, the matching control unit 1122a sets -1 to the CNT counter 1124 as an initial setting, or
Set 12 swings in the RO counter 1125 (step 21
01).

Step 2102 is an initial setting step for actual matching processing after step 2103. In this step, the matching control unit 1122a sets the value of the ADRO counter 1125 in the ADRA counter 1126, sets the value of the ADRB counter 1127 to 0, and
Set the register (flag) to 1.

Next, the matching control unit 1122a checks whether the FLG register is in the reset state (0) (step 2103).
, if it is in the reset state, jump to step 2106.

If the FLG register is set (1), the matching control unit 1122a checks whether the output code of the B register 1114 (code at address O) is 0, that is, a special code, based on the output of the comparison circuit 1120 (step 210).
4) If the code is not a special code, jump to step 2106. If it is a special code, the FLG register is reset, the ADRB counter 1127 is incremented by 1 (step 2105), and the process returns to step 2103.

Step 2103 or 2104 to step 2106
When jumping to A, the matching control unit 1122a
D E< Add 2 to A counter 1126, set value of ADRA counter 1126 to ADR1 counter
D The value of the RB counter 1127 is transferred to the ADR2 counter 11.
18 respectively.

As a result, ADRI counter 1 is set from A register 1112.
The code of the address specified by 116 and the code of the address specified by ADR2 counter 1118 of B register 1114 are output, and the code of the address specified by ADR2 counter 1118 of B register 1114 is output.
Compare by 0.

The matching control unit 1122a checks the output of the comparison circuit 1120 (step 2107), and if the comparison shows a match, the process proceeds to step 2108, and if the comparison does not match, the process proceeds to step 2115.

If the codes match, in step 2108, the value obtained by adding 1 to the value of the ADRA counter 1126 is set to A.
Set in DRI counter 1116.

ADR is the value of ADRB counter 1127 plus 1.
By setting the 2 counter 1118, information on the length or amount of the matched code in the unknown code string and the dictionary code string is output from the A and B registers 1112 and 1114, and the difference in length or amount is outputted from the A and B registers 1112 and 1114. Calculate the absolute value d by Jf. Then, in step 2109, it is checked whether d is less than or equal to TII. If d is less than or equal to TH, the soft match of the code is determined to be valid and the process proceeds to step 2110; however, if d exceeds TH, the code comparison match is not evaluated to be valid and the process proceeds to step 2115. .

If the process proceeds to step 2110, the matching control unit 11
22a sets a value obtained by adding 1 to the value of the ADRB register 1127 in the ADH2 counter 1118, and also sets the FLG register. Next, it is checked whether the value of the ADR2 counter 1118 is less than or equal to the value of the BN register (step 2111). Note that the BN register contains the search control unit 11.
By 06/1? The number of code strings of the writing code strings is set in advance.

If the ADR2 counter value≦BN register value, the matching control unit 1122a checks whether the output code of the B register 1114 (the code at the address specified by the ADRZ counter 1118) is O, that is, a special code (
Step 2113).

If the code is not a special code, the process returns to step 2106, but if it is a special code, the value of the ADRB register 1127 is set to +2 in step 2114, the FLG register is reset, and the process returns to step 2106. If RL
The T register is set to Sera 1- or Sera 1- (the content of this determination is as shown in FIG. 17).

If it is determined in step 2107 that the code comparison does not match, or if it is determined that the comparison match is not valid in step 2109, the matching control unit L122a
It is checked whether the G register is in the reset state (0) (step 2115), and if it is in the reset state, the process returns to step 2106.

On the other hand, if the FLG register is not in the reset state, the MO
It is checked whether the value of the b1 portion of the DE register is 0 (step 2116). In the case of b1~O.

In other words, if the dictionary code string in question is specified for matching of "whole correspondence", the matching is determined to be a mismatch, and
The RLT register (result flag) is set to O (step 2118), and the matching ends. If b1=0,
In other words, in the case of j corresponding to one part, the CNT counter 11
After adding 1 to 24, the value of CNT counter 1124 is increased by 2.
The multiplied value is set in the ADRQ register 1125 (step 2117), and matching processing in step 2102 is started (1).

Note that when the process proceeds to step 2112 or step 2118, the matching control unit LL22a notifies the search control unit LIQ6 of the end of the matching process.

For example, the unknown code string shown in FIG. 27 is stored in the A register 1112 with the address relationship shown, and the dictionary code string shown in FIG. 28 (a) or (b) is stored in the B register 1114 with the address relationship shown. Assume that matching of r part correspondence J is performed. The odd addresses of each code string are length or quantity information.

In the case of matching with the dictionary code string in Figure 28 (a),
The transition of addresses, codes, and flags is as shown in FIG. 29, and the matching result is a match.

, and the determination threshold TH in step 2109 is set to 2.

The dictionary code string in FIG. 28(B) differs from the dictionary code string in FIG. 28(B) only in the length or amount of address 9, but this difference causes a mismatch in the matching results.

By adding length or quantity information to the code string in this way and introducing length or quantity comparison into matching,
Matching accuracy can be improved. This will be explained using a specific pattern example.

For example, the handwritten number "0" pattern shown in Figure 30 (a) and the handwritten number "6" pattern shown in Figure 30 (b) are given a direction code as explained in Figure 8. Then, as shown in Figure 1, exactly the same direction codes are assigned to (a) and (b) in the same order.Therefore, when matching a code string consisting only of direction codes, both patterns are assigned the same dictionary. Matching with the code string is established and identification is impossible.

