JPS5856905B2 - Pattern extraction method and device in pattern recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern cutting method and apparatus for cutting out unit patterns that are separated and arranged in a predetermined direction in a pattern recognition device such as a character reading device.

In a character reading device, a document is scanned line by line, and one character pattern is detected from the resulting image pattern.
Characters are read by cutting out each character pattern and comparing the cut out character pattern with a standard pattern.

For example, a one-line pattern area containing multiple character patterns as shown in Figure 1 is first scanned in the vertical direction (j direction) using a scanner (corresponding to the eye that performs photoelectric conversion), and photoelectric conversion is performed. Convert it, then move it by one scanning line (one column) in the horizontal direction (i direction), scan again in the vertical direction, perform photoelectric conversion, and perform these operations to scan the entire pattern area by one row. Continue until.

Each character pattern is separated and cut out from the video pattern obtained in this way. Conventionally, in such a cutting method, first, the vertical peripheral distribution is calculated, and then the horizontal direction of the character pattern is calculated. Find the center of the character frame, refer to the pitch of the character frame given in advance or the character pitch, and find the horizontal separation between the character patterns in the pattern area for one line, as shown by the dotted line in Figure 1. , Next, refer to the character patterns in the one-line pattern area stored in memory, find the horizontal peripheral distribution of each character pattern for separation purpose, and find the vertical center point of the character pattern. .

For example, in order to cut out the character ``田'' as shown in Figure 2, when calculating its horizontal and vertical peripheral distributions PH and Pv, the left end coordinate i1, right end coordinate 12, and top end coordinate of the character pattern are calculated. The coordinate j and the lower end coordinate j2 are determined, and thereby the horizontal and vertical center points of the character pattern are determined, and based on the center points, the character pattern can be cut out into a frame of a desired size.

In such conventional pattern extraction methods, the peripheral distribution in the vertical direction is determined simultaneously with scanning by detecting the state of the binary pattern within the vertical scanning line, but it is not easy to obtain the peripheral distribution in the horizontal direction. .

That is, after first determining the horizontal delimiter using the vertical peripheral distribution, the corresponding horizontal bit is retrieved from the memory and the state of the horizontal binary pattern is detected to determine the horizontal peripheral distribution. I'm looking for a distribution.

Therefore, there is a drawback that it takes a very long time to obtain the peripheral distribution in the horizontal direction, and a great deal of time is required to cut out the pattern.

In addition, since the pattern that has been stored in the memory in a predetermined order is read out again in a different order to determine the horizontal peripheral distribution, it is necessary to provide a special memory for this purpose, which makes the equipment very expensive. There was a problem with becoming.

The main object of the present invention is to provide a pattern cutting method that can cut out patterns at extremely high speed.

Another object of the present invention is to provide a pattern cutting device that has a very simple configuration and can cut out patterns at high speed.

In order to achieve such an object, in the present invention, when sequentially scanning a scanning area including a plurality of unit patterns arranged separately in a predetermined direction along a scanning line in a direction perpendicular to the predetermined direction, In addition to detecting the presence of unit patterns in each scanning line, the distance from the start and end of each scanning line to each unit pattern is verified, and the unit pattern is extracted based on these detection results. There are certain characteristics.

For example, in the above-mentioned character reading device, when scanning a pattern area for one line, the peripheral distribution of scanning lines in the vertical direction, which is the scanning direction of the scanning area, is determined, and the portion of each character pattern is calculated from the top of each scanning line. , and the distance from the bottom edge of each scanning line to the part of each character pattern. From these values, find the left edge, right edge, top edge, and bottom edge of the character pattern, and then cut out the character pattern. Let's do it.

Further, in the present invention, a counting means for counting the number of bits of a pattern on each scanning line in the above-described scanning area, a counting means for counting the number of bits from the starting end of the scanning line until reaching the pattern portion, and The feature is that a counting means is provided to count the number of bits from the end of the pattern to the pattern part, and the information necessary for extraction is obtained by hardware.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 3 is a diagram for explaining an example of the pattern cutting method according to the present invention.
An example of a line pattern area is shown.

In the figure, as described above, the image is first scanned in the vertical direction (j direction), then moved in the horizontal direction (i direction) by one scanning line, and then scanned again in the vertical direction. It has become.

