JPS6232588A - Online consecutive-character recognizing device - Google Patents

Abstract

PURPOSE:To approximately halve the amount of calculation and the capacity of storage by recognizing an input pattern based on a character-string standard pattern-connecting standard patterns. CONSTITUTION:The titled device consists of a tablet 100, a standard pattern storage part 130 to store the standard pattern that is made by connecting an inter-character stroke from the end-point of a character to the start-point of the next character to an isolated character pattern showing one character, an end-point table 200 to store the end-point of the isolated character pattern, and a recognizer part 240 that recognizes an input pattern based on the character-string standard pattern made by connecting a standard pattern whose end-point is the one stored in the table 200 to a pattern made by connecting zero-piece or more standard patterns. The character- string standard pattern is synthesized by connecting one or more consecutive-character pattern(s) provided that the one connected last only is an isolated character pattern in which the inter-character stroke until the end-point stored in the end-point table is not included. Such a combination of consecutive-character patterns that minimizes the inter-pattern distance between the character-string standard pattern synthesized a aforementioned and the input pattern, is made the result of the recognition.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an online continuous character recognition device, and particularly to an online continuous character recognition device that recognizes one or more consecutive characters written from a basket of characters input online. It relates to a character recognition device.

[Prior art and its problems]

Conventionally, a device that recognizes a string of consecutively written characters without the user having to indicate the end of one character is disclosed in Japanese Patent Application No. 110743/1983 (
There was an online continuous character recognition device described in the document (hereinafter referred to as document (1)). Hereinafter, a conventional online continuous character recognition device will be explained using this device as an example.

In this device, the input character string is a time series pattern Amal, a2 . ...-a:,--,aI-
Conversion to (1). This time series pattern is taken as an input pattern person. The standard pattern B for each character is a time series pattern B"-b"l, b with the same direction angle as the input pattern.
"2.-, b"j, ・, b", a ・-・-(2)
It is held in the device as a. Here, n represents Bunkine. Moreover, this standard pattern B is an isolated character pattern BI"BI'wb'bc', which represents one character of the character name n.
, ,b”Jla −・−・−・(3)tt
'tz' and a continuous character pattern BC"BC"-BI'■b% , bmu, which is a combination of isolated character pattern and character strokes (bK, b, ,..., bQJ211) that connect the characters. ..., b', 21
...(4)-babll..., bl1.
It consists of bl,...baJ2%...(5)111 Proverbs. However, ■ represents a concatenation of patterns.

Here, the distance d(i, j) between the i-th data of input pattern A and the j-th data of standard pattern B is defined as the direction angle ai
, : defined as the angle formed between b. The pattern distance between the input pattern person and standard pattern B is the distance d (i, j) between the corresponding data by aligning the time axes of each time series pattern.
It is obtained by accumulating

The inter-pattern distance between the input character string and a standard character string pattern that is a concatenation of standard patterns for each character is determined. In other words, a character string standard pattern is synthesized by connecting zero or more continuous character patterns BC'' and one isolated character pattern BI'' at the end, and the inter-pattern distance between this and the input pattern person is calculated. The input string can be recognized by

As an example of a method for recognizing an input character string based on the standard character string pattern synthesized in this way, the above-mentioned document (1) describes the specification of "Japanese Patent Application No. 199098-1981" (hereinafter referred to as document (
2) DP of automaton control as shown in K
A method using a matching method is shown. In this method, a finite state automaton as shown in FIG. 3 is used to control the concatenation of the continuous character pattern BC'' and the isolated character pattern BI'' so that there is no contradiction. FIG. 3 is a diagram showing a finite state automaton. The finite state automaton has two states, an initial state $ and a final state f. When the character pattern BC" is accepted, the finite state automaton transits from the initial state $ to the same initial state 8. A state transition TC occurs, and when the isolated character pattern BI'' is accepted, a state transition TI occurs in which the initial state S transitions to the final state f, and the process ends.

