JPH0668307A - Handwritten character recognizing method - Google Patents

Abstract

PURPOSE:To sharply shorten a time required for searching a character code even at the time of searching the character code by referring to the data base of a basic stroke in which plural nodes are arrayed at each stage of a tree structure by providing an inferring mechanism and an encoder. CONSTITUTION:A handwritten character is successively inputted to an inferring mechanism 1 by each one stroke. In the inferring mechanism 1, the set of basic stokes T0, T1... is preliminarily prepared so as to be standard, and which basic stroke corresponds to the stroke including the individuality of the inputted handwritten character is searched. The searched basic stoke string is inputted to an encoder 2. In the encoder 2, the plural basic strokes are arranged at each node while a reappearance is permitted, a number is added to each node, the branch of the tree structure is traced, and the character code is searched from the output data added to the node existing at the destination of the trace, so that the basic stroke string can be converted into the character code.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten character recognition method, and more particularly to a handwritten character recognition method capable of recognizing handwritten characters at high speed.

[0002]

2. Description of the Related Art Recognition of handwritten characters has been actively attempted in the past. As a method of recognizing handwritten characters,
In particular, the salient features of Kanji, that is, (1) the shape of one stroke (stroke) is generally determined. (2) For example, if you set a limit to "write in regular writing",
Stroke breaks can be clearly determined. There is a method that utilizes this feature. That is, a set of basic strokes that model each stroke (stroke) of characters such as Chinese characters is prepared, and when each stroke of a handwritten character is input, each stroke is pattern-matched with each basic stroke, and Let's be represented by a close basic stroke. As a result, a basic stroke sequence in which basic strokes are arranged in the order of strokes is obtained, which includes information in which stroke order each stroke of a handwritten character was input. Next, this basic stroke sequence is converted into a character code, that is, a handwritten character is recognized.

As a method for obtaining a basic stroke from a stroke of a handwritten character, for example, forward inference, backward inference, a blackboard method, a method using a neural network,
Various methods such as fuzzy inference can be adopted. For example, “Journal of the Institute of Electronics, Information and Communication Engineers D-II Vol.
74-D-II No. 2 pp. 166-174 1
February 991 'Online handwritten character recognition method applying fuzzy inference "Hirofumi Tamori et al." (Hereinafter simply referred to as "references") has a method of handwritten character recognition using fuzzy inference as shown in this title. Has been introduced.

When recognizing handwritten characters using the above-mentioned various methods, how to convert a basic stroke sequence into a character code becomes a problem. Hereinafter, an example of a conventional method for converting a basic stroke sequence into a character code will be described using a specific example.

[0005]

[Table 1]

[0006]

[Table 2]

Table 1 shows each basic stroke T0, T1, T
2, T3 and these basic strokes T0, T1, T
2, T3 is a correspondence table with a basic stroke code consisting of two bits that are identified with each other. Table 2 shows a correspondence between a basic stroke string in which basic strokes are arranged and a character code attached to the basic stroke string. It is a table. FIG. 11 shows
It is a figure showing an example of the basic stroke arranged in the tree structure.

First, two branches extend from the uppermost node (vertex) in the figure, and basic strokes T0 and T1 are arranged at the nodes ahead of each branch. Of these, the character code C1 is attached to the node where the basic stroke T0 is arranged, while the character code is not attached to the node where the basic stroke T1 is arranged. From each of these nodes, the node in which the basic stroke T0 is arranged is 3 more.
A book branch extends, and basic strokes T0, T1, and T3 are arranged at the nodes at the ends of the three branches, respectively. The character code C is assigned to each of these nodes.
3, C4 and C5 are attached. Two branches further extend from the node where the basic stroke T0 is arranged among these nodes, and the basic strokes T1 and T2 are respectively arranged at the nodes at the ends of these two branches. Character codes C8 and C9 are attached to the respective nodes. Further, two branches extend from the node on the uppermost stage where the basic stroke T1 is arranged, and the basic strokes T0 and T are respectively provided to the nodes at the ends of the two branches.
2 is arranged, and of these nodes, the character code C7 is attached to the node in which the basic stroke T2 is arranged. Furthermore, these basic strokes T0, T
One and two branches respectively extend from the node where 2 is arranged, and the basic stroke T0 is arranged at the node at the tip of the branch which extends from the node where the basic stroke T0 is arranged, and the character code C10 is attached. The basic strokes T1 and T3 are arranged at the nodes at the ends of the two branches extending from the node where the basic stroke T2 is arranged, and the character codes C11 and C1 are arranged at these nodes.
2 is attached.