The characteristic feature of the embodiment of the present invention is as shown in FIG. 3a,
Each time scanning is performed along each scanning line in the vertical direction, the peripheral distribution Np (i) in the vertical direction as shown in Fig. 3d is determined, and a part of the character pattern is and the distance Nb(i) from the lower end of the pattern area, that is, the end of the scanning line, to reaching a part of the character pattern (shown in FIGS. 3 and 3C).

).

In this way, the marginal distribution Np(i), the distance N t (i)
, Nb(i), the marginal distribution N p (i)
The left end 11 and right end 12 of the character pattern are obtained, and the upper end j1 and lower end J2 of the character pattern are calculated by calculating the minimum values of the distances Nt(i) and Nb(i) within the pattern area for one line. The frame for cutting out the character pattern is set accordingly.
The character pattern is cut out.

If the pattern area shown in Figure 3a is scanned, photoelectrically converted, and then sampled and converted into binary data, the marginal distribution Np(i) is the number of bits in the character pattern part in scanning line i. Also, the distances N t (i) and Nb (i) are determined by counting the respective number of bits from the top and bottom edges of the pattern area to a part of the character pattern in scanning line i. .

Furthermore, a further characteristic feature of the embodiments of the present invention is that the values Np (i) and N t (i
) and Np(i) through a median filter to reduce the influence of noise contained in the pattern and to perform more accurate cutting.

This median filter sorts the signal values at a point of interest and multiple points adjacent to that point in order of magnitude, registers the median value as the new value of the signal at the point of interest, and performs these operations. This process is repeated to remove spike-like noise from the signal.

Therefore, we use this median filter as the marginal distribution Np(i)
, distances Nt(i) and Nb(i), respectively, and the values obtained thereby are Npm(i) and Ntm(i
), Nbm(i) as information for extraction, more accurate extraction can be performed.

FIG. 4 shows the configuration of an embodiment of the character reading device according to the present invention, in which 1 is a host microcomputer that controls the entire device, 2 is a microprocessor, and the character reading device 2 includes a microprocessor for cutting out character patterns according to the present invention. The pattern observation device 6, which includes an extraction circuit, 3 a character pattern recognition circuit that recognizes character patterns, 4 an output device such as a floppy disk that outputs the recognition results, and 5 a scanner that scans a predetermined pattern area, is as described above. Al is an interface bus that connects each circuit.

In such a configuration, first, a pattern area corresponding to one line is scanned by the pattern observation device 5, and the pattern within the area is photoelectrically converted and transferred to the character pattern cutout circuit 2, where the character pattern cutout circuit 2, which will be described later, is A cutting process is performed, and the pattern cut out for each character is transferred to the memory in the microcomputer 1 through the interface bus 6.

Next, this cutout pattern is connected to the interface bus 6.
After being transferred to the character pattern recognition circuit 4 via the interface bus 6 for recognition processing, it is once returned to the microcomputer 1 for editing processing, and the result is output to the output device 4 via the interface bus 6.

The microcomputer 1 controls the transfer of various data and edits recognition results.

FIG. 5 shows an example of a specific configuration of the character pattern extraction circuit 2 shown in FIG. 4, in which 20 is an internal data bus;
1 is a microprocessor that controls the entire circuit; 22 is a read-only memory (hereinafter referred to as ROM) that stores a microprogram executed by the microprocessor 21;

), 23 is an interface circuit for connecting to the interface bus 6 in FIG. 4, and 24 is a pattern observation device 5.
25 is a counting circuit for calculating the vertical peripheral distribution Np(i) of the pattern of each scanning line in the scanned pattern area; 26 is the upper end of each scanning line in the scanned pattern area; A counting circuit 27 calculates the distance from the bottom of each scanning line in the scanning pattern area to a part of the character pattern, and a counting circuit 28 calculates the distance from the bottom of each scanning line in the scanning pattern area to a part of the character pattern. A control circuit for storing data and controlling the counting circuits 25 to 27; 29 a memory for storing the results obtained by the counting circuits 25 to 27; and 30 a memory for storing the results obtained by the counting circuits 25 to 27. This is a median filter circuit for performing median filtering.

