JPH039507B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (1) Description of the field to which the invention pertains The present invention relates to a deformed pattern generation method and a deformed pattern generation device that have a simple configuration and generate deformed patterns at high speed and in large quantities.

(2) Description of conventional technology A large amount of training data is usually used to design identification dictionaries for character reading devices, but as an alternative means of data collection, there is a method using artificially generated modified character patterns. be. Conventionally, methods for artificially generating deformed character patterns include methods that combine various deformation factors, such as linear and nonlinear deformation, global and local deformation, to standard glyphs, and methods that analyze human writing motion. , a method of configuring a character generation model using a feedback system that imitates this is known.

In the former method, linear transformation is realized by rotation, expansion and contraction, etc. using a two-dimensional matrix, and nonlinear transformation is realized by high-order polynomials and other special relationships.

In addition, in the latter method, the human writing movement is understood as a series of minute strokes 110 as shown in Figure 1, the stroke at a certain point is compared with a standard character shape, and the results are fed back to the next step. A method is used to determine the length and direction of the stroke at the point in time.

These conventional methods a) have a large number of parameters and are difficult to control deformation; b) require a large amount of processing to generate deformed characters; and c) are difficult to quantitatively express the degree of deformation. There were drawbacks such as .

(3) Purpose of the Invention In order to solve these drawbacks, the present invention employs a simple configuration in which irregular perturbations are applied to standard character shapes to generate deformed characters, thereby representing character deformations with a small number of parameters. This also makes it possible to quantify the amount of character deformation, and the drawings will be described in detail below.

(4) Description of structure and operation of the invention FIG. 2 shows an embodiment of the invention, in which 210 is a random number generation circuit, 220 is a smoothing circuit, 230 is a standard shape storage memory, 240 is a perturbation value calculation circuit, 25
0 is a deformation amount calculation circuit, 260 is a deformation amount output terminal,
270 is a thick circuit, 280 is a modified pattern memory, and 290 is a control section.

To operate this, a signal for initialization is sent from the control unit 290 to the smoothing circuit 220, the deformation amount calculation circuit 250, the thickening circuit 270, and the deformation pattern memory 280, and then the timing signal is sent through the connection 2901. According to the signal, the random number generation circuit 21
Generate one uniform random number with a value of 0 (-1, 1),
It is sent to the smoothing circuit 220. Also, this timing signal is transmitted to the perturbation calculation circuit 2 through the radio 2902.
40, and the perturbation calculation circuit receives this signal and reads out the coordinate value of one point on the standard character shape and the normal direction angle from the standard character shape storage memory 230.

The smoothing circuit 220 performs a smoothing operation based on the random numbers newly sent from the random number generation circuit 210 and the already stored random numbers, generates one perturbation value as a result, and performs the perturbation calculation. circuit 240,
and sent to the deformation amount calculation circuit 250.

The perturbation calculation circuit 240 calculates new coordinate values from the sent perturbation value and the coordinate values and normal angles that have been read out in advance from the standard shape storage memory 230, and sends them to the thickening circuit 270.

The deformation amount calculation circuit 250 uses the perturbation value sent from the smoothing circuit 220 to update the character deformation amount. On the other hand, the thickening circuit 270 performs thickening processing based on the coordinate values sent from the perturbation calculation circuit 240, and stores the results in the deformation pattern memory 28.
Write to 0. When the above processing is completed, random numbers are generated again using the next timing signal, and the same processing is repeated.

When there are no more coordinate values and normal angles to be read out in the standard glyph storage memory 230, the control unit 290 sends a termination signal to the random number generation circuit 210, the smoothing circuit 220, the perturbation calculation circuit 240, and the deformation amount calculation circuit. 250, and send it to the thick circuit 270. The deformation amount calculation circuit 250 receives this signal and outputs the deformation amount to the output terminal 260, and other circuits stop subsequent processing.

At this point, the deformed pattern memory 280 has the following information:
A modified character pattern corresponding to the amount of character modification output to the terminal 260 is stored. To continue generating modified character patterns, the controller 290 sends an initialization signal and repeats the same process as described above.

The random number generation circuit, smoothing circuit, perturbation calculation circuit, deformation amount calculation circuit, and thick circuit used in the present invention can all be realized with a simple configuration, and therefore,
It is possible to generate various modified characters at high speed and in large quantities.

FIG. 3 is a block diagram of the smoothing circuit 220.
The purpose of this part of the operation is to smooth the random number sequence sequentially sent from the random number generation circuit 210, convert it into a perturbation value sequence, and send it to the perturbation calculation circuit 240 and the deformation calculation circuit 250. be. The smoothing circuit 220 includes the deformation parameter storage memory 2
21, smoothing coefficient generation circuit 222, smoothing coefficient storage memory 223, random number sequence storage memory 224, multiplier 225, adder 226, perturbation value register 22
7, an address register 228, and an adder 229.

