JPS6261990B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a handwritten character recognition method for recognizing handwritten characters drawn on a tablet. In this type of conventional recognition method, an area for writing each character is defined on the tablet.
By determining the pen coordinates of the tablet,
I was cutting out characters. Therefore, it was necessary to set a sheet of paper with a writing frame printed on the tablet and write characters within the writing frame. An object of the present invention is to provide a handwritten character recognition method that can extract characters even when the characters are not written within the writing frame or even when the characters are written outside the writing frame. This was achieved by making the character strokes perform the same number of strokes, and then cutting out the next character stroke to be determined based on the determination results. In the present invention, as in the prior art, the pen coordinate locus from pen down to pen up in a handwritten character is treated as a stroke, and is sequentially divided into strokes and stored in a coordinate memory. In the present invention, without preceding character extraction, a specific number of temporally preceding strokes are selected as a unit, and the character to which the temporally most preceding stroke belongs within that unit is determined. The number of strokes selected as a unit is determined by the maximum number of strokes of the recognition target, and is determined by taking into account the type of recognition target such as numbers, alphabets, kana, etc., and modified stroke order and modified characters due to individual differences. In the case of number recognition, the unit can be two strokes, and the unit selected for character recognition is two strokes of numbers,
For example, combinations of numbers such as "4", "5", and one stroke, "11", "12", etc., and in the case of a combination of two numbers such as "12", the number to which the temporally preceding stroke belongs "1" is determined. Characters can be judged in the same way as before, but the feature items in the judgment dictionary and the feature extraction necessary for this are slightly different. For example, since a specific number of strokes is used as a unit, the number of strokes of a character itself is not a feature of character determination. Also, for example,
If the preceding stroke is similar to another digit ``1'', such as ``7'', a feature is required to distinguish it from a two-stroke combination that includes the digit ``1'', such as ``11''. . Selection of a specific number of strokes for recognizing the next character is performed based on the output of the stroke number table memory, and this corresponds to character segmentation. Stroke numbers corresponding to character codes are stored in advance in the stroke number table memory, taking into account modified characters and modified stroke order. The number of strokes corresponding to the character code recognized immediately before is read out, and based on this number of strokes, the determined stroke in the coordinate memory is determined,
A specific number of temporally preceding strokes are selected from the remaining undetermined strokes, and the next character is recognized. Note that if no new strokes are input to the coordinate memory for a certain period of time or more, character recognition is performed for the remaining strokes. FIG. 1 shows an embodiment of a handwritten character recognition device to which the character recognition method according to the present invention is applied. In FIG. 1, 10 is a character input tablet, 20 is a data register (R), and 30 is a coordinate memory. 40
is a write counter (WC) 41, a status display register (FLAG) 42, and a valid coordinate determination unit 43.