The microprocessor 21 controls these circuits as a whole and also performs character pattern cutout processing using the results obtained by the counting circuits 25 to 27. The operation of the circuit shown in FIG. 5 will be explained mainly.

It is assumed that the pattern observation device 5 photoelectrically converts the pattern in the scanning pattern area, as well as performs a process of binarizing and converting it into binary data, and outputs it to the character pattern cutout circuit 2 as a digital signal.

First, from the pattern observation device 5, one vertical scanning line (1
binary data (for example, 128 bits).

) is taken in via the input/output control circuit 24, which serves as a normal interface, and stored in the memory of the control circuit 28.

Next, among the data stored in the memory of the control circuit 28,
the desired data portion (eg, consisting of 96 bits).

) is extracted, read, and written to the memory 29 via the internal data bus 20, and simultaneously transferred to the counting circuits 25-27.

The counting circuits 25 to 27 calculate information Np (i) necessary for cutting out character patterns based on the transferred data.
Find N t (i) and Nb (i).

In this way, the information Np(t) y N t (i)
Once t Nb (t) is determined, the result is stored in memory 2.
Write to 9.

The above-described processing is for data for one scanning line, and similar processing is performed for data for the next scanning line.

Then, such processing is repeated over the pattern area for one line.

For example, if one row consists of 1500 scanning lines (columns), it will be repeated 1500 times.

Next, the median filter circuit 30 generates one row of information Np(iLNt(i)) obtained in this manner.

Apply median filtering to Nb(i),
Pattern cutting information N p m (i), N t
m (i) , obtain Nbm(i).

For example, in the case of the marginal distribution Np(i), the marginal distribution Np(n) at the nth scanning line is taken as the value of the point of interest, and from the n-Nth scanning line to the n+Nth scanning line centering on that point, Retrieve the marginal distributions Np(n-N) ~NP(n+N) from the memory 29, rearrange those values in the order of their magnitudes, select and retrieve the marginal distribution with the N+1st magnitude, and focus on its value. It is registered in the memory 29 as cutout information N p m (n) for the n-th scanning line.

Such processing is repeated for each scanning line to obtain a new marginal distribution N P m (i).

Such processing is also performed for the distance information Nt(i) and Nb(i), and the cutout information Ntm (
i) , obtain N b m (i).

Cutting information Npm(i), Ntm(i), Nbm(
For i), the corresponding N p (i), N t
(i) and Nb (i) may be used as they are.

Regarding the principle of the median filter, for example, L, R, Rabiner, et ad: Appli
cations of a Non1inear
Smoothing Algorithm to
5peech Processing, IEE
E Transaction As5P-23, A6.
552-557 (published December 1975).

Thus, according to embodiments of the present invention, information Np(i)
, Nt(i), Nb(i) using a median filter 30
The influence of noise contained in character patterns on pattern extraction can be significantly reduced.

Furthermore, the microprocessor 21 uses the thus obtained cutout information Npm(i) and Ntm(i).

By N b m (i), the left end 1 of each character pattern
1. Find the right edge 12, top edge J1, and bottom edge J2, find the horizontal and vertical center points of each character pattern from these values, set a frame for cutting out, and cut out each character pattern.

FIG. 6 shows an example of a specific configuration of the circuits 25 to 28 in FIG. 5.

The counting circuit 25 for calculating the marginal distribution Np(i) consists of an AND gate 251.8-bit binary counter 252 and a bus driver 253, and the counting circuit 26 for calculating the distance Nt(i) consists of an inverter 261, a flip-flop 262, ANDGATE263, 8-bit binary
Consisting of a counter 264 and a bus driver 265,
The counting circuit 27 for calculating the distance Nb(i) is an OR gate 27
1.8-bit binary counter 272 and bus
The control circuit 28 further includes a control pulse decoder 281 and a 256-bit random access memory (hereinafter referred to as RAM).

) 282.8-bit address counter 283 16-bit shift register 284 capable of parallel input/output, bus
It consists of a driver 285 and a control signal line 286 into which control signals from the microprocessor 21 are input.

It is assumed that the internal data bus 20 has a 16-bit configuration.

The operation shown in FIG. 6 will be explained in detail below.

It is assumed that the binary data of each scanning line (one column) of a one-line character pattern area consists of eight words (one word is 16 bits).