Two deformation parameters A and W are previously stored in the deformation parameter storage memory 221 from the outside. A is a real value satisfying A≧0, and W is a positive integer and an odd number. Prior to the deformation pattern generation process, the smoothing coefficient generation circuit 222 generates parameters A and W stored in the deformation parameter storage memory 221.
With reference to , W smoothing coefficients g ₁ , g ₂ , . . . gw are generated. The smoothing coefficient gk is stored in the smoothing coefficient storage memory 223 at address K (K=1, 2, ..., W).
are stored sequentially.

As the value of gk (K=1, 2, . . . , W), the following Gaussian function value is used.

gk=C・Aexp [−18(K−w+1/2) ² /(W −1) ² ] …(1) However, C= _W 〓 ^K=1 [−36(K−w+1/2) ² /(W -1) ² ] ...(2) (K=1, 2, ..., W) In addition, initial values for random number generation are set in advance in the random number generation circuit.

When the random number generation circuit 210 receives an initialization signal from the control unit 290 via the connection 2901, the random number generation circuit 210 generates W random numbers and sends them to the random number series storage memory 224 via the connection 210.
are sequentially stored from address 1 to address W. Thereafter, the memory contents at address i of the random number sequence storage memory 224 are set as fi (i=1, 2, ..., W).
It is expressed as

Next, the perturbation value register 227 and address register 228 are each reset to a value of 0 by a timing signal sent from the control section 290 via the connection 2903. Also, in response to the same timing signal, the random number generation circuit 210 sends one random number to the random number sequence storage memory 224 via the connection 2101.

After moving f _W-1 to address W, f _W-2 to address W-1, ..., f ₁ to address 2, the random number sequence storage memory 224 stores the one random number sent as follows. Store at address 1.

The adder 229 counts up the contents of the address register 228 by 1 at timing 2291, and sequentially sends the contents of the random number sequence storage memory 224 and the smoothing coefficient storage memory 223 indicated by the contents of the register to the multiplier 225. send. Multiplier 225 calculates the product of both, and sends the result to adder 2.
26 sequentially. The adder 226 sequentially adds this result to the perturbation value register 227.

When the contents of the address register 228 reach W+1, these additions and multiplications are stopped. At this point, the perturbation value register 227 stores a perturbation value γ expressed by the following equation.

γ= _W 〓 ^K=1 gk·fk (3) The perturbation value γ is sent to the perturbation calculation circuit 240 and the deformation calculation circuit 250 via the connection 2201.
Thereafter, every time a timing signal is received, generation of random numbers, shifting of the random number sequence storage memory contents, sum-of-products calculation using equation (3), and sending of perturbation values are repeated in the same manner as described above. The process is stopped by the end signal.

The series of perturbation values γ output from the smoothing circuit 220 via the connection 2201 is obtained by convolving and smoothing the random number series sent via the connection 2101 by the function expressed by equation (1). The degree of smoothing is the parameter W, and the series of perturbation values γ are A, A, and A in the deformation parameter storage memory.
It can be controlled by simply rewriting the value of W.

FIG. 9 shows a specific example of the smoothing process. Here, the smoothing effect when the parameter A is set constant and the parameter W is varied is shown. In the figure, a is a sequence of uniform random numbers that take values from -1 to 1, and b, c, d, e, and f are obtained by smoothing this sequence of uniform random numbers using the calculation of equation (3). It is a series. The parameters W of b, c, d, e, and f are respectively W=3, W=7, W=11, W=
15, corresponds to the case of W=19, and as is clear from the figure, it can be seen that the smoothing effect becomes larger as W becomes larger.

FIG. 4 is a configuration diagram of the perturbation calculation circuit 240. The purpose of the operation of this circuit is that the smoothing circuit 220
Based on the perturbation values sent from the standard character, perturbation is applied to the standard character shape to cause deformation. The perturbation calculation circuit 240 includes a perturbation calculation section 241, an address register 242, and an adder 243.

FIG. 5 is an explanatory diagram of an embodiment, taking the standard character shape "5" as an example, for explaining the operation of FIG. 4.

In FIG. 5, it is assumed that a standard character form 510 is composed of N character points. The standard character shape storage memory 230 stores in advance the x and y coordinate values and normal directions of N character points shown in FIG. The normal direction is expressed by the angle between the normal and the x-axis. The coordinate values and normal direction are in the order in which characters are actually written by humans, that is, on the side shown in Figure 5, they are ABC.
Then, they are stored in AD order from address 1 to address N. The x, y coordinate values and normal direction stored at address j (1≦j≦N) are respectively
Let Xj, Yj, and θj.