An input data write control section having the following. 5
0 includes a segment unit 51, a feature calculation unit 52,
This is a character recognition section consisting of a character determination section 53. The segment unit 51 includes a polygonal line approximator, a rotation amount calculator, a segment detector, and a segment registration memory. The feature calculation unit 52 includes a feature calculator, feature ROM, and feature calculation result memory. In addition
ROM shall mean read-only memory. The character determination unit 53 is a character determiner, a determination
Equipped with ROM. 60 is the stroke number table
In ROM, 70 is read counter (RC) 71
and the first stroke start address storage register (SR
1) 72 and a second stroke start address storage register (SR2) 73. The pen coordinate data input process by the handwritten character input tablet 10, data register 20, coordinate memory 30, and input data writing control unit 40 shown in FIG. 1 is combined with the flowchart of the pen coordinate data input process shown in FIG. I will explain. The input data writing control section 40 performs control to write the coordinates of the pen position pressed on the character input tablet into the coordinate memory, and as described later, interrupts and controls the character cutting recognition control section at any time. It is something. In the input data writing control section 40, in the processing step 2A, first, the pen-up and pen-down signals are drawn in from the character input tablet 10 at certain regular intervals, and in the processing step (2B), the pen-up and pen-down signals are determined. If the pen down signal is "1", it is determined that the pen is down, and if it is "0", it is determined that the pen is up. Further, the pen-up and pen-down states at the time of the previous input processing are set in the status display register (FLAG), and if the pen is in the previous pen-up state, FLAG=0, and if it is the previous pen-down state, FLAG=1. There are five types of processing contents for this pen coordinate data input processing:
The details of each process are described below. Processing state 1: After the device is started, there is no data input, or after the end point data of a character stroke is input, there is no data input. That is, pen up/down signal = “0”, FLAG =
In the case of "0", interrupt processing is started at regular intervals, pen up/down signal = 0,
Identify FLAG=0 and end the interrupt processing. Processing state 2 This is processing when the first character coordinate data of a stroke, which is the starting point of a character stroke, is detected. In other words, pen up/down signal = “1”, FLAG =
In this case, after identifying the pen up-down signal = “1” and FLAG = “0”,
"1" is set in FLAG, and the coordinate data (orthogonal coordinates x, y) of the detected character input is set in the data register 20. Processing state 3 This is a process when the pen-down state is continuous and a character stroke is drawn, and when the pen-up-down signal = “1” and FLAG = “1”, the judgment result in the effective coordinate judgment unit This process is performed when the currently input coordinate data differs from the previous coordinate data by a certain value or more. Whether the coordinates are valid or not is determined by inputting the previously inputted pen coordinate data stored in the data register (R) 20 and the currently inputted pen coordinate data into the input data writing control section 40. In this case, after identifying the pen up-down signal = "1" and FLAG = "1" and further identifying that it is valid coordinate data, write the previously input pen coordinate data stored in the data register 10 to the counter WC. The pen coordinate data pointed to is written to a predetermined address in the coordinate memory 30, and the pen coordinate data input this time is set in the data register (R) 20. Furthermore, the write counter (WC) is incremented by 1, and the +
In order to indicate the beginning of the unwritten area, "-1" is written at a predetermined address in the coordinate memory 30 that is addressed by the contents of the write counter (WC) that has been set to 1. Processing state 4 This is a case where the pen down state continues, but the difference between the pen coordinate position data and the previously input pen coordinate position data is within a certain value. In this case, the pen up down signal = "1", FLAG =
However, it is identified as invalid coordinate data, and the pen coordinate data input last time is written to the coordinate memory 30, and the pen coordinate data input this time is set to the data register (R) 20. do not have. Processing state 5 This is processing when the last character coordinate data of a stroke, which is the end point of a character stroke, is detected. That is, pen up/down signal = “0”, FLAG =
1, and the pen up/down signal =