First, binary data scanned by the pattern observation device 5 shown in FIG. 5, photoelectrically converted, and further binarized is input word by word to the RAM 282 via the human output control circuit 24 as an interface.

To do this, first, the address counter 283 is reset to 0 by the control signal C1 from the control pulse decoder 281, and then one word of binary data is transferred to the shift register 28 by the control signal C2 from the decoder 281.
Write in 4.

Next, the content of the address counter 283 is incremented by 1 by the control clock pulse CP from the decoder 281, and the shift register 284 is shifted, so that the binary data in the shift register 284 is transferred bit by bit to the decoder 281. The data are sequentially stored in the RAM 282 in accordance with the write control signal C3 from .

When one word of binary data has been stored in the RAM 282, the address counter 283 is reset again by the control signal C1, and the next one word of binary data is transferred from the pattern observation device 5 to the internal data bus 20. The signal is inputted to the shift register 284 via the signal line, and stored in the RAM 282 by repeating the same operation as described above.

This operation is continued until one scanning line (eight words) of binary data is stored in the RAM 282.

In this way, the binary data for scanning lines is
Once written in M282, the written 8th edict (
Only 6 words (96 bits), mainly character patterns, are extracted from the 128 bits) and written into the memory 29 via the shift register 284.

The pattern observation device 5 can obtain binary data of 8 words for each scanning line, taking into account scanning positional deviations, etc. However, the actual number of words required for extraction is 6 words, mainly in the character pattern part. It is sufficient.

Therefore, the microprocessor 21 detects the actual center of the character pattern, sets the range to be extracted, and selects the first bit position for the first bit position. The address is set in address counter 283 via internal data bus 20.

Subsequently, a read control signal C4 is applied from the decoder 281 to the RAM 282 to start reading from the RAM 282.

At this time, a control clock pulse CP is applied from the decoder 281 to the address counter 283 to increment its contents one by one, and the contents of the shift register 284 are shifted one bit at a time.

As a result, data read from the RAM 281 is sequentially stored in the shift register 284.

When data for one word is stored in the shift register 284, the bus buffer 285 is opened by the control signal C5 from the decoder 281, and the data is stored in the shift register 284.
Data for one word of 4 is sent to the memory 29 via the bus 20 and written there.

Similarly, when the next word is stored in shift register 284, bus buffer 285 is reopened and the data is written to memory 29.

This operation is repeated over 6 words (96 bits).

On the other hand, prior to reading from the RAM 282, the control signal C6 from the decoder 281 causes the counter 252,
264 and 272, the flip-flop 262 is set, and the "l" signal is applied to the AND gate 263.

Subsequently, the decoder 281 outputs the control clock pulse CP, and the counting circuit 25 outputs the control clock pulse CP and R.
An AND gate 251 performs a logical product with the readout output of the AM 282, and a counter 252 counts the output pulses.

That is, the counting circuit 25 counts the number of 1'' data bits read from the RAM 282, in other words, the number of bits in the character pattern portion.

Next, in the counting circuit 26, the control clock pulse CP,
Set output of flip-flop 262 and RAM 282
The AND gate 263 performs a logical AND operation with the output of the inverter 261, and the output is sent to the counter 26
Count by 4.

Furthermore, since the flip-flop 262 is reset by the 1" output from the RAM 282, the period from the beginning of the binary data for the scanning line until the "■" signal (signal representing the character pattern part) appears. It is in a set state, and the counter 264 counts the number of clock pulses CP during that period.

Therefore, the counting circuit 26 calculates the distance from the starting end of the scanning line to the character pattern portion.

In addition, the counting circuit 27 repeatedly resets the counter 2γ2 while the "1" signal is output from the RAM 282, and starts counting the control clock pulses CP from the time when the "1" signal disappears. Counting continues until the CP is sent out from the decoder 281.

Therefore, the counting circuit 27 calculates the distance from the f end of the character pattern to the end of the scanning line.

In this way, when the reading of the binary data for the scanning line is completed, the control signal C7 from the decoder 281 is
, C8 and C9 to bus buffer 253.256
and 273, the information stored in counters 252, 264 and 272 N p (1) + N
t(') and Nb(i) are sent via data bus 20 to memory 29 and written there.