In each of the above state transitions, automaton-controlled DP matching between the input pattern A and the synthesized standard pattern is performed as follows. By solving the recurrence formula for each time i-1゜2, . . . between j-1, 2, . . .
The minimum cumulative distance can be found.

gCn(tej) is state transition TC, character name n1 time i.

represents the recurrence formula value at j. However, as the initial value gC
, (oeo) −o * gca (o Ij) =■
(J-1*21...-Jfi).

At the same time as this recurrence formula operation, the necessary path value hc (minimum time for i, D
... Perform the assignment operation (7). Since the number is 9, the initial value is h's (0*O) −〇. route value h
'rs (', J) has (
6) When performing the recurrence expression calculation of the expression, the boundary value where the boundary distance is given, that is, where the starting end of the standard pattern corresponds to the input pattern is stored.

Also, in the same way as in the state transition TC, in the state transition TI, the minimum cumulative distance in the state f can be found by solving the recurrence formula between J-1, 2, ... J7.e gi
ll (i, j) is state transition TI, character name n side time
represents the recurrence formula value at j, where g is the initial value
i, (0,0)Woe gII(Olj)−ω(j−
1, 2, ... Jt).

Similarly to the path value hc, (i, j), at the same time as the recurrence formula calculation, for the path value hi (i, j), the following substitution operation is performed: (9). However, the initial value is hi (0,0)
=O.

In state transition TC, minimum cumulative distance gc up to time i
, (i, to) is the smallest for n T = mi
n gc, (i, to) ・・・・・・(
10) becomes the boundary distance of the recurrence formula calculation at the next time i+1. That is, when a state transition TC occurs in the finite state automaton in FIG. 3, gcI, (i, o)
=T... As in (11), if a state transition TI occurs, gi・(i・0
)−T...(12) and the boundary distance are given, and the recurrence formula calculations of equations (6) and (8) are performed. The boundary value of the path at that time is hc, (i, O) am i for each state transition.
= (13) hia(i, O) = i
It is given by (14).

In addition, the character name n that satisfies formula (10) and the path value h c7 (i + J ; ) at that time are respectively expressed as the boundary character name N
(i) and the boundary route L (i) are substituted as follows.

N(i)-M ・・・・・・(
15) L (i) -hc~(i, J:)
......(16) However, for time i-I, in state transition TI, min gi (i, J:) ・
...The character name W that satisfies (17) and the path value hi~(1°JT) at that time are similarly substituted as follows.

N(I)-n...(18
)L(I)-hi~(i, JT)...
...(19) Boundary character name N (i) and boundary path L(i
) stores the character name five of the standard pattern whose cumulative distance is less than the minimum at the end of the standard pattern and the position of the starting end of the standard pattern.

In this way, for the two transitions state transition TC and state transition TI, the recurrence formula calculation of the standard butter / and input pattern is performed at time im1.2. ...By performing the machining, the distance between the patterns becomes
: ) =-(20).

From the above, the recognition results are obtained as follows. In i-■, the value n of boundary character name N(i) = N (i)...
...(21) becomes the character name n of the standard pattern matching the part from time L(i) to I. Then the highest value is L
(i) and similarly find the value of N(i). L(
Recognition results are obtained in reverse order of time by tracing L(i) until it becomes i)-0.

By the way, in order to perform the above-mentioned recurrence formula calculations and path value substitution calculations on the device, each recurrence formula value gcI,j),
gi, (i, j) and each route value h c a (', J
), hia(i,j) need to be saved for one time because they will be referenced at the next time i+1. In other words, for character name n and time j, each recurrence formula value gca (i, j), gi, (i, j) and each path value hc at time i
, (i, j) hi, table storage GC (n, j) for storing (i, j) (n=1 in (j-1, 2, ·)
．． 2. ”・N)−(22)GI(n, D Cjmx,
2+-y:nwxx*z*-N)HC(n,j) (
j=1.2. -- to n-1, 2,...-N)...(Z
3) HI (n, j) (j=1.2.- to n-1,2
, =・N) is required.