However, the stroke lengths of the umbrella marks in each pattern differ considerably. Therefore, if the dictionary code string and unknown code string have a structure in which the stroke length is added to the direction code, and the matching process also compares the stroke lengths, the difference in the stroke length of the umbrella mark part (
The code string of the pattern (b) is matched with the dictionary code string of the number "0", but not matched with the dictionary code string of the number "6", and the opposite is true for the pattern (b). In this way, the accuracy of identifying similar patterns can be improved.

〔effect〕

As is clear from the above description, according to the present invention, by inserting special symbols into a symbol string, it is possible to extremely easily change the correspondence between symbols to be compared, and also to add '?' to a dictionary. By adding control information to the recorded symbol string when inserting special symbols, it is possible to easily control the matching method and dictionary search flow in dictionary searches, realizing pattern recognition that is resistant to pattern deformation. It has a great effect on doing so. Furthermore, by adding length information to dictionary symbol strings and unknown symbol strings, the accuracy of identifying similar patterns can also be improved.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram of a matching processing device according to an embodiment of the present invention, FIG. 2 is a diagram showing an example of a first symbol string (code string), and FIG. 3 is an example of a second symbol string. Figure 4 showing
6 to 6 are diagrams showing transitions of addresses, codes, and flags in the example of FIGS. 2 and 3, and FIG. 7 is a diagram showing another example of the second symbol string. FIG. 8 is a diagram showing an application example of the present invention, FIGS. 9 and 10 are diagrams showing an example of the processing flow of FIG. 1, and FIG. 11 is a function of the i1F search device according to another embodiment of the present invention Block diagram, 1st
2 is a flowchart of dictionary search operation, FIG. 13 is a diagram showing an example of the dictionary configuration, FIG. 14 is a diagram showing a specific example of the dictionary, FIG. 15 is a diagram showing an example of an unknown code string, and FIG. 16 is a diagram showing an example of the dictionary configuration. 17 is a detailed flowchart of step 1316 in FIG. 16, FIG. 18 is a diagram showing a dictionary code string for explaining the "partial correspondence" matching process, and FIGS. 21 is a diagram showing the transition of addresses, codes, and flags in the matching process between the unknown code string in FIG. 2 and each dictionary code string in FIG. FIG. 23 is a functional block diagram of a dictionary search device according to another embodiment of the present invention, FIG. 24 is a diagram showing an example of the dictionary configuration, and FIG. 25 is a diagram showing an unknown code string. A diagram showing an example, FIG. 26 is a flowchart of the matching process, FIG. 27 is a diagram showing an unknown code string to explain the matching process, and FIG. 28 is a dictionary code to explain the matching process of "1 partial correspondence" Diagram showing columns,
FIG. 29 is a diagram showing the transition of addresses, codes, and flags in matching "partial correspondence between the unknown coder section in FIG. 27 and the dictionary code string in FIG. FIG. 2 is a diagram showing a handwritten pattern for explaining the effect of introducing comparison of . 11...A register, 12...B register, 1
3...Flag, 14...A address counter. 15... B address counter, 16... Comparison circuit, 17... Control circuit, 1100... Dictionary memory, 1102... Input register, 1104... Output register, 1106... Search control unit , 1108...Register group. 1110.1110a...Matching processing unit, 111
2...A register, 1114...B register. 1116...ADR1 counter, 1118...ADR2 counter, 1120...Comparison circuit, 1122.1122a...Matching control unit, 112
4...CNT counter, 1125...ADRO register, 1126...ADRA register, 1127...ADRB register. Fig. 1 Fig. 3 Fig. 4 Fig. 5 Fig. 6 (A) Figure 25

Claims (7)

[Claims] (1) When checking whether a first symbol string matches a pre-registered second symbol string, insert a special symbol into the second symbol string, and use the special symbol to , a symbol string matching method characterized by controlling the order of correspondence of symbol strings to be compared. (2) A mode is set in the second symbol string to specify whether it is partial correspondence or whole correspondence, and when all the second symbol strings are present in the first symbol string, whether it is partial correspondence or total correspondence. 2. The method for matching symbol strings according to claim 1, further comprising the step of checking the following to finally determine whether they match or not. (3) The first and second symbol strings include length or quantity information associated with each symbol constituting the first and second symbol strings, and the symbols and their associated lengths or quantities are compared, respectively, in matching each symbol string. A method for matching symbol strings according to claim 1 or FIG. 2, characterized in that: (4) Each symbol in the first and second symbol strings is a direction code indicating the direction of the stroke of the pattern, and each symbol string has stroke length information added thereto, 3. The method for matching symbol strings according to claim 1, wherein symbols and lengths are compared in matching each symbol string. (5) When inserting special symbols into each dictionary symbol string registered in the dictionary, control information is also added, and the method of matching between the dictionary symbol string and the unknown symbol string is added to the dictionary symbol string. In addition to controlling according to the control information, the procedure for the correspondence of the symbols to be compared is controlled by a special code inserted into the dictionary symbol string, so that the result of matching the dictionary symbol string and the unknown symbol string is a match. A dictionary search method characterized in that the flow of subsequent dictionary searches in the case of a dictionary symbol string is controlled in accordance with control information added to the dictionary symbol string. (6) Dictionary symbol strings and unknown symbol strings contain length and quantity information related to each of the constituent symbol strings, and the symbols and their associated lengths and quantities are compared in matching each symbol string. A dictionary search method according to claim 5, characterized in that: (7) Each symbol in the dictionary symbol string and unknown symbol string is a code that indicates the stroke method of the pattern, and the stroke length information corresponding to each symbol is added to the dictionary symbol string and unknown symbol string. 6. The dictionary search method according to claim 5, wherein symbols and lengths are respectively compared in matching a dictionary symbol string and an unknown symbol string.