The binary data for the next scanning line is transferred to the pattern observation device 5.
, the information Np is obtained by processing similar to that described above.
(i), N t (i) and Nb(i) are determined and the results are written into the memory 29.

This operation is repeated over all the scanning lines in the character pattern area for one line, and the results are written into the memory 29.

FIG. 7 shows an example of the memory configuration of the memory 29 portion in FIG.
The value obtained in 7. Np ('') + N t (i),
An area 292 for storing Nb(i) and cutout information Npm(i), Ntm(i), obtained by passing through a median filter 30, which will be described later.
It consists of an area 293 for storing Nbm(i), an area 294 for storing data of a single character pattern area cut out as described later, and various working memory areas 295.

FIG. 8 shows an example of a specific configuration of the median filter 30 shown in FIG.
303 is a control circuit that controls the operation of the switching circuit group; 304 is a switch circuit group that compares the contents of adjacent registers 3011 to 301-5 and replaces the contents depending on the size; 303 is a control circuit that controls the operation of the switching circuit group; A shift register 305 outputs data manually inputted from the data bus 20 in parallel, and an input line 306 for a control signal from the microprocessor 21 represents a bus buffer.

Further, FIG. 9 shows an example of a specific configuration of the exchange circuit group 302-1 in FIG. 8, where 310 is a comparison circuit, 311 to
1 and 311-2 are registers 301-1 and 3 (H
-2 registers 312i and 312- that temporarily store the contents of
2 indicates an AND gate.

Although the control signal to the comparison circuit 310 and the output signal from the comparison circuit 310 are multi-bit signals, they are shown as 1-bit signals in the figure for the sake of simplicity.

Further, although one AND gate 311 to L3122 is shown, it goes without saying that in reality, it is only the number of signal bits.

In such a configuration, a process for obtaining the marginal distribution N p m (i) for extraction from the marginal distribution Np(i) will be described.

Marginal distribution of the first to fifth scanning lines Np1 to Np
5 is sequentially input to the shift register 304, and the control circuit 3
The contents of the shift register 304 are sequentially stored in the registers 301-5 to 3011 by the clock pulse CP1 from 03.

As a result, information Np1 to Np5 of the first to fifth scanning lines are stored in the registers 3011 to 301-5, respectively.

Therefore, first, the switching circuit groups 302-1, 302-3 are operated by the control signal C11 from the control circuit 303, and then the switching circuit groups 302-2.
302-4 to perform necessary data exchange.

Now, considering the operation of the exchange circuit 302-1 for exchanging the contents of the registers 301-1 and 301-2, as shown in FIG.
The contents of 2 are stored in registers 3111 and 311-2, compared by comparison circuit 310, and stored in register 301-2.
The comparison circuit 310 outputs the control signal C13 only when the content of the register 301-1 is greater than the content of the register 301-1.
11 and 311-2.
-1 and 301-2 data to AND gate 3121
and 301-2 respectively by exchanging via 312-2.
and stored in 301-1.

By repeating such operations, the outputs 01 to 04 of each comparison circuit 310 of the switching circuit group are transferred to the control circuit 303.
By monitoring the data, it is detected that no data exchange is being performed, thereby terminating the reordering operation, opening the Yoribus buffer 306 in response to the control signal C14, and redirecting the data in the central Ludisk 301-3 to the data. bus 2
Send to 0.

At this time, the registers 301-1 to 3015 store the values of the sizes in that order, so the central register 301-3 stores the data of the third value. The area 293 of the memory 29 is used as the new peripheral distribution Npm3 for cutting out in the third scanning line.
is memorized.

Next, the marginal distribution Np(
i) sequentially through the shift register 304 to the register 30
1-1 to 301-5, and performs the same processing as described above to obtain a new peripheral distribution Npm4 for the fourth scanning line, and stores the result in area 293 of memory 29.

Hereinafter, similarly, the peripheral distribution N p (i) of each scanning line
A new marginal distribution Npm(i
) can be obtained.

The distances N t (i), Nb (i
), a new distance N for extraction is similarly set.
tm(i) and −N b m (i) are determined and the results are stored in area 293 of memory 29 .