After identifying "0" and FLAG="1", the pen coordinate data of the previous input set in the data register (R) 20 is determined to be the stroke end, and the pen coordinate data of the data register (R) 20 is added. Set the stroke end bit at the specified bit. Further, FLAG="0" is set, and then the pen coordinates of the stroke end set in the data register (R) 20 are written to a predetermined address in the coordinate memory 30 indicated by the write counter 41. Set the current pen coordinates of the pen up in the data register (R) 20, further increment the write counter (WC) by 1, and write "-1" to the predetermined address of the coordinate memory 30 indicated by the +1 write counter (WC). Set. FIG. 3 shows the coordinate memory 30,
The input pen coordinate data (x coordinate, y coordinate) is stored based on the pen coordinate data input process shown in FIG. This coordinate memory 30
has a stroke end bit for each input data, stores the stroke end bit set in processing state 5, and uses it to indicate the end point of each stroke of a character required during the cutout recognition process described later. There is. In the character cutout recognition process, the coordinate data of the character stroke stored in the coordinate memory 30 is input to the character cutout identification unit 50 under the control of the character cutout identification control unit 70 to proceed with the process and determine the character. It is something to do. This process will be explained with reference to the flowchart of the character cutout recognition process shown in FIG. In FIG. 4, when the apparatus is activated, that is, started, the character cutout identification control section performs initial settings. That is, a read counter (RC) 71 for setting the address of the internal coordinate memory;
The start address of the coordinate memory is stored in the first stroke start address storage register (SR1) 72 for indicating the start address of the first stroke stored in the coordinate memory, and the address of the coordinate memory 30 indicated by the read counter (RC) 71 is stored. (ie, starting address), "-1" is written to indicate the starting address of the unwritten area. Then, a stroke counter (SCNT) for calculating the number of strokes and a status display register (FLAG) for indicating pen-up/pen-down during the previous pen position input processing are each set to "0". Furthermore, the input data write control section 4
The start address of the coordinate memory 30 is set in the write counter (WC) 41 in 0. In this way, after setting each register, each counter, etc. of this device to OFF, the character cutting identification control unit 70 sequentially reads out the contents of the address of the designated coordinate memory of the reading counter (RC) 71, and sets it to "-1". Determine whether it exists. If it is "-1", it is determined that no more coordinate data has been input, and the same location is read out repeatedly, waiting until the next coordinate data is input. If it is not "-1", it is determined that coordinate data has been input, and then it is determined whether it is the end of the character stroke. This is done by referring to the stroke end bit in the coordinate memory. If the stroke end bit is not set, it means that one stroke has not been completed, and the read counter (RC) is incremented by 1 and the process waits until the next coordinate data is input. If the stroke end bit is set, it is determined that the stroke is at the end point, and the stroke counter (SCNT) is set to "+".
1" and determines whether the number of strokes is "2". When the number of strokes, that is, the value of the stroke counter (SCNT) is 1,
Add 1 to the value of the read counter (RC) and store it in the second stroke start address storage register (SR).
2). This is because the next read address indicates the first address in the coordinate memory where the second stroke is stored. In this case, the contents of the read counter are subsequently incremented by 1, and the process waits until the next coordinate data is input. When the number of strokes is "2", a start signal is sent from the character cutout recognition control section 70 to the character recognition section 50. As a result, the segment section 51 in the character recognition section 50 executes segment detection processing, the feature calculation section 52 executes feature calculation processing, and the character determination section 53 executes character determination processing, respectively. First, in the segment detection process, the first stroke start address stored in the first stroke start address storage register (SR1) in the character cutout recognition control unit 70 is set in a read counter, and the contents of the coordinate memory 30 are sequentially read. After reading, a polygonal line approximation process is executed in the polygonal line approximator, and as a result, the input character stroke is approximated on a polygonal line by a straight line having a directional quantum coefficient of 8. FIG. 5 shows directional quantum coefficients approximated by broken lines. Since this technique of approximating the input stroke using a straight line as a polygonal line is a widely used technique, it will be omitted here. Next, in order to segment the putter approximated by polygonal lines, first, a rotation amount calculator calculates the amount of rotation between adjacent straight lines within the same stroke. n in the pattern approximated by a polygonal line