Upon receiving the initialization signal from the control section 290, the address register 242 is reset and the value 0 is entered into the register. Thereafter, every time the adder 243 receives a timing signal from the control section 290, the adder 243 counts up the address register 242 by one. Now, taking the case where the j-th timing signal is received as an example, after the adder 243 counts up the address register 242 by 1, the address of the standard glyph storage memory indicated by the contents of the register is referenced, The coordinate values Xj, Yj and the normal direction θj are sent to the perturbation calculation unit 241 via a connection 2301.

The perturbation calculation unit 241 uses the perturbation value γj sent from the smoothing circuit 220 via the connection 2201 and the above-mentioned Xj, Yj, θj by the same timing signal, and calculates the coordinates after perturbation according to the following equation. Values X′j, Y′j
is calculated, and the result is sent to the thick circuit 270 via the connection 2401.

X′j＝Xj＋γj cos θj ……(4) Y′j＝Yj＋γj sin θj ……(5) Equations (4) and (5) are expressed by the points on the standard glyph (Xj, Yj ) is perturbed by γj and moves (X′j, Y′j). When the sending of X'j and Y'j to the thick circuit 270 is completed, the same process as described above is repeated according to the timing signal of number j+1 from the control section 290, and X' _j+1 and Y' are sent to the thick circuit 270. The value of _j+1 is sent. and,
When the contents of the address register 242 exceed N,
A stop signal is sent from the control unit 290, and the above processing is stopped.

In this way, the perturbation calculation circuit 240 has a function of generating a modified character by perturbing each point on the standard character shape using the perturbation values sequentially sent from the smoothing circuit 220. The parameter A in equation (1) represents the magnitude of perturbation, that is, the average magnitude of deviation from the standard character shape, and the larger A is, the greater the deviation from the standard character shape. Further, the parameter W in equation (1) represents the degree of change around the standard character shape of the perturbed character line, and the larger W is, the smoother the change is.

FIG. 8 shows a specific example of generating modified characters. The character frame shown is 64 x 64 meshes (9.6 x 9.6mm), and the character height is approximately 30 meshes (4.5mm).
It is.

Here, regarding the perturbation parameters W and A, W
We selected three values of W = 3.45, 5.85, and 8.25 mm as the values of W, and furthermore, for each value of W, we set the value of A to A = 0.3, 0.6, and 0.9. As shown in the figure. Note that the value D in the figure is the amount of character deformation, which will be described later, and indicates the average value of the amount of deviation from the standard character shape.

As is clear from the figure, as the value of A increases,
It can be seen that the deviation from the standard character shape becomes larger and the value of D also increases. On the other hand, it can be seen that as the value of W increases, the changes around the standard character shape become gentler, resulting in smoother character lines.

FIG. 6 is a configuration diagram of the deformation amount calculation circuit 250. The purpose of the operation of this circuit is that the smoothing circuit 220
The purpose of this is to calculate the amount of character deformation based on the perturbation values that are sent sequentially.

An initialization signal sent from the control unit 290 via the connection 2904 causes the deformation amount register 252 to
is reset and set to the value 0. Adder 25
1 is the absolute value of the perturbation value sent from the smoothing circuit 220 via the connection 2201 and the deformation amount register 2
By adding the contents of 52 and writing the result, the contents of the deformation amount register 252 are rewritten.
The above operation is repeated while the perturbation value is sent from the smoothing circuit in response to the timing signal from the control unit 290, and when the end signal is received from the control unit 290, the operation is stopped and the deformation amount register is stored. 2
The contents of 52 are sent to the divider 253, where:
It is divided by 1/N and outputted to output terminal 260 as a character deformation amount.

The value output to the terminal 260 is 1/N _N 〓 ^j=1 |γj|, which represents the amount of character deformation as the average value of the magnitude of perturbation applied to each point on the standard character shape. It is something. With this configuration, the amount of character deformation can be calculated quickly with simple processing.

FIG. 7 is a configuration diagram of the thick circuit 270 and the modified pattern memory 280. The purpose of this circuit's operation is to perform thickening processing based on the coordinate values sent from the perturbation calculation circuit 240, and to write the results into the deformation pattern memory.

The thick circuit 270 includes a coordinate register 271, an adder 272, a thick area storage memory 273, an address register 274, and an adder 275.

In the modified pattern memory 280, each point of the pattern is represented by 1 bit, that is, a point on a character line is represented by 1, and other points are represented by 0.

The thick area storage memory 273 stores pairs (u, v) (Ul, Vl) (l=1, 2, . . . ) that satisfy the following equation.
L) is stored from address 1 to address L.

u ² +v ² ≦H (6) Here, L is the number of pairs of (u, v), and H is a parameter indicating the degree of thickness, which is specified from the outside.