In addition, each information Npm in the first and second scanning lines from the beginning and the first and second scanning lines from the end
(i), Ntm(i), Nbm(i) are the corresponding information Np(i), Nt(i), Nb(i)
can be obtained by using as is.

Next, the information Np(
Cutting information N pm(i) , N tm(i) , N b obtained by passing t) , N t (t) , Nb (t) through the median filter 30
Based on m(i), the microprocessor 21 cuts out each character pattern from one line of binary data stored in the area 291 of the memory 29.

The character pattern extraction process by the microprocessor 21 will be described in detail below.

Figure 10 shows the state of the marginal distribution N p (i) in the character pattern area of a one-line sentence, and Figure 11 shows the state of the marginal distribution N p (i) in the character pattern area of a one-line sentence.
This is an enlarged view of a part of the figure.

In the figure, B is the distance (number of bits) from the left edge of the pattern area for one line to the left edge of the first character pattern frame set in advance, and P is the horizontal pitch of each character pattern frame ( n is the maximum number of characters in the pattern area for one line (for example, n = 25), W is the maximum width of each character pattern in the horizontal direction (number of bits) (for example, W = 44), k is Character pattern frame number (1≦≦n
).

Moreover, FIG. 12 shows the cutout state of the character pattern, where the dotted line indicates the cutout frame of the character pattern, 13 and i4 are the left and right ends of the cutout frame of the character pattern, respectively, and J3 is the cutout frame of the character pattern. Show the edge.

FIGS. 13a, b, and 13c show a flowchart of an example of character pakun extraction processing.

The character pattern extraction process will be described in detail below according to this flowchart.

First, in step S1 shown in FIG. 13a, the number k of the character pattern frame is reset to 0, and in step S2, this k is set to 1.
In step S3, it is determined whether ≦n, and the process proceeds to step S4.

Step S4. In S5, the total sum Sk of the peripheral distribution Npm(i) within the set character pattern frame is calculated as follows (1)
Calculate according to the formula.

Next, in step S6, the value Sk obtained in this way
is compared with a certain threshold value St, and when Sk<St, it is determined that there is no character pattern within the frame, the process returns to step S2, and the value of k is increased by 1.

Then, when it is kin, the above-mentioned processing ends.

Further, when Sk≧St in step S6, it is determined that there is a character pattern within the frame, and the process moves to the next step S7.

Here, St is a threshold value for testing the presence of characters, and is, for example, Stγ10.

In step S7, the left end i5 and right end i6 of the range in which the marginal distribution Npm(i) is to be examined are determined as follows (2), (3
) is calculated using the formula.

When 15 t 16 is determined in this way, Npm
(15) Check Npm(to) and proceed to the next step S8
Now, it is checked whether the following equation (4) is satisfied, and when the equation (4) is satisfied, the process proceeds to step S21 in FIG. 13, and when the equation (4) is not satisfied, the process proceeds to step S9. In S9, the following (
5) Check whether the formula is satisfied, and when the formula (5) is satisfied,
The process proceeds to step S31 in FIG. 13, and if the condition is not satisfied, the process proceeds to step S10.

In step S10, it is checked whether the following equation (6) is satisfied, and when the equation (6) is satisfied, that is, Npm(i5)
- If 〇 and NPm(i6)>0, step S1
1, i is incremented by 1, and the process returns to step S8.

Moreover, when the formula (6) is not satisfied, that is, Np+i−
When +(i5)>O and Npm(i6)=O, 15 is decreased by 1 in step 812, and the process returns to step S8.

In step S21 of FIG. 13, il is set to i5+1, and in step S22, it is checked whether the following formula (7) is satisfied, and if formula 7) is satisfied, Stella 7'S24
If the equation (7) is not satisfied, proceed to step S23.
The process proceeds to step 822, increments 11 by 1, and returns to step 822.

In step 824, 12 is set to 16-1, and in the next step S25, the following equation (8) is checked. When equation (8) is satisfied, the process moves to step 826, where i2 is decreased by 1, and the process proceeds to step S25. Returning, if the equation (8) is not satisfied, proceed to step 841.

In addition, in step 831, 11 is set to 15-1, and in the next step 832, the following equation (9) is checked, and when the equation (9) is satisfied, the process moves to step S34, and if the condition of the equation (9) is satisfied. If there is not, the process moves to step S33, il is decreased by 1, and the process returns to step S32.