Here, a case will be described in which the character code C4 corresponding to the basic stroke sequence T0 → T1 in which the basic strokes T0 and T1 are arranged in this order is obtained.
When the first basic stroke T0 (basic stroke code '00') is input, the basic strokes T0, which are arranged at the first two nodes of the tree-structured data,
Each address in which T1 is stored is sequentially given to the storage device, the basic stroke codes' 00 'and' 01 'are sequentially output, and the output basic stroke codes'00' and '
01 'is sequentially compared with the input basic stroke code' 00 '. Here, it is assumed that the basic stroke T0 (basic stroke code '00') is stored at the first address of the storage device. In that case, the basic stroke input by one address access and one comparison operation is input. The basic stroke T on the storage device that matches the stroke T0
The address in which 0 is stored is specified. This is shown in Figure 1.
It means that, out of the two branches in the uppermost stage shown in 1, the branch on the left side of the figure is followed to the node in which the basic stroke T0 is arranged.

Next, when the basic stroke T1 (basic stroke data "01") is input, the contents of the address specified above on the storage device are referred to, and the branch of the basic stroke T0 indicated by this contents is specified. The respective addresses of the basic strokes T0, T1, T3 in the above nodes are sequentially given to the storage device, the basic strokes T0, T1, T3 stored in the storage device are sequentially obtained, and these basic strokes obtained sequentially T0, T1, T3 are sequentially compared with the input basic stroke T1. If the basic strokes T0, T1 and T3 are stored in the storage device in this order, the stored basic stroke T1 is the second one, so the basic stroke T1 is input by two address accesses and two coincidence comparison operations. Coincides with the basic stroke T1,
The address where the basic stroke T1 is stored on the storage device is specified. This means that the process proceeds to the node in which the basic stroke T1 is arranged, which constitutes the second stage from the top of FIG. 11, whereby the character code C4 attached to this node is obtained, and this character code C4 is Is output.

Although the basic stroke sequence T0 → T1 has been described here, in this case, the character code C4 is obtained by a total of three address accesses and three coincidence comparisons.

[0012]

The above conventional example shows a considerably simplified example. In the case of a tree structure having an extremely large number of nodes in each stage shown in FIG.
There is a problem that a desired character code can be obtained only after an extremely large number of times of address access and matching comparison, and it takes a long time to obtain the character code.

Further, in the above-mentioned document, a tree-structured character dictionary is adopted in recognizing a character code after recognizing a basic stroke sequence, and fuzzy reasoning is also incorporated in recognizing this character code. Therefore, it takes a longer time to recognize the character code than when the simple tree structure as described above is adopted.

The present invention solves the above problem, and even if a character code is obtained by referring to a basic stroke database in which a large number of nodes are arranged in each stage of a tree structure, the character code It is an object of the present invention to provide a handwritten character recognition method that requires an extremely short time to obtain.

[0015]

FIG. 1 is a block diagram showing the configuration of the present invention. Handwritten characters are sequentially input to the inference mechanism 1 for each stroke (stroke). The stroke sequence of strokes is called a stroke sequence. Reasoning mechanism 1
, A standard stroke set is prepared in advance as standard, and it is required to determine which basic stroke in the basic stroke set each stroke including the input handwritten personality corresponds to. As this inference mechanism 1,
As described above, various inference mechanisms such as forward inference, backward inference, fuzzy inference, etc. can be adopted, and this inference mechanism 1 has a similar function even if it is not called'inference '. Various neural networks for pattern matching and pattern identification are also included. Further, the inference mechanism 1 may definitively associate each input stroke with each basic stroke and obtain a unique basic stroke sequence corresponding to the input stroke sequence. Using fuzzy reasoning, neural networks, etc., for example, when the'stroke 'of the input stroke is so bad that it is difficult to definitively associate that stroke with one of the basic strokes, For example, a plurality of corresponding basic stroke sequences may be associated with probabilities and the like.