In this way, in the conventional method, the GC, HC, and GI stores the recurrence formula values and path values for the state transitions TC and TI, respectively.
, HI four table storages are required, and recurrence formula operations and assignment operations must be performed for each.
In addition, the standard pattern also has the disadvantage of requiring a large amount of calculation and memory since it is necessary to store both the isolated character pattern and the continuous character pattern.

[Purpose of the invention]

An object of the present invention is to realize a high-speed, small-sized online continuous character recognition device by reducing the amount of calculation and memory required in the conventional online continuous character recognition device.

[Structure of the invention]

The online continuous character recognition device of the present invention has an input section that reads the traces of written characters as a time-series input pattern, and an isolated character pattern representing one character, and a character interval from the end point of a character to the start point of the next character. a standard pattern storage unit that stores standard patterns in which strokes are connected; an end point table that stores end points of the isolated character pattern; and end points that are stored in the end point table for patterns in which zero or more of the standard patterns are connected. A recognition unit is configured to recognize the input pattern based on a character string standard pattern which is a concatenation of standard patterns.

[Principle of approved invention]

The principle of the present invention will be explained based on the above-mentioned online character recognition device. In the present invention, the continuous character pattern BC'! is used as a standard pattern. Using only ``, the end point table Tj (n) that stores the end point Ja of the lone two-character pattern BI'' included therein is set as TJ(n) - JT.
・・・・・・Prepare as shown in (24). A character string standard pattern is synthesized by concatenating one or more of these continuous character patterns BC', but only the last concatenated pattern excludes intercharacter strokes up to the end point listed in the end point table Tj. The combination of the continuous character pattern BC'' with the minimum inter-pattern distance between the character string standard pattern synthesized in this way and the input pattern A is taken as the recognition result.

In the conventional online character recognition device, in the data of continuous character pattern BC11, j-1, 2,...Jl
The part is the same as the data of the isolated character pattern BI'' as shown in equations (4) and (5) above.

Further, the process is performed for these two character patterns (6).

The boundary distance of the recurrence formula operation of equation (8) is (11), (12)
equal as shown in the equation, so in the recurrence formula operation j
-1, 2, ... In the range of JT, gc, (+, j
)-gi.

(i, D holds. Therefore, the data in the table storage GI (n, j) shown in equation (22) that stores the recurrence formula value for the character name n99 time in the state transition TI of the automaton is the state transition TC j-1, 2, . . . Ja in table storage GC (n, j) stored similarly in
It is the same as the data in the part.

Therefore, in the present invention, the recurrence formula calculation is performed only on the continuous character pattern BC''.

Continuous character pattern B at the time and gate of input pattern A
For DP matching with C", the minimum cumulative distance up to time i can be obtained by calculating a recurrence formula similar to formula (6) for j-1, z, ... JH. However, if the initial value As g, (0,0)=0. g, (0,j)−ω(j
-1, 2,...J).

Simultaneously with this recurrence formula calculation, the following substitution calculation is performed for the path value h a (iI J ) necessary to later trace the combination of standard patterns that gives the minimum inter-pattern distance. However, the initial value is (0,0)-0. The path value ha '' l J ) contains where the boundary distance was given when performing the recurrence formula calculation of equation (25) at time 'tJ, that is, where the starting edge of the standard pattern corresponds to the input pattern. The boundary value is stored.

Character name n1 Minimum cumulative distance to time i g a (' v
J'2) is the smallest for n T -min ga(i, J:)
(27) becomes the boundary distance of the recurrence formula calculation at the next time ++X. That is, gll(i, O) = T ・
...(28), and the boundary value of the route at that time is h (i, O) −i ・
...It is given by (29).