An initialization signal sent from the control unit 290 via connections 2905 and 2906 resets the coordinate register 271 and the deformation pattern memory 280, respectively, and stores the value 0 therein. Next, when coordinate values are sent from the perturbation calculation circuit 240 via the connection 2401, they are stored in the coordinate register 271.

When the storage is completed, the timing 276 becomes active and starts sending timing signals to the adder 275 at regular intervals thereafter. When the adder 275 receives the timing signal 276, the adder 275 counts up the contents of the address register 274 by 1, and refers to the address of the thick area storage memory 273 indicated by the value.

The coordinate values sent from the perturbation calculation circuit 240 are
X′j, Y′j, referenced thick area storage memory 273
For example, if the contents of the address are Ul and Vl,
The adder 272 calculates X″j=X′j+Ue...(6) Y″j=Y′j+Ve...(7), and adds new coordinate values (X″j, Y″j) to the connection 2.
701 to the modified pattern memory 280, and sets a predetermined bit on the memory corresponding to the coordinate to 1. Thereafter, the same process is repeated every time the timing signal 276 is received. Adder 275
When the content of becomes L+1, timing 276 stops sending the signal.

Next, when a new coordinate value is sent from the perturbation calculation circuit 240, the coordinate value is stored in the coordinate register 271, and the timing 276 is activated to repeat the same process as described above.

Upon receiving the end signal from the control unit 290, the thick circuit 270 stops processing. At this point, the modified pattern memory 280 stores a modified character pattern having a character line width of 2H. Therefore, H
can be used to control the character line width.

(5) Explanation of Effects As explained above, the deformed pattern generation according to the present invention is basically composed of a random number generation circuit, a smoothing circuit, a perturbation calculation circuit, and a thick circuit, and the processing of each part refers to the memory. It is possible to perform only simple addition and product-sum operations, and does not require complex processing such as nonlinear transformation. Furthermore, control of the deformation is only possible with two parameters.

Therefore, there is an advantage that various modified character patterns can be generated in large quantities at high speed with a simple device configuration.

In addition, if the character deformation amount output from the deformation calculation circuit is used together with the deformed character pattern and applied to the recognition test of a character recognition device, the relationship between character deformation and recognition performance will become clear, and the character recognition device will stabilize against character deformation. You can check the gender.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram for explaining the conventional deformed pattern generation method, FIG. 2 is a block diagram of one embodiment of the method of the present invention, and FIG. 3 is a block diagram of one embodiment of the smoothing circuit of FIG. 2. 4 is a block diagram of one embodiment of the perturbation calculation circuit in FIG. 2, FIG. 5 is an explanatory diagram for explaining the operation of FIG. 4, and FIG. 6 is an implementation of the deformation amount calculation circuit in FIG. Example block diagram, FIG. 7 is an example block diagram of the thick circuit in FIG. 2, FIG. 8 is an explanatory diagram showing a specific example of deformed character generation, and FIG. 9 is an explanatory diagram showing a specific example of smoothing processing. It is. In the figure, 110 is a minute stroke, 210
2 is a random number generation circuit, 220 is a smoothing circuit, 230 is a standard shape storage memory, 240 is a perturbation calculation circuit, 2
50 is a modified calculation circuit, 270 is a thick circuit, 280
is a deformation pattern memory, 290 is a control unit, 510
indicates a standard glyph.

Claims (1)

[Claims] 1. A process of smoothing a uniform random number sequence by convolution with a function having a reduction filtering function to calculate a perturbation value sequence, and using the perturbation value to calculate each point on a standard character shape. A process of applying a perturbation in the same order as the writing order and calculating the coordinate value of each point of the character after perturbation, a process of calculating the amount of character deformation from a series of perturbation values, and a process of calculating the coordinate of each point of the character after perturbation. A deformed pattern generation method characterized by thickening character lines based on a value. 2 A random number generation circuit that sequentially generates uniform random numbers by giving an initial value, and a series of perturbation values that are smoothed by convolving the generated random number series with a function that has parameters for controlling deformation and output a series of perturbation values. A smoothing circuit that stores the coordinate values and normal direction angles of each point on the standard glyph in the order of writing, and a standard glyph storage memory that stores the perturbation values and the coordinate values and normal direction angles read from the standard glyph storage memory. A perturbation calculation circuit that calculates new coordinate values based on the perturbation calculation circuit, a deformation calculation circuit that calculates the amount of character deformation by averaging a series of perturbation values, and a perturbation calculation circuit that calculates new coordinate values based on the coordinate values output from the perturbation calculation circuit. The present invention further includes a thickening circuit that thickens a character line by a specified amount, a deformation pattern memory that stores the generated deformation pattern, and a control section that sends necessary control signals to each circuit. A deformed pattern generator characterized by: