JPS5994181A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To recognize exactly a pattern with simple constitution by reading the tip of a sheet of paper and measuring the tilt of the paper by line sensor and correcting the tilting by using measured quantity of the position of left and right ends. CONSTITUTION:A sheet of paper 10 is charged in a paper feeding mechanism 20, and a picture pattern on the paper 10 is photoelectrically converted by an observing mechanism 30 using a line sensor. A signal that indicates this photoelectrically converted image is supplied to a tilt detecting circuit 40, an X-axis direction position detecting circuit 50 and a correcting circuit 60. The circuit 40 works when the paper 10 is fed newly, and reads the tip of the paper 10 and measures the tilt of the paper 10. The circuit 50 detects left and right ends of the paper 10 and measures the position and width of the paper 10 and inputs the result of detection to the circuit 60. The inclination measured by the circuit 40 is added to a recognition processing section 70, and at the same time, the corrected value of deviation in the direction X is added to the processing section 70 by the circuit 60. Recognition of character pattern is performed by the processing section 70 and supplied to a display 80 and a memory 90.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a pattern recognition device, particularly an optical character recognition device.

[Background of the invention]

□ Conventionally, paper fed to a photoelectric character recognition device was only allowed to be tilted relative to the character observation unit within a certain tolerance range, and moreover, within a fairly small tolerance range. As shown in FIG. 1, the above-mentioned certain tolerance range is usually such that a character pattern on another line above or below does not fit within a rectangle 3 surrounding one line of character pattern. Therefore, in the above range, when cutting out a character pattern, it is possible to cut out each line horizontally, and the cutting out method and its implementation circuit are relatively simple.

However, in optical character recognition devices where paper is fed in a wide variety of sizes, transported at high speed, and recognized at high speed, the paper is fed at an inclination within the above tolerance range. It's hard to do. Therefore, in order to keep the paper feeding mechanism within the above-mentioned tolerance range, the paper feeding mechanism becomes complicated and expensive.

Furthermore, for a piece that is as inclined as shown in FIG. 2, the cutting process is more complicated than in the past, and it is extremely difficult to cut it out at high speed.

Therefore, in the past, a flying spot scanner (flying spot scanning mechanism) was used as an observation means to change the scanning form according to the inclination and capture the pattern in a form that removed the influence of the inclination.

However, in this case, there was a drawback that the observation section was expensive.

[Purpose of the invention]

Therefore, the purpose of the present invention is to make it possible to wait for customers even on a large inclination as shown in Fig.
And it should be cheap.

[Summary of the invention]

In order to meet the requirements of high speed and low cost, the apparatus according to the present invention uses a one-dimensional photoelectric conversion element (hereinafter referred to as a line sensor) as an observation means, and also enables tilt correction.

When using a flying spot scanner as an observation method, it is possible to randomly access any point on the paper and photoelectrically convert it, making it relatively easy to capture character patterns while correcting the inclination. Met. However, since the device according to the present invention employs a line sensor as an observation means for the above purpose, the conventional method using a flying spot scanner is useless.

Therefore, in the apparatus according to the present invention, the line sensor also reads the leading edge of the paper to measure the inclination of the paper (relative inclination to the line sensor), and the line sensor further measures the positions of the left and right edges of the paper. By doing so, a means for correcting the inclination using the same measured quantity is provided.

In general, when paper is fed at an angle, a pattern shift also occurs in the lateral direction (hereinafter referred to as the X-axis), so the correction of the inclination referred to here also includes the correction of the shift in the X-axis.

[Embodiments of the invention]

Hereinafter, the present invention will be explained in detail with reference to Examples.

FIG. 3 shows a functional block diagram of the device according to the invention. In the figure, a paper 10 is loaded into a paper feed mechanism 20, and photoelectrically converted by an observation mechanism 30 using a line sensor. A signal representing the photoelectrically converted image is supplied to a tilt detection circuit 40, an X-axis direction position detection circuit 50, and a tilt correction circuit 60.

The tilt detection circuit 40 operates when paper is newly fed and measures the amount of tilt by reading the leading edge of the paper.

The X-axis position detection circuit 50 operates constantly when the paper is under the observation mechanism, detects the left edge and right edge of the paper, and measures the position and width of the paper.

A tilt correction circuit 60 receives the above three signals.

The recognition processing unit 70 converts the tilted image into an image without tilt, and further applies correction for the deviation in the X-axis direction.
Transfer to. Here, the recognition process also includes a cutting process. The recognized results are transferred to a display 80 for displaying the results or a memory 90 for storing the recognition results.

Here, a paper feeding mechanism 20, an observation mechanism 30 using a line sensor, a recognition processing section 70. Display 80, memory 9
0 can be configured using a known technique, so the explanation will be omitted.

Inclination detection circuit 40, which is important in the present invention, is as follows: The X-axis direction position detection circuit 50 and the tilt correction circuit 60 will be explained in detail.

First, the tilt detection circuit will be explained. FIG. 4 is a diagram for explaining the principle of tilt detection. In the figure, a frame 101 is an area that can be observed by a line sensor, and a white part 10
2 is an image of paper, and the shaded area 103 is an image of the area where no paper exists. By taking appropriate measures, the pattern image (binary pattern) in the areas without paper is made to have the value 1, and the white areas are made to have the value 0.

In the figure, the vertical axis corresponds to the time of observation. Since the paper passes at a constant speed past the fixed line sensor, the line sensor's image of the paper appears to be tilted as shown in the figure.

Therefore, there is a time t when there is no paper at several locations on the line sensor. The length of black from t1 to time t1 where there is paper 11, 12,
If you measure ......, the inclination of the paper is 13-11
, 14-12... can be easily obtained.

In this embodiment, the length of black is measured at four points as shown in FIG. The length of the line sensor is 2048 bits, and the distance between measurement points is 512 (bits) between 13 and 'I'4 and the measurement points of the pillars. Therefore, the slope θ
8 can be found as in equation (1).

A more detailed block diagram of the circuit 40 for determining the above-mentioned slope θS is shown in FIG.

The operation and configuration of the tilt detection circuit 40 will be explained with reference to FIG.

First, the signal 5YNO201 is a basic clock signal,
This is a pulse signal synchronized with sampling each bit of the line sensor. Signal 5LNO205 is one scan (
(2048 bits) This is a row synchronization signal output to each row synchronization signal.

Signal 5PF204 is an initial reset signal that is output when a new sheet is fed. The signal P2O2 is a video signal (obtained by subjecting the output of the line sensor to appropriate binarization processing) and is 0.0 when the pattern is white (or when there is paper). It takes a value of 1 when it is black (or when there is no paper).

The circuit 120 is an address counter indicating the address of the point currently being sampled in the X-axis direction, and is 0 to 20.
It takes a value of 47. Address counter 120 is reset to signal 205 for each new row and incremented by one by signal 201 for each sampling.

Circuit 131 stores the contents of address counter 120 and the value 70.
4, and if they match, n is a comparison circuit that outputs the signal 211 as 1. Circuits 132, 133, and 134 are respectively comparison circuits that similarly compare with the values 1216, 832, and 1344. Therefore, signals 211, 213, 21
2. and 214 are signals whose value is 1 only when sampling measurement points 11, 12, 13, and t4, respectively.

The AND circuit 141 ANDs the signals 211.201 and the inverted video signal 203, and outputs a signal 241 of 1 only when white, that is, a paper portion is detected at the corresponding measurement point. Circuits 142, 143 and 144 are similar to circuit 141.

Circuit 151 is reset by signal 204.

This is a counting circuit that counts the number of times that the signal 241 is 1 after that, and sets the signal 221 to 1 when the number of times that the signal 241 is 1 is greater than a predetermined number (nl). This is provided to avoid the influence of the unstable state of the paper immediately after feeding, and nl is 16
It is sufficient to set it to a certain degree. That is, signal 22
1 is a signal that is O from when a new sheet of paper begins to be fed until the leading edge of the same sheet passes through the 11 measurement points of the sensor, and then becomes 1. Circuits 152, 153, and 1
54 is similar to the circuit 151.

Circuit 161 is! , is reset by the signal 204, is incremented by 1 by the signal 205, and furthermore, when the signal 221 becomes 1, further counting up is prohibited. Since the signal 205 is a pulse output for each line, the counter 161 counts the number of black lines until the white one arrives. Circuits 162, 1638 and 164 are similar to circuit 161.

The circuit 171 is 13-1. , the circuit 172 is! This is a subtraction circuit that calculates 4-12, respectively. Circuit 173 is circuit 1
This is an addition circuit that further adds the subtraction results of 71 and 172. The addition result 251 is sent to the latch circuit 183 as a latch signal 2.
31 and latched. The signal 252 is a binary number representing the value θ8, and the decimal point is LSB (Least,
It can be considered as the 10th bit counting from 51gn1jic, anti Bit). However, in reality, the latch circuit 183 uses the O latch signal 23 consisting of 6 bits.
1 is created by an AND gate 181 and a delay circuit 182 having a delay time equal to or greater than the delay time in the adder circuit.

The AND gate 181 outputs 1 when the leading edge of the original sheet passes through all measurement points (4 points). Also signal 231
is also output to the recognition processing section 70 as a signal PPR.

Next, the X-axis direction position detection circuit 50 will be explained. A detailed block diagram of the same circuit is shown in FIG. The position information to be detected is shown in FIG. Andori, (Lo.

L is the length of black ground on the left and right sides within one scan. It is. However, L3-L, -L.
(2) The circuit operates as follows. In other words, the video signal P
321 and the basic clock signal 5YNO 322 are generated by an AND gate 301 to generate a pulse signal 331 which becomes 1 on a black background. Further, an AND gate 302 generates a pulse signal 332 that becomes 1 on a white background. Lo is counted by a counter 31]. The counter 311 uses the row synchronization signal 5LNO32
3 is reset to O, and signal 331 counts up (
+1). Further, once the signal 334 becomes 1, the counter 311 becomes in a count amplifier inhibited state.