The basic stroke sequence output from the inference mechanism 1 is input to the encoding device 2. This encoding device 2
Is a plurality of basic strokes arranged in each node of the tree structure allowing reappearance, a character code is attached to each node as necessary, and a node existing at the destination after tracing the branch of the tree structure By calculating the character code from the output data attached to, the basic stroke sequence in which the basic strokes are arranged in the order along the traced path is converted into a character code. And a node number input section which has a latch circuit for latching the node number attached to each node and which inputs the node number latched in this latch circuit, and these inputs based on the input basic stroke and node number. A coincidence detection circuit unit including a coincidence detection unit that detects a coincident node defined by the basic stroke and the node number. And the node number output section for outputting the node number attached to the node detected by the match detection section and inputting it to the latch circuit, and the output data attached to the node detected by the match detection section. And an encoding circuit unit having an output unit for outputting the output data.

Here, in principle, each node of the tree structure is given a different node number in order to identify each node from each other. For example, when a plurality of stroke orders for the same character are permitted, If the same basic stroke or basic stroke sequence with the same character code is present on the tip side of the multiple nodes after branching once after branching, from the multiple nodes once branched to the tip side It is preferable to combine the basic strokes or basic stroke sequences on the tip side into one so that the branches extend to the same node, and assign the same node number to the plurality of nodes that extend the branch to the same node. .

A character code corresponding to the basic stroke sequence input by using the encoding device 2 having the above configuration is output. In the case where the inference mechanism 1 is configured to obtain a plurality of basic stroke sequences corresponding to one stroke sequence, even if only the basic stroke sequence having the highest probability is converted into a character code according to a predetermined rule. Of course, after obtaining a plurality of character codes corresponding to each of a plurality of basic stroke sequences, one character code may be selected, or a plurality of characters may be displayed and the user may manually select the character code. Alternatively, one character code may be automatically selected from the character sequence identified so far, the context, etc. In the present invention, regarding the handling when a plurality of basic stroke sequences are obtained, It is not particularly limited.

[0019]

In the handwritten character recognition method of the present invention, since the character code corresponding to the basic stroke sequence is obtained after the basic stroke sequence is obtained, the encoding device 2 having the above-mentioned configuration is employed, and therefore, a large number of nodes from the node currently located are obtained. Even if the branch of the node is extended, unlike the case of the above-described conventional example in which it is sequentially searched, the node to be moved to the next is immediately found by one search operation, and therefore, the basic stroke sequence can be used in a very short time. Converted to code.

In the present invention, when the same basic stroke or basic stroke sequence to which the same character code is added is present at the tip side of the plurality of nodes at which the branch has been made once, after the branch has been made, , Even if the same basic stroke or basic stroke sequence with the same character code is added, they may be treated as different nodes or node sequences once the branches have been split, but the same node from these multiple nodes If a tree structure is configured to extend a branch to the same node and the same node number is given to the plurality of nodes that extend the branch to the same node, the tree structure is simplified as a whole, and therefore the circuit of the encoding device 2 is The scale can be reduced.

[0021]

EXAMPLES Examples of the present invention will be described below. FIG. 2 is a diagram showing an example of an input handwritten character “large” and a stroke sequence thereof, and FIG. 3 is a diagram showing a part of a set of basic strokes prepared in advance by the inference mechanism.

In order to avoid complexity, an embodiment will be described with reference to the block diagram of the present invention shown in FIG. 1, but the present invention is not limited to the embodiment described below. As shown in FIG. 2A, the handwritten character "large"
Shall be entered. At this time, the stroke sequence of S1 → S2 → S3 shown in FIG. 2B is input. Here, each stroke that constitutes this stroke sequence may be a stroke that includes information about the temporal movement direction of the pen as indicated by an arrow indicated by a broken line. In this case, prepare a set of basic strokes that includes information about the direction of movement of the pen, and enter a stroke sequence that also includes information about the direction of movement of the pen. For example, ignoring the information whether a horizontal bar is drawn from left to right or a horizontal bar from right to left, one stroke as a result is defined as one stroke as a result, for example, a horizontal bar is drawn as a result. To do.

When a stroke sequence as shown in FIG. 2 (B) is input to the inference mechanism 1, basic strokes corresponding to the strokes S1, S2 and S3 constituting this stroke sequence are obtained. The method for obtaining this is not particularly limited, and may be, for example, a method based on simple pattern matching, or a neocognitron suitable for image recognition, which is a type of neural network, or may be used. Fuzzy inference as described in the above-mentioned document may be adopted.

Here, the basic stroke sequence T3 → T7 → T9 corresponding to the input stroke sequence S1 → S2 → S3 is obtained by any of the above methods.
The obtained basic stroke sequence T3 → T7 → T9 is input to the encoding device 2. Hereinafter, an embodiment of the encoding device 2 will be described. However, for simplicity, in the following, the basic strokes T0, T1, ..., T11, ... Shown in FIG. T7
A further simplified example, that is, an example corresponding to the above-described conventional example, apart from T9 will be described. Therefore, here, Tables 1 and 2 in the above-mentioned conventional example are used as they are.

FIG. 4 is a diagram showing a data structure (tree structure) according to an embodiment of the present invention. This FIG. 4 corresponds to FIG. 11 in the above-mentioned conventional example, and is different from FIG. 11 in that a node number (number in parentheses in the figure) is attached to each node. Here, the method of encoding using the tree-structured data will be specifically described as follows.

First, consider the case where the basic stroke sequence T0 → T1 is input. At this time, the desired character code (10 bits) is' 0000 as defined in Table 1.
000100 '. In order to obtain this result, first, the basic stroke sequence '00' corresponding to the basic stroke T0 is obtained.
Is entered. The basic stroke T0 on the tree structure is arranged at the node at the branch destination of the node number (0) (node number (0) is the first node number for character code search). The basic stroke T0 connected to the tip of the branch extending from the node with the node number (0) is 1
Although there is only one, there are some other nodes in which the basic stroke T0 is arranged, which are connected to nodes other than the node assigned the node number (0).

Therefore, here, in order to recognize that the basic stroke input to this tree structure data is the first basic stroke T0 of the tree structure, this basic stroke code '00' and the node number (0)
Search for both (data "0000"). Although the character code C1 (character code '0000000001') is given to the node in which the basic stroke T0 of the branch point is arranged, this is not the character code to be obtained this time.

Next, when the basic stroke data "01" corresponding to the basic stroke T1 is input, the node number (1) (data ') of the immediately preceding basic stroke T0 is input.
0001 ') and the basic stroke code entered this time'
Both 01 'and the tree structure database are searched. As a result, the basic stroke T1 at the other branch tip
And the basic stroke T1 at the tip of the node with the node number (1) can be clearly distinguished.

With this, the branch of the basic stroke sequence T0 → T1 is determined, and the character code C4 (character code (10 bits) '00000000100') attached to the tip of the branch of this T1.
Will be output as the character code to be obtained. Similarly, any basic stroke sequence C1, ..., C12 defined in Table 1 can be obtained. FIG. 5 is a diagram showing an example of an encoding device in which the tree structure database is implemented as hardware.

The coincidence detection circuit section 20 of this encoding device has basic stroke code input terminals TD0 and TD1.
Node number input terminals ND0, ND1, ND2, ND3 are provided. This node number input terminal ND0, ND
The data input from 1, ND2 and ND3 are input to the node number setting circuit 22. The node number data output 26 from the encoding circuit section 25 is also connected to the node number setting circuit 22, and the output is switched by the output switching terminal SW1.