Also, substitute the character name n that satisfies equation (27) and the path values h- and (+-JH) at that time to the boundary character name N (i) and boundary path L (i), respectively, as follows. .

N (i) = ? r...
...(30)L (i) -b-(i, J, :)
・・・・・・(31) Boundary character name N
(i) and the boundary path L (i), the character name five of the standard pattern whose cumulative distance is the minimum at the end of the standard pattern and the position of the starting end of the standard pattern are stored.

In other words, the standard pattern for character name N(i) is time L(i)
It can be seen that the parts from to i are matched.

By the way, in order to execute the recurrence formula operation and assignment operation in equations (25) and (26), the recurrence formula value g5(
i, j) and route value ha (il J) respectively in table storage G(n, j) (jwsa1+2e−”J:: ne
=1.2e'・N)'-(32)H(n,j) (j
=1.2.・J';:nm1.2+・N)・(33)

The above recurrence formula calculation is performed at time i=1.2. . . , I-1 to obtain the minimum inter-pattern distance for the continuous character pattern BC11. This process is inconvenient to performing recurrence formula calculations on the state transition TC in the conventional automaton-controlled Dr matching method.

On the other hand, for the last part of the input pattern A, the minimum inter-pattern distance with respect to the isolated character pattern BI' must be determined. Therefore, at the end point i=I of the input pattern A, the isolated character pattern BI11 is obtained from the end point table Tj(n).
End point Js' −TJ (n)...
(34) is obtained (25), and the recurrence formula operation and assignment operation of equation (26) are performed only for the isolated character pattern BI' portion of j-1, 2, . . . JW. This results in
The inter-pattern pressure @D equal to equation (20) is D = mln g, (1,)...
...It is found as (35).

From the above, the recognition results are obtained as follows. The aforementioned boundary character name N(1) stores the character name n of the standard pattern whose cumulative distance is the minimum at time i. Further, the boundary value is stored in the boundary route L(1). i-
In I, the value of the boundary character name N (i) is n - N (t)...
...(36) becomes the character name n of the standard pattern that matches the part from time L(i) to . Next, the value of i is set to L(
Similarly, the value of N (i) is determined by replacing the value of i). L
By tracing L(i) until (i)-0, the recognition result can be obtained in terms of time and time.

The principles of the present invention have been described above, and according to the present invention, automatic control can be performed with a small amount of calculation and memory.
Results similar to those obtained using the matching method can be obtained. When using the conventional automaton-controlled DP matching method, the recurrence formula operation and assignment operation are gc, (i, j), gi, (i
, j) and hc, (i, j), hill(i.

j), and the table storage for that is also done for GC (n, j), GI (n, j) and HC (n, j).
.

Four HI(n,j) were required. However, in the present invention, this can be achieved only by calculations on g, (i, D and hll(ilJ) and table storage G(n, j) and H(n, D. In other words, the amount of data for inter-character strokes is small. Therefore, if JrX and JF are approximately equal, both the amount of calculation and the amount of memory can be reduced by almost half.Similarly, since the standard pattern only needs a continuous character pattern, the amount of memory for the standard pattern can be halved. .

In the detailed explanation of the present invention described above, the method of expressing the input pattern and the standard pattern and the method of using direction angles to define the distance between the patterns have been described. A variety of methods can be used, such as those described in the literature entitled ``Online handwritten kanji recognition moving towards relaxing handwriting restrictions such as kuzushiji,'' December 5, 1958, pages 115-133. Furthermore, many forms of the recurrence formula (25) can be applied, such as the one shown in the above-mentioned document (2).

〔Example〕

Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the online continuous character recognition apparatus of the present invention, and FIG. 2 is a detailed block diagram of the Dr matching section in FIG. 1.

In FIG. 1, the continuous character pattern BC'' is stored in the standard pattern storage unit 130, and the isolated character pattern B in the continuous character pattern BCI' is stored in the end point table Tj200.
The end point Ji of the part corresponding to I' is.

