JPS5924470B2 - Online character feature extraction method - Google Patents

Description

[Detailed Description of the Invention] The present invention relates to an online character feature extraction method, and in particular, extracts features based on the shape and stroke order of each character by analyzing the pen point movement of characters written on paper on a writing board. The present invention relates to an online character recognition method for identifying written characters.

To explain the prior art and the present invention, some terms will first be explained. (1) One stroke: One stroke is defined as a figure from the state in which the writing pen is "attached" to the surface of the paper until the state in which the shape is written by the pen and the state in which the writing pen is "separated" from the paper surface.

The number of character strokes is the number of strokes written in one character entry area.
It is the algebraic sum of strokes and corresponds to the "number of strokes". (2) Amount of change d: XY coordinate point series of the figure V) One coordinate of three points (Pi
, Pi+1, Pi+2), and vector Pi2Pi
The curvature was approximated by the product of the quantization coefficient (clockwise 0 to -8, counterclockwise 0 to 7) obtained from the direction difference between +l and Pi+1 and Pi+2 using a direction quantization circuit and the length 1Pi+1 and Pi+21. thing.

i20~(n-1). (3) Elementary feature code: When one stroke is divided into straight and curved partial strokes, the starting point P
Let D be the sum of the amount of change from s to the division point (to Ps+). (D=D8+Ds+,+...+Ds+r), and the sum of the distances Ti between each coordinate point is set to L (L=T8+DtS+1+
...+T8+r), l[l>. If THt is satisfied, it is a curve, and if the above formula is not satisfied, it is a straight line.In the case of a curve and the direction from P8 to Ps+r, a straight line 8 as shown in Figure 1 from the sign of D
With 16 types of seeds and curves as elementary features, elementary feature codes of O~7, 16~31 are given.

THt is a constant value. Figure 1 shows the elementary feature code. (4)
Discrimination feature: By combining the elementary feature codes in the time series appearance order, the appearance order in the stroke, the connection with the next elementary feature,
The final elementary feature code, etc., is specified as a distinguishing feature of a character.

In the conventional technology, extraction of elementary feature codes is performed as follows.

(1) If there is a coordinate Pj such that the amount of change 1djI≧THO, then P
One stroke was divided by i, and elementary feature codes were extracted between the starting point and point Pj.

This divides the stroke into partial strokes at points where the stroke suddenly changes. (2) If there is no sudden change point: Djl
) > THO is not detected, and in that case, the partial stroke length χL from the starting point to Pq is calculated using a fixed length THt, which is obtained by dividing the total length of one stroke into 3 to 4.
When t is satisfied, an elementary feature code is extracted between the starting point and point Pq, and using Pq+1 as the next starting point, the elementary feature code of Sugitsugi's partial stroke is extracted in the same manner.

Note that the elementary features are extracted by dividing one stroke into straight/curved partial strokes, so if one stroke is divided into 3 to 4 parts and the elementary features of the partial strokes are extracted, the same elementary features will be extracted. When codes are extracted adjacent to each other, one of them is representative to avoid duplication. Due to the above method, there will be fluctuations in the extraction of elementary feature codes for characters and figures that are drawn as either straight or curved lines, or characters that are relatively large.
Combinations and combinations of elementary features within the identification features were absorbed by logical 0R combinations, but this required a large capacity memory circuit to store the contents of the identification features, and at the same time it took a long time to calculate the identification features, making it difficult to write characters. The drawback was that speed was limited.

Therefore, an object of the present invention is to improve the above-mentioned drawbacks of the conventional technology, and by performing an elementary feature integration process on fluctuations in the detection of elementary features caused by deformation of written characters, the drawn character figures can be divided into straight lines and curved lines. This provides an economical feature extraction device that effectively divides and approximates the following. The present invention divides one stroke into partial strokes according to predetermined criteria as in the past, detects the elementary feature codes of these partial strokes in the order of appearance, and then extracts three strokes in the order of appearance. Extract the elementary feature codes, determine the combination form of their linear elementary features and curved elementary features, and combine 3 elementary features in a predetermined combination with 2 elementary features into 2 elementary features. The number of elementary features used for character identification is reduced by performing integrated conversion into feature codes and repeating the same integration using the integrated elementary feature code series. Note that this integration process consists of 1
When one stroke is divided into approximately 4 to 5 partial strokes, curves tend to be detected as a set of multiple straight line features, and curve features tend to be detected between two straight lines. Furthermore, considering that straight line features tend to be detected between two long curves, character identification processing is simplified by expressing it with a constant and small number of features. This is what we are trying to do. Table 1 is a table explaining the appearance order and combinations of elementary features to be subjected to integration processing. In Table 1, "straight" is a straight line feature,
A "song" represents a curve element feature. Order 1, 2, 3, 4, 5
The "-" in columns 4 and 5 means that whether the elementary feature at this position is direct or melody is irrelevant to the explanation of the integration process. The ○ mark in the processing column indicates that the process is performed, and the x mark indicates that the process is not performed. FIGS. 2 to 4 are diagrams illustrating the principle of integration. 2 to 4 (α) The dotted line represents the combination of original elementary features, and the solid line represents the combination of elementary features after processing. FIG. 5 is a block diagram of the first embodiment of the present invention, 400 is a coordinate memory, 401 is a change amount detection circuit, 402 is a change amount memory, 403 is a control circuit A, 404 is an elementary feature code judgment detection circuit, 405 is a switch, 406 is an elementary feature code and the start of each elementary feature code.・End point coordinate memory, 407 is an integrated processing mode determination circuit, 408 is an intermediate coordinate point detection circuit, 40
9 is a maximum value detection circuit; 410 is a control circuit Bl4ll is an elementary feature code detection circuit, S4OO is an input coordinate signal line 4
Enter 00. S4Ol, S4O2, S4O3 are 40
S4Ol inputs to 401 and 408 on the 0 output line. S
4O4 is input to 402 through the output line of 401. S4O5
, S4O7, S4O6, and S4O8 are each memory 40
It outputs from 403,410 with the control line of 0,402. S
4O9 is output from 403 and input to 404. S4
lO is the output of 402 and input to 404 and 409. S4
ll is output from 403 via the control signal line 404. S4
l2, S4l5 are the output lines of the elementary feature code, each with 40
4,411. S4l3, S4l4 is the beginning.
It is output from 404 and 411 on the end point coordinate signal line. S4l
6 is the control line of 405 and is output from 403 and 405 and 41
Enter 0. S4l7 and S4l8 are 4 at the output of 405
Enter into 06. S4l9 is the output of 406 and input to 410. S42O is the output of 406 and is input to 407, and is also input to the identification feature calculation unit. S42l, S422
, S423 are input to 410 as outputs of 40r, 408, and 409, respectively. S424, S425 are 41 control lines
It is output from 0 and input to 408 and 409. S426,
S427 is a control line that is output from 410 and 403 and input to 406, respectively. S428 and S429 are output from 410 and input to 411. S43O inputs to 407 a constant value given from the outside. S43l has 410 output lines and 40
Enter to 3. In FIG. 5, the parts within the dotted line belong to the conventional technology, and are schematically shown as a block diagram in order to explain the operation of this embodiment.

The procedure, operation, and effect of the integration process will be explained using Table 1 and FIGS. 2 to 5.

Integration processing is performed when one stroke is 3 or more, and in the case of combinations shown in Table 1, the combinations between orders 1 and 2 are integrated, and if the integrated result is 1, then the combinations between 1 and orders 3 and 4 are performed. Integrate by. (Mode 1) For combinations 1 and 2: Starting point T of straight line t1 as shown in Figure 2
, S and the end point T2E of the straight line.

That is, from the elementary feature code r of the straight line T, and the elementary feature code 6 of the straight line T2, one set of curve elementary feature codes (codes 16 to 23 corresponding to clockwise and counterclockwise rotation or code 2
4 to 31), and one curve element feature code similar in that direction is extracted from the coordinates of the start and end points Tls, t2E. (Mode 2) Case of combination 3: As shown in FIG. 3, there is a curve C narrowed by straight lines t1 and T2.

11! ,,i)-T2 (to) When l≧THd is satisfied, if the coordinates of the maximum change point of C are Pd, then the linear feature (Tls
, Pd), and the linear feature code (Pd, t2E).

Here, Tlb) and T2 (b) are each linear feature code number, and THd is a constant value, which is usually 3. (Mode 3) For combination 6: As shown in Figure 4, if the coordinates of the midpoint of the straight line t sandwiched by curves C, , C2 are Pm, then the curve element feature (ClS
, Pm) and the curved feature code (Pm, C2E)
Extract the elementary feature code.

That is, from the codes 22 and 30 of the curve element features Cl and C2 to the code set of the curve element features (ClS, Pm) and (Pm, ClS) (codes 16 to 23 corresponding to clockwise and counterclockwise rotation)
Or specify code 24-31) and start and end point Cls:
One curve element feature code each having similar directions of Pm, pm, and c2s is extracted. The operation and effect will be explained step by step using FIG.

In FIG. 5, the control unit A4O3 counts and temporarily stores the number of elementary features of each stroke, and when the number of elementary features is 3 or more, starts the control unit B41O with the control line S416 and switches the switch 405 in S414. , S4l5 are controlled to output to S4lr and S4l8. 410 receives the activation of the operation and performs S4.
26 controls 406 to partially read out the memory contents and determines the combination mode of the integration process in 407, and then in S43
When a constant value THd is input to O and mode 2 is detected, the absolute value of the difference between two straight line feature codes is detected, and when the condition is met, mode 2 is output.

From 40r, the mode number is input to 410 through S42l, and 410 performs processing control operations according to the processing mode.