The direction of the straight line Ln is expressed as Ln, and the direction of the straight line Ln is expressed by a direction code divided into 8 parts as shown in Figure 5 as (Ln).The direction change in two consecutive straight lines is expressed as a quantity with a sign, and this is expressed as (Ln). As the amount of rotation, the amount of rotation G (Ln, Lo ₊₁ ) is defined as follows. |(L _o+1 )−(Ln) | When <4 G(Ln, Lo ₊₁ )=(L _o+1 )−(Ln) When (L _o+1 )−(Ln)>4 G(Ln, Ln+1)=(L _o+1 )−(Ln)−8 When (L _o+1 )−(Ln)<−4, G(Ln, L _o+1 )=(L _o+1 ) −(Ln)+8 However, when |(L _o+1 )−(Ln)|=4, there are the following two cases. As shown in Figure 6, the straight line Ln
When the relative angle in the counterclockwise direction of the adjacent straight line L _o+1 to the line is 180° or less calculated from the coordinate position data of the starting and ending points of each straight line, G (Ln, L _o+1 ) = 4 L with respect to the above Ln The counterclockwise relative angle of _o+1 is
When the angle exceeds 180°, G(Ln, L _o+1 )=-4. According to the above definition of the rotation amount, the rotation amount calculator calculates the rotation amount of two consecutive straight lines. The segment detector 4 detects the amount of rotation G(Ln, L _o+1 ) and G(L _o+
1 , Lo ₊₂ ) are different in sign, the end point of the straight line Lo ₊₁ is set as the dividing point of the segment, and the rotation amount G, starting line direction, ending line direction, and segment starting point coordinate position for each segment are stored. Store the address and segment end point coordinate position storage address in the segment registration memory. Additionally, for the first segment of each stroke, a stroke start bit is set in the segment registration memory. FIG. 7 shows the segment registration memory. An example of segment division is shown in FIG. Using each data set in the segment registration memory, the feature calculation section 52 performs feature calculation processing. Here, in this embodiment, for example, recognition processing of numbers 0 to 9 is performed, and the case where the stroke order is written as shown in FIG. 9 will be considered. The feature calculator calculates the following nine types of features for each stroke and segment in correspondence with the stroke order features numbered 0 to 9 shown in FIG. The number of these features is required to be even greater in cases where the stroke order is different, in the case of kana characters, in the case of English characters, etc. First feature N(ST ₁ )=1 This means that the number of segments in the first stroke is 1.
Indicate that it is. Second characteristic R(ST ₁ , SG ₁ )>1 This indicates that the sum of the rotation amounts of adjacent straight lines within the first segment within the first stroke is 1 or more. Third feature R(ST ₁ , SG ₁ )>4 This is the second feature in which the total amount of rotation is replaced with 4 or more. Fourth feature R(ST ₁ , SG ₁ )<-1 This is the result of replacing the sum of the rotation amounts of the second feature with -1 or less. Fifth feature R (ST ₁ , SG ₂ ) <-1 This means that the total amount of rotation between adjacent straight lines in the second segment within the first stroke is -
Indicates that it is 1 or less. 6th feature R (ST ₂ , SG ₁ ) <-1 This means that the sum of the rotation amounts between adjacent straight lines of the first segment in the second stroke is -1
Indicates that: Seventh feature D(ST ₁ , SG ₁ , 1 ; ST ₁ , SG ₁ , 2) > {L(ST ₁ , SG ₁ )/4} This is the first stroke from the starting point of the first segment in the first stroke This indicates that the distance to the end point of the first segment in the first stroke is 1/4 or less of the total length of all straight lines in the first segment in the first stroke. Eighth feature D(ST ₁ , SG ₁ , 1 ; ST ₂ , SG ₁ , 1) <{L(ST ₁ , SG ₁ )/3} ∧R(ST ₂ , SG ₁ ) <−1∧N(ST ₂ ) = 1 This means that the distance from the start point of the first segment in the first stroke to the start point of the first segment in the second stroke is 1/3 of the total length of all straight lines in the first segment in the first stroke. The following is
And, the total amount of rotation between adjacent straight lines in the first segment in the second stroke is -1
and the number of segments in the second stroke is one. Ninth feature D(ST ₁ , SG ₁ , 1 ; ST ₂ , SG ₂ , 1) <{L(ST ₂ , SG ₁ )/3} ∧R(ST ₂ ,SG ₁ )=0 ∧F(ST ₂ , SG ₁ , 1 ; ST ₂ , SG ₁ , 2) = 0 This means that the distance between the start point of the first segment in the first stroke and the start point of the second segment in the second stroke is is 1/3 or less of the total length of all straight lines in
It also indicates that the direction from the start point of the first segment in the second stroke to the end point of the first segment in the second stroke is the 0 direction shown in FIG. For the features 1 to 9, if the pattern of each character satisfies the condition, the feature is determined to be present (1), otherwise it is determined to be (0), and the following judgment dictionary is created. However, × means that no attention is paid to whether it is 1 or 0.