In step S34, 11 is increased by 1, and in the next step S35, 12 is changed to i6+1.

In the next step S36, the relationship of the following αO) formula is examined,
If the formula (10) is satisfied, proceed to step 838 and i
2 is decreased by 1, and if the equation 00) is not satisfied, i2 is increased by 1 in step S37, and the process returns to step S36.

Thus, steps 821-826 and step 8
31-838 is the left end i of the character pattern.

and the right end 12 can be found.

In step S41, the value 11 obtained in this way is
, 12 and the cutout information N tm(i) and N bm(i) stored in the memory 29, the upper end j1 and lower end J2 of the character pattern are determined based on the following equations (11) and (12>). Calculate.

That is, the distances Ntm(i) and Nb in the scanning line from the left end 11 to the right end +2 of the character pattern found
Find each minimum value of m(i) and use it as the value of the upper end j1 and the F end j2 of the character pattern.

Next, in step 842, the above-mentioned values J1, 32 and
The horizontal and vertical widths of the character pattern cutting frame (
number of bits) WH and Vvv (for example, WH=42,
Wv-48), the top end j3, left end i3, and right end i4 of the cutting frame are calculated according to the following equations 43) to (15).

However, if j3<0 in steps 843 and 844, j
3=0 in step S45, and in step S46, when j3>W■, the replacement is performed with j3=W■.

Next, in step 847, the six words of binary data of each scanning line within the range of i3≦i≦i4 are sequentially read out from the memory 29 in FIG. 5 and stored in the RAM 282 in FIG. 6 in the same manner as described above. Forward.

When the transfer is completed in this way, the value j3 is set in the address counter 283, the three words (48 bits) after that address value are read out from the RAM 282, and the memory 2
It is transferred to the cutout pattern area 294 of No. 9.

These operations are performed on the binary data of all scanning lines within the range i3<i<i4.

Note that the same control signals as shown in FIG. 6 can be used to perform these controls, so their explanation is omitted here.

Finally, in step 848, 1 in the area of memory 29
The cutout pattern for characters is read out and sent to the host microcomputer 1 via the internal data bus 20 and interface 23.

In this way, when the cutout pattern is sent to the host microcomputer 1 shown in FIG. 4, the pattern is sent to the pattern recognition circuit 3.

The pattern recognition circuit 3 compares the cutout pattern with the standard pattern, recognizes the cutout pattern, and then stores the result in the memory of the output device 4.

As this pattern recognition circuit 3, for example, one having a structure as shown in Japanese Patent Application No. 53-68791 by the inventors of the present invention can be used.

As can be seen from the embodiments described above, according to the present invention,
Since all the information necessary for cutting out the pattern can be obtained when scanning each scanning line in the scanning pattern area, the pattern can be cut out in an extremely short time.

Also, the values Np (i), N t (i), Nb
By using a hardware configuration as shown in FIG. 6 to obtain (i), character patterns can be cut out not only with an extremely simple circuit configuration, but also at higher speed.

Note that in the embodiment described above, the information Np(iLNt(i)
, N b (i), the counting circuit 25 ~
27. Although a special circuit consisting of the control circuit 28 is provided, this information may also be obtained by a microprogram of the microprocessor 21.

Also, the information Np (i), N t (i), N
b From (i), extract information Npm(i), Ntm(
i), Nbm(i), we use a special median
Similarly, without providing a filter, microprocessor 2
The processing may be performed by software using the microprogram No. 1.

Furthermore, the specific configurations of the counting circuits 25 to 27, the control circuit 28, and the median filter 30 are not shown in FIGS. 6, 8, and 9, and various modifications are possible.

In short, these circuits count the number of bits of the pattern on each scan line in the scan pattern area, count the number of bits from the beginning of the scan line to the pattern part, and count the number of bits from the end of the scan line until reaching the pattern part. Any configuration may be used as long as it counts the number of bits and performs median filtering on the results of these counts.

Furthermore, in the above example, the information obtained by the counting circuit is passed through the median filter to obtain the cutting information, but the information obtained by the counting circuit is used as it is as the cutting information, - The filter part may be omitted.