For mode 1, 410 is connected to t1 through control line S426.
The start point coordinates (TlS) and the end point coordinates (T2E) of T2 are read out, T, s, t2E are output to 411 in S429, the curve elementary feature code is extracted by activating the control line S428, and the elementary feature code, Tls,t at S4l4
The 2E coordinate is input to the switch 405 and stored in 406 through S4lr and S4l8. For mode 2, control line S4
Through O6, 410 reads out the change amount Di of a partially necessary section from memory 402 and inputs it to 409.

409 is the coordinate point Pd that takes the maximum value from the input Di
is detected and input to 410 through S423.

410 is connected from memory 406 to AS through control line S426.
and read T2E (Tls, Pd), (Pd, t2E)
The start and end point coordinates of the combination are input to 411 through S429, and the straight line feature code is extracted by activating the control line S428. The subsequent steps are the same as in mode 1. Mode 3
In this case, 410 is connected to memory 400 through control line S4O5.
Read out the coordinate points of the more partially required section and input them to 408? let

408 detects the midpoint coordinates from the input number of coordinate points and performs S4
22 to 410.

410 is the Cl from the memory 406 through the control line S426.
Read S, C2E (C, S, Pm), (Pm, C2E
) are input into 411 through S429, and the curve element feature code is extracted by activating the control line S428.

The rest is the same as in mode 1. In this way, memory 4
06 stores the new elementary feature code and start/end point coordinates that have been subjected to the integration process.

410 inputs to 403 through S43l that one integration process has been completed.

If a mode other than the specified mode is detected by 40r, 410 is considered to be the end of the process, and the end is sent to 4 through S43l.
Enter into 03. Since 403 temporarily stores the number of elementary features in one stroke, 410 is activated through S4l6 until all integration processing is completed.

When all the integration processes are completed, the elementary feature code is read out from the memory 406 in S42O and calculation of identification features begins. As explained above, by performing the integration process as in the first embodiment, fluctuations in the extraction of elementary features due to size deformation of characters and shapes that are difficult to distinguish between straight lines and curves can be absorbed. The amount of variation absorbed by the logical OR combination is reduced, making it easier to create identification features, reducing the storage capacity of identification features, and making the device more economical.

The computing time of the distinguishing features is also reduced, which has the advantage of making online character recognition devices using such devices more effective. In the case of mode 2, the curve is divided by the coordinates of the maximum value point of the amount of change Di of the curve and integrated into a straight line, but if the curve is smooth and there is no obvious difference in the amount of change, the midpoint of the curve is can also be integrated as Pd.

In addition, in the case of mode 3, the 7th
As shown in the figure, even if the curves of Cl and C2 are integrated as they are, they approximate the original figure, so it is expected that a similar effect will be obtained.

The present invention extracts partial shapes of straight lines and curved lines from the shape of characters, and combines straight lines and curved lines depending on the order of appearance to form character identifying features. It can be used for online recognition devices such as part symbol characters.

Brief explanation of the drawings: Figure 1 is an explanatory diagram of elementary feature codes; Figures 2, 3, and 4 are diagrams diagrammatically explaining the integration process of the present invention; and Figure 5 is an illustration of the implementation of the present invention. Example block diagrams, FIGS. 6 and 7, are diagrams illustrating other embodiments of the integration process.

400, 402, 406... memory circuit, 401...
Change amount detection circuit, 403, 410... control circuit, 40
4... Elementary feature code judgment detection circuit, 405... Switch, 40r... Integrated processing mode judgment circuit, 408...
- Intermediate coordinate point detection circuit, 409... maximum value detection circuit,
411...Elementary feature code detection circuit.

Claims (1)

[Claims] 1 a) The process of detecting the pen point movement of writing characters on the writing board as a coordinate point series of XY coordinates for each stroke and storing it in memory, and b) The process of sequentially shifting the coordinate point series of one stroke one point at a time. A process of extracting three points from the memory one by one and detecting the product of the quantization coefficient of the direction difference between the two vectors given by each three points and the length of one vector as the amount of change d, and c) change in one stroke. When the amount of change is inspected and there is a change amount larger than a predetermined value THo, divide it at the coordinate point corresponding to the change amount, and if there is no change amount larger than the predetermined value, d) dividing each stroke into partial strokes by dividing each stroke by a length determined according to the total length of one stroke; and d) making each partial stroke correspond to predetermined elementary features of straight lines and curves. In this online character feature extraction method, the process of extracting three elementary feature codes for each stroke and determining whether each elementary feature code is a straight line or a curved line is performed. e) If the sequence is a line-straight line-straight line or a straight line-straight line-curve elementary feature code, the first two straight line elementary feature codes are integrated and converted into one curve elementary feature code. , f) If it is a straight line-curve-straight line elementary feature code series, integrate the three elementary feature codes and convert it into two straight line elementary feature codes, g) In the case of a curve-straight line-curve elementary feature code series An online character feature extraction method characterized by including, in certain cases, a process of integrating three elementary feature codes and converting them into two elementary feature codes.