[Table] Feature data 1 to 9 are stored in the feature ROM, and the feature calculator reads them one feature at a time, and reads the stroke group data required for feature calculation from coordinate memory 1 or segment registration memory. and determine whether the feature is present (1) or absent (0). All the results are stored in the feature calculation result memory for each feature. Next, character determination processing is performed in the character determination section, but the character determination ROM
A judgment dictionary is stored in , and the character judger sequentially reads out this judgment dictionary and the feature judgment results and outputs characters in which all (1) and (0) match as recognition results. As described above, the processing in the character recognition section 50 is completed, and the identification result is output as a character code. The character code that is the identification result is input into the stroke number table ROM 60. The stroke number corresponding to the character code is stored in this stroke number table ROM 60, and the stroke number corresponding to the input character code is outputted and inputted to the character cutting recognition control section 70. If the number of strokes of the determined character is "1", the character extraction recognition control unit 70 sets the second stroke start address stored in the second stroke start address storage register SR2 to the first stroke start address storage register SR.
1, and the read counter (RC), and set the stroke counter (SCNT) to "1".
The process then continues to cut out and recognize the next character. Also, if the number of strokes of the judged character is "2", the content of the read counter (RC) is set to "+1".
The value is stored in the first stroke start address storage register.
SR1, the content of the stroke counter is set to ``0'', and the value of the read counter itself is also updated by ``+1'', and the process continues to cut out and recognize the next character. Above, when identifying numbers, we have explained the case where each character requires a maximum of two strokes, but by increasing the number of strokes per unit, it is possible to identify numbers, kana characters, or alphabetic characters that differ from the stroke order shown in Figure 9. Of course it can be recognized. As explained above, in this embodiment, since it is not necessary to perform recognition using data for each character, there is no need for special extraction processing, and the number of strokes of the determined character is depicted and processed next in the coordinate memory. Since the address of the data to be processed can be determined and processed, character cutout recognition processing can be performed continuously. Therefore, there are advantages in that character pitch is allowed to vary, and characters can be input using a form without an entry frame (for example, one with only horizontal lines).

[Brief explanation of the drawing]

Fig. 1 shows an embodiment of a handwritten character recognition device to which the character recognition method according to the present invention is applied, Fig. 2 shows a flowchart of pen coordinate data input processing, and Fig. 3 shows a coordinate memory storing coordinate data, etc. Figure 4 is a flowchart of the character extraction recognition process, Figure 5 is a diagram showing directional quantum coefficients approximated by broken lines, and Figure 6 is for explaining the positive and negative signs of the rotation amount depending on the rotation direction between adjacent straight lines. , FIG. 7 is a diagram showing a segment registration memory storing various data, FIG. 8 is a diagram showing an example of segment division, and FIG. 9 is a diagram showing numbers when implementing the character recognition method of the present invention. It is a figure showing an example of stroke order of 0-9. DESCRIPTION OF SYMBOLS 10...Character input tablet, 20...Data register, 30...Coordinate memory, 40...Input data writing control section, 50...Character recognition section, 60
...Stroke number table ROM, 70...Character cutout recognition control unit.

Claims (1)

[Claims] 1. A pen coordinate trajectory from pen down to pen up in a handwritten character is stored as a stroke in a coordinate memory in order, and a certain number or more of undetermined characters belonging to one or more characters are stored in the coordinate memory. In a state where the strokes are stored, select the predetermined number of temporally preceding undetermined strokes as a unit, and determine the character to which the temporally most preceding stroke belongs based on the selected unit of strokes; 1. An online handwritten character recognition method characterized by selecting a stroke unit for recognition of the next character based on a predetermined number of strokes corresponding to the character code of the character.