Furthermore, in the above example, the peripheral distribution N p (i) of each scanning line is determined by the counting circuit 25, but it is only necessary to detect the existence of a character pattern in each scanning line using a flip-flop or the like. Any method may be used as long as the horizontal center of the cutout pattern frame can be determined.

Further, in the above-mentioned embodiment, an example of a character reading device that recognizes a character pattern is shown, but the present invention is not limited to this. For example, when object patterns exist separately in a predetermined direction, It goes without saying that the present invention can also be applied to an object recognition device that cuts out an object pattern and recognizes it.

As described above, according to the present invention, when scanning is performed along each scanning line, the existence state of the unit pattern in each scanning line and the distance from the start and end of each scanning line to the unit pattern are determined. By finding the information and setting the pattern cutting frame based on that information, pattern cutting can be performed very quickly.

Further, since this information can be obtained using a counting circuit, pattern cutting can be performed with a very simple configuration and at high speed.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram for explaining a method of scanning a pattern area for one line, FIG. 2 is an explanatory diagram for explaining a conventional pattern cutting method, and FIG. 3 is an explanatory diagram for explaining an example of a pattern cutting method according to the present invention. FIG. 4 is a block diagram showing an example of the overall configuration of the pattern recognition device according to the present invention, and FIG. 5 is a block diagram showing an example of the specific configuration of the pattern extraction circuit portion of FIG. 6 is a circuit diagram showing an example of a specific configuration of the counting circuit and control circuit portion of FIG. 5, FIG. 7 is a diagram showing an example of a memory configuration of the memory portion of FIG. 5, and FIG. 9 is a circuit diagram showing an example of a specific configuration of the median filter portion of FIG. 8, and FIGS.
12 and 12 are a peripheral distribution diagram and a pattern diagram for explaining the pattern extraction process, and FIG. 13 is a flowchart showing an example of the pattern extraction process. 2...Pattern cutting circuit, 5...
Pattern observation device, 21...Microprocessor, 25-27...Counting circuit, 28...
- Control circuit, 30...median filter.

Claims (1)

[Scope of Claims] 1. Scanning for obtaining electrical signals by sequentially scanning a pattern area including a plurality of unit patterns arranged separately in a predetermined direction along a scanning line in a direction substantially perpendicular to the predetermined direction. a pattern cutting means for separating and cutting out each unit pattern in the pattern area based on an electric signal from the scanning means, and a recognition means for recognizing the unit pattern cut out by the pattern cutting means. In the pattern recognition device, for each scan along each scanning line by the scanning means, a first detection signal indicating the existence state of the unit pattern in each scanning line, and a first detection signal indicating the existence state of the unit pattern in each scanning line, and The second and third detection signals indicating the distance to reach are determined, and the pattern is cut out by setting a cutting frame for each unit pattern based on the signals corresponding to the first to third detection signals. ,
A pattern cutting method characterized in that the cut out pattern is input to the recognition means. 2. Median filtering is applied to each of the first to third detection signals to obtain fourth to sixth detection signals, and a cutting frame for each unit pattern is set based on the detection signals. A pattern cutting method according to claim 1, characterized in that: 3. The pattern cutting method according to claim 1 or 2, wherein each of the pattern areas is composed of a single line character pattern area including a plurality of character patterns arranged separately in the row direction. . 4. a scanning device that obtains a binary electric signal by sequentially scanning a pattern area including a plurality of unit patterns arranged separately in a predetermined direction along a scanning line in a direction substantially perpendicular to the predetermined direction; A pattern recognition device comprising a pattern cutting device that separates and cuts out each unit pattern in the pattern area based on an electric signal from the scanning device, and a recognition device that recognizes the unit pattern cut out by the pattern cutting device. The apparatus includes: a first counting means for counting the number of bits of an electrical signal in a unit pattern portion on each scanning line in a scanning pattern area; a second counting means for counting; a third counting means for counting the number of bits of the electrical signal from the end of each scanning line to the unit pattern portion; and counting results of the first to third counting means. A pattern cutting device comprising a cutting means for setting a cutting frame for each unit pattern based on a corresponding value and cutting out the pattern. 5. The pattern cutting device according to claim 4, wherein the cutting means has a median filter for applying median filtering to the counting results of the first to third counting means.