The numbers written at the leftmost end of the match detection circuit section 20 represent the node numbers (1), (2), (3), ..., (12) of each node of the tree structure shown in FIG. ing. For example, the node number (1) in the bottom row represents the node in which the basic stroke T0 is arranged in the upper row of the tree structure data in FIG. In addition, the match detection circuit unit 20
In, in the left-right direction, it extends to the coincidence detection circuit 21,
The line segment corresponding to each node intersects the data lines from the basic stroke code input terminals TD0 and TD1 and the data line from the node number setting circuit 22 which extend in the vertical direction. A black circle is displayed at this intersection when the data of the data line is the normal data "1", and a black circle is displayed when the data line is the inverted data "0". The output of 21 is configured to be "1" (active). That is, the node number (1) has the basic stroke code input terminal TD.
When all the outputs from 0, TD1 and the node number setting circuit 22 are "0", the output of the coincidence detection circuit 21 corresponding to the node number (1) becomes "1".

Here, the node number setting circuit 22 is set in FIG.
Node number (0) of the vertex node shown in (data '00
00 ') is set and the data' 00 'of the basic stroke T0 is input from the basic stroke code input terminals TD0 and TD1 in that state, the output of the coincidence detection circuit 21 corresponding to the node number (1) becomes'1'.'Becomes. Then, the lowermost row of the encoding circuit unit 25 on the right side of the figure becomes active, and the output of the output circuit 28 having the intersection where the white circle 27 exists becomes "1". Specifically, the node number data output 26 to '0001', the code valid bit output 29
To '1' and character code output 30 to '0000
000001 'is output. Here, the code valid bit output 29 indicates whether the data output from the character code output 30 is valid or invalid, that is, the node with the node number output from the node number data output 26 is assigned a character code. It indicates whether or not Since the code valid bit output 29 is "1" here, the data output from the character code data output 30 is valid, but here the data output from the character code output 30 is the character code to be obtained. do not use. The data “0001” output from the node number data output 26 is input to the node number setting circuit 22.

Next, the basic stroke code input terminal TD
When the code “01” of the basic stroke T1 is input to 0 and TD1, the output of the node number setting circuit 22 is the value “0001” from the search result of the previous basic stroke T0, and therefore the match detection circuit unit The input of 20 is now '
It becomes 010001 '. The match in this pattern is the node number (4) in which the basic stroke T1 is arranged (see FIG. 4). As a result, the output of the match detection circuit 21 corresponding to the node number (4) becomes "1", and the fourth row from the bottom of the encoding circuit section 25 becomes active. Therefore, the code valid bit output 29 becomes "1" and the character code output 30 becomes "00000000100", and finally this 10-bit code is obtained as a character code. At this time, node number data "0100" is obtained at the same time, but this is not used here.

As described above, by referring to both the basic stroke code and the node number, even if a large number of branches are branched from each node, high-speed search and match comparison are performed, and eventually High-speed character code search is realized. Here, it is desirable that the match detection circuit unit 20 be configured so that only the output of the match detection circuit 21 of one row is always “1” for each search data. if,
If this is not realized and the outputs of the plurality of coincidence detection circuits 21 are set to 1'at the same time, the method of determining the data of the encoding circuit unit 25 becomes complicated. Of course, the outputs of the plurality of coincidence detection circuits 21 are “1” at the same time.
It is also possible to configure the hardware to sequentially output a plurality of outputs that are "1" at the same time, but at the same time, only one output can be output. The hardware cost can be reduced by devising the method of creating the tree-structured data so that it becomes 1 '.

FIG. 6 is a circuit diagram showing a part of the encoding device shown in FIG. 5, and FIG. 7 is a circuit diagram further embodying the circuit shown in FIG. This FIG. 6 and FIG.
Input data of 111 '(basic stroke code' 11 '
And the output from the node number setting circuit 22 is' 011.
6 is a circuit diagram corresponding to the node number (12) in FIG. 5 in which the output of the coincidence detection circuit 21 becomes “1” for 1 ′).