Each character name n (n = 11 2+ . . . N) is stored.

(32) table storage G(n,j), H(n,j
), the recurrence formula value ga (' * J ) and the ayu value ha (' * j) are stored, and the time j and the character name n are stored.
G memory 170 and H memory 18 addressed by
Similarly, (27), (30), (31
) A T memory 210 that stores boundary distances, boundary character names, and boundary routes corresponding to the formula and is addressed by time t.
An N memory 230 and an L memory 220 are each prepared.

The character string data input from the taplet 100 is converted into time-series data of direction angles in the preprocessing section 110.

This time series data is sent to the input pattern buffer 120 and held as the input pattern ai. At the same time, j=m112. ..., Jl and n=1
．． 2. ..., a sufficiently large value (the maximum value allowed by the memory) is set for H. Similarly, T memory 2
10 has T(0)=Q,TO)−” (i=1+
2. ".*I) is set. Also, the end point of the input pattern is sent from the preprocessing section 110 to the end point detection section 260 and held there.

When the above initialization is completed, the time signal i is sent from the control unit 250 to 1, 2, . . . , is output, and processing is performed in synchronization with this. At this time i, the character designation signal nFil from the control unit 250, 2. ....

It changes like N. The input pattern ai is transferred from the input pattern buffer 120 to the distance calculation unit 14 according to the time signal i.
Sent to 0. At the same time, the standard butter yBll-b” b” ・= read out from the standard pattern storage unit 130 according to the character designation signal n and the standard pattern time signal j
b'' is sent to the distance M calculation section 140.The distance calculation section 140 calculates the distance d(i, j) between the input pattern ai and the standard pattern bj.

This distance d(i, D is sent to the DP matching unit 160, and the minimum cumulative distance is calculated for the character name n. Here, the operation within one cycle for the character name n is also calculated using FIG. explain.

When the character designation signal n is designated, the T memory 210 fl'
(i-1) is read out to the signal line T2 and the g register 16
01 and i-1 is written in the sh register 1606.
A numerical value will be written.

Next, when the standard pattern time signal j is set to 11', the contents of the sgl register 1601 become gz register 1.
The contents of the sh_t register 1606 are transferred to the hl register 1607. At the same time, G(n, 1) is read from the G memory 170 and written into the G register 1601, and H(n, 1) is read from the H memory 180.
is read out and written into the h register 1606. At this point, gz register 1602 and hl register 160
The values held in 7 are the recurrence formula (25), respectively.
Taking the boundary condition gm(11,0) of (26), (i-1,
0) Compatible with K. Also, gl register 1601
The values held in the and hl registers 1606 are g, (i-1°1) and ha(t-1,1), respectively.

Thereafter, the calculations of equations (25) and (26) are repeatedly executed while the standard pattern time signal j is sequentially increased.

Next, the operation when the % standard pattern time signal generally takes a value of j will be explained. At this point, gl register 1601. The gt register 1602 has a go
(i-1,3), ga(i-1,3-1) is also h
l register 1606. The hl register 1607 holds ha(i-1, 0, ha(i-1, J-1)).The minimum value detection unit 1604 stores the contents of the gs register 1601.gz register 1602. The minimum value of the signals gt and g- is determined and output as the minimum value signal g1.At the same time, 11#, respectively, corresponds to whether the minimum value is gt or gz.

A switch signal C having a value of 121 is output.

The distance calculation unit 140 calculates the distance d(t) according to the specification of the standard pattern time signal j from the control unit 250.

j) is output as a signal d. An adder 1609 calculates the sum of this signal d and the minimum value signal gm. The signal 1gz obtained in this way is the calculation result g of recurrence formula (25)
Il(i,j). This signal sg2 is a new G(n
, j) in the G memory 170. The multiplexer 1605 is controlled by the switch signal C and h
l register 1606. h, - Signals hl and h2 from register 1607 are selected. That is, the signal hl for Cm'l'.