As shown in FIG. 7, the match detection circuit section 20
Are six transistors T connected in series with each other.
Data lines DL1, DL2, DL3, DL4, DL are connected to the respective gates of r1, Tr2, Tr3, Tr4, Tr5, Tr6.
5, DL6 and data bar lines DBL1, DBL2, D
Any of BL3, DBL4, DBL5 and DBL6 is connected. Further, precharging transistors Tr10 and Tr11 are provided at two locations, and the transistor Tr10 is a P-channel transistor for precharging the potential at the point A. The transistor Tr11 is an N-channel transistor provided between the transistor Tr1 and the ground line, and suppresses discharge of the potential at the point A during precharge. The coincidence detection circuit 21 is composed of an inverter 20 'and a feedback P-channel transistor Tr12.

In the encoding circuit section 25, the output of the inverter 20 'is the memory transistors MTr1 and MTr.
2, the gates of MTr3 and MTr4 are connected.
These memory transistors MTr1, MTr2, MTr
One of the data output lines DOL3, DO
The other of L4, DOL5, and DOL6 is connected to the ground line.

Further, each data output line DOL1, ...
, DOL15 has P-channel transistors PTr1, PTr15 for precharging formed at one end, and an inverter 21 'and a feedback P-channel transistor Tr13 formed at the other end.
Here, first, the precharge control terminal 31 for initialization
When “0” is applied to the point A, the potential at the point A is set to “1”. Along with this, the output of the coincidence detection circuit 21, which is the inverter output of the point A potential, becomes "0". When the output of the coincidence detection circuit 21 becomes "0", the memory transistors MTr1, MTr2, MTr3, MTr4 are turned off, and when "0" is applied to the precharge control terminal 31, the precharge P The channel transistors PTr1, ..., PTr15 are turned on, the respective data output lines DOL1, ..., DOL15 maintain the state of "1", and the inverter output thereof is the inverter 21 '.
The output of keeps the state of '0'. At this time, the code valid bit output also outputs "0". This signal output makes it possible to know that the output of the character code (10 bits) is invalid data.

Next, the basic stroke code input terminal T
When desired input data is applied from D0 and TD1 and the output from the node number setting circuit 22 is determined, and then "1" is applied to the precharge control terminal 31, the search state is entered. By repeating the initialization state and the search state in synchronization with the application of each input data, a desired character code can be obtained.

Here, as an example of the matching search circuit according to the present invention, a NAND type dynamic decoder in which transistors are connected in series and which consumes relatively little power has been described, but this encoding device does not necessarily have this structure. The circuit configuration is not limited, and may be a general associative memory type circuit configuration. Next, an example in which the same node number is assigned to a plurality of nodes will be described.

FIG. 8 is a diagram showing a data structure (tree structure) according to another encoding device of the present invention. The difference from the data structure (tree structure) shown in FIG.
The only difference is in the structure of the branch extending from the node of node number (6) in which 0 is arranged and the branch tip extending from the node of node number (7) in which basic stroke T2 is arranged. The node number (13) has the same character code C11 as the character code C11 given to the node number (11).
11 is attached.

That is, in this tree structure, the basic stroke T
The same basic stroke T1 is arranged and the same character code C11 is attached to the tip of the node having the node number (2) in which 1 is arranged and branched to the node having the node number (6) and the node having the node number (7). The two nodes (the node having the node number (11) and the node having the node number (13)) are arranged. In such a case, a more compact structure can be achieved by changing the data structure (tree structure) as follows.

FIG. 9 is a diagram showing a changed data structure (tree structure). In the tree structure shown in FIG. 9, instead of the node of the node number (13) having the basic stroke T1 and having the character code C11 shown in FIG. 8, the node number having the basic stroke T0 shown in FIG. 8 is arranged. The node number of the node (6) is the same as the node number (7) in which the basic stroke T2 is arranged.
And the node having the node number (11) where the basic stroke T1 is arranged is connected by a branch. Thus, the same basic stroke T1
The number of nodes is reduced by grouping the nodes in which is arranged and the same character code C11 is attached into one node.