(Signal h2 is selected for sw '2'. As a result, the output signal sh2 is the calculation result h of equation (26)
11(i,j). This signal 5sh2 is written into the H memory 180 as a new value of H(n,j).

When the above process is repeated up to j, , ga(i
, to), (i, resistance) is obtained. These signals are output as signals T, , h, respectively. In this way, one operation of the DP matching section 160 is completed.

Signal Tl from DP matching section 160 and T memory 21
Signal T2 from 0 is sent to comparison circuit 190 and compared. When T1 (T), a write pulse W is output from the comparator circuit 190. Due to the stiffness, the signal TI = gn ('eJ2) is output to the T memory 210, and the signal TI = gn ('eJ2) is output from the L memory 22.
0 has the signals hh, (r + J;), and the N memory 230 has the character designation signal n, respectively T (
1), L(1), and N(i). On the other hand, when T1>T2, these entry processes are not performed.

Character designation signal n is 1.2. ...N, the same process as above is repeated, and the process at time i ends. As a result, the minimum value of g, (i, J:) for n is held in the τ memory 210 as shown in equation (27). Furthermore, this process continues for 1, 2, and 1 at time 1. ...
, and are repeated, but the processing at the end point is different from before.

The end point detection unit 260 compares the time signal i with the end point work, and
-■ At the point in time, the end point signal e is sent to the end point table Tj200. When the end point table Tj 200 receives the end point signal e, it sends the end point Ji corresponding to the character designation signal n to the standard pattern storage section 130. Standard pattern storage section 1
30 uses this value sent to the distance calculation unit 140.
Let JS be the end point of the standard pattern to be output. In other words,
The data for standard pattern j is B10-b7. b
The following distance calculations and recurrence formula calculations are performed for j-1, z, ... JH.The subsequent processing is the same as before, and the single character designation signal n About 1.2. ...By changing it to N, the processing in the time adjustment is completed.If it is 0 or more, the distance between the patterns is T.
This means that it has been found as the value of (I).

Next, the 0 determination unit 240 starts the determination process to obtain the recognition result, and the 0 determination unit 240 reads the information from the L memory 220 and the N memory 230 to
i) and N (i) to output the recognition result n as in equation (36). The value of i-I is given to the N memory 230, the value of N (i) at that time is outputted as a result, and the value of i is then outputted to the L memory 220 to obtain the value of L(i). Newly set the value of L(i) to i and write N memo IJ 23
0, and similarly output the value of N(i) as a result. By repeating this process at t, which is L(i)-0, the identification result n is output in the reverse order of time i.

〔Effect of the invention〕

As described above, according to the present invention, by reducing the amount of calculation and memory in the conventional online continuous character recognition device by approximately half, it is possible to realize a high speed and small online continuous character recognition device.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the online continuous character recognition device of the present invention, FIG. 2 is a detailed block diagram of the DP matching section in FIG. 1, and FIG. 3 is a finite state diagram for explaining the prior art. It is a diagram showing an automaton. 100...Tablet, 110...Preprocessing unit, 120...Input pattern buffer, 1
30...Standard pattern storage section, 140...
...Distance calculation section, 160...DP matching section, 170...G memory, 180...H
Memo IJ, 190... Comparison section, 200...
...End point table, 210...T memory, 22
0...L memory, 230...N memory, 240...determination section, 250...control section, 260...end point detection section . ＼−一〆′

Claims (1)

[Claims] An input section that reads the handwriting of characters to be written as a time-series input pattern, and a standard pattern that stores a standard pattern in which inter-character strokes from the end point of a character to the start point of the next character are connected to an isolated character pattern representing one character. storage section and
Based on an end point table that stores the end points of the isolated character pattern, a string standard pattern that is a pattern in which zero or more of the standard patterns are connected, and a standard pattern whose end point is the end point stored in the end point table. An online continuous character recognition device comprising: a recognition unit that recognizes the input pattern.