FIG. 10 is a diagram showing an example of an encoding device in which the tree structure database shown in FIG. 9 is implemented as hardware. The difference between the circuit shown in FIG. 10 and the circuit shown in FIG. 4 is that there is no node number 10 and node number 12 and that the node number (6) shown in FIG. 4 is rewritten to (7). Regarding the points, there is no change in the substance of the circuit structure), and a white circle 27a is added.

As described above, the tree structure shown in FIG. 8 is replaced with hardware as it is, and the same basic stroke T is used.
By configuring a tree structure (FIG. 9) in which 1s are arranged and nodes to which the same character code C11 is attached are combined into one (FIG. 9), equivalent functions can be realized with a small-scale circuit configuration. Further, in the above embodiments, the case where the character code data is directly stored in the encoding circuit unit 25 is described as an example. However, this is not always necessary, and the output data that can be converted into a character code is stored in the encoding circuit unit 25, the external memory is accessed using this output data as address data, and the character code is output from the output. It goes without saying that you can get the output.

[0046]

As described above, according to the handwritten character recognition method of the present invention, the basic stroke sequence is obtained by a predetermined inference mechanism, and then the character code corresponding to the obtained basic stroke sequence is obtained. Since the encoding device having the above configuration is adopted, the basic stroke sequence is converted into a character code in an extremely short time, and therefore, high-speed recognition of handwritten characters becomes possible.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of the present invention.

FIG. 2 is a diagram showing an example of an input handwritten character “large” and a stroke sequence thereof.

FIG. 3 is a diagram showing a part of an example of a set of basic strokes prepared in advance by an inference mechanism.

FIG. 4 is a diagram showing a data structure (tree structure) of an example of an encoding device according to the present invention.

5 is a diagram showing an example of an encoding device in which the tree-structured database shown in FIG. 4 is implemented as hardware.

FIG. 6 is a circuit diagram showing a part of the encoding device shown in FIG.

FIG. 7 is a circuit diagram in which the circuit shown in FIG. 4 is further embodied.

FIG. 8 is a diagram showing another data structure (tree structure) of the encoding device according to the present invention.

FIG. 9 is a diagram showing a changed data structure (tree structure).

10 is a diagram illustrating an example of an encoding device in which the tree-structure database illustrated in FIG. 9 is implemented as hardware.

FIG. 11 is a diagram showing an example of basic strokes arranged in a tree structure.

[Explanation of symbols]

 1 Inference Mechanism 2 Encoding Device 20 Match Detection Circuit Section 21 Match Detection Circuit 22 Node Number Setting Circuit 25 Coding Circuit Section 26 Node Number Data Output 29 Code Valid Bit Output 30 Character Code Output

Claims (1)

[Claims] 1. A predetermined inference mechanism for classifying each stroke of a handwritten character into one or more predetermined basic strokes, and a plurality of said basic strokes allowing each node to reappear in each tree structure node. The output data is arranged to each node as necessary, and the character code is obtained from the output data added to the node existing at the end of tracing the branch of the tree structure. It has a basic stroke input section for converting the basic stroke sequence in which the basic strokes are arranged in order into a character code, for inputting the basic stroke, and a latch circuit for latching the node number attached to each node. A node number input section for inputting the node number latched in the latch circuit, and a base stroke based on the input basic stroke and the node number. Te,
A match detection circuit unit including a match detection unit that detects a node that matches a node defined by the input basic stroke and node number, and outputs a node number attached to the node detected by the match detection unit And a node number output section for inputting to the latch circuit, and an output section for outputting the output data when the output data is attached to the node detected by the coincidence detection section. An encoding device comprising: a stroke sequence of handwritten characters is converted into one or a plurality of basic stroke sequences by using the predetermined inference mechanism, and the basic stroke sequence is converted into characters by using the encoding device. A handwritten character recognition method characterized by converting to a code.