Signal 334 is created as follows. The circuit 303 receives the pulse signal 332 that becomes 1 on a white background and the basic clock signal 322, and generates a signal 334 that becomes 1 only when the white background pulse continues for a predetermined number n2. This is to avoid detecting a black dot added as noise on a white background and erroneously recognizing the white background as a black background, and a value of about 8 is sufficient for n2.

Ll is counted by a counter 312. counter 31
2 is reset to O when the signal 334 or the signal 323 is 1, that is, when the signal 333 is 1, and is counted up by the signal 331.

The signal 5LN1324 is a pulse that becomes 1 only at the end of one scan, and the contents of the counters 311 and 312 are latched to the latch 31.
3 and 314, respectively.

It is latched by latches 313 and 314. and Ll are input to the subtraction circuit 315. The output of the subtraction circuit 315 is L3
It is.

Next, the stubborn correction circuit 60 will be explained. FIG. 7 shows a block diagram of the same circuit. 8 and 9 are waveform diagrams of main signals of the circuit. Further, FIG. 10 shows the pit configuration of the main registers in the same circuit.

In FIG. 7, the tilt correction circuit 60 is a 2048 x 32 bit RAM (random access memory IJ) 4.
It consists of an address calculation circuit section for 01 and so on.

From this, it is possible to perform tilt correction by looking at an area corresponding to 32 one scanning line consisting of 2048 bits in the X-axis direction on the paper.

The principle of tilt correction is based on the synchronization signal (SY) that specifies whether the write mode is the read mode for the AM401.
N2) In synchronization with 483, video signal P471 is R, AM4
01 sequentially, and the contents are read out in synchronization with the signal 5YN2 while changing the address according to the slope. The address for writing is (yo, xo
), the read address is (y, x). These addresses are Y.

Or the top 5 pits and X to represent Y. or X
It has a total length of 16 bits including the lower 11 bits that represent the .

When writing, write Xo sequentially from O to 2047 +
1, and Y when Xo returns to 0 again from 2047. Ha+
oXo and Yo that are set to 1 are stored in register 402. and 4
The contents of these registers are set to 1 by signals 5YNO and 5LNO, respectively.

When reading, X corresponding to the lower 11 bits of the read address is calculated using the following equation.

X=Lo+Xo (3) The addition of equation (3) is performed by the addition circuit 411.

Lo is the output of the aforementioned X-axis direction position detection circuit 50 and indicates the address of the left end of the paper. In Equation (3), since the blank reading starts from the left edge of the paper, the positional deviation in the X-axis direction is effectively removed.

Note that the width of the paper is small compared to the length of the line sensor (2048 bits), and normal misalignment will not cause the paper to go outside the scanning area of the line sensor.
11 bits is sufficient.

Next, Y corresponding to the upper 5 bits of the read address
You can ask for it in the following way.

If the shift in the X direction per 1 bit in the X direction, that is, the value of
, Xo), but the read address (y, Well calculated.

(nl When 5KEW>0, write address (Yo
, Xo), so that the read address (y, x) is set so that the content B2 of the part written on the paper 102 is read out, instead of the content A2 of the content A2 of Explode.

The circuit that calculates equations (4) and (5) is the addition circuit 413, and the circuit that calculates the initial value H1 is the subtraction circuit 412.

The operation of the circuit unit 484 for multiplying 2048=2 only requires shifting the decimal point position, and there is no need to perform multiplication.

The multiplexer 415 uses the sign bit 461 of the register 405, which holds the value 5KEW, to set the initial value Y in equation (4). The initial value of equation (5) (Yo-2048xSK
f2W).

The multiplexer 416 selects one of the outputs of the multiplexers 413 and 415 according to the selection signal (8LCT) 462, but selects the output of the multiplexer 415 only when setting an initial value.

The upper bit Y of the read address of the RAM 401 is latched into the register 404 by the signal 463.

The latch signal 463 is activated by the OR gate 426 when either the set signal (SgTY) 464 or the output of the AND gate 426 between the above-mentioned signal (SYNO) 466 and the synchronization signal (PPRI) 465 described below becomes 1. The output will be 1.

5KEW is the output θ8 of the tilt detection circuit 40, and if the tilt exceeds 32 scanning lines in the X-axis direction, it is corrected as the tilt equivalent to 32 scanning lines.
It can be obtained by performing the following calculation using the limiter 421.

However, the sign of sign(θ8) is shown.

It is also possible to select an appropriate value other than 32 as the numerical value stored in the register 422, depending on the degree of slope that actually occurs.

The signals in FIG. 8 will be explained. Signal SPF is a pulse signal that becomes 1 when paper is newly fed. Signal PPR is a level signal that becomes 1 when paper reaches below the line sensor. Signal 5LNO is synchronized with scanning one line. The scanning start signal, signal PPRI, which becomes 1 when signal PPR becomes 1, becomes 1 when 32 scans are completed, that is, when the contents of RAM 401 are filled with new video signals. It's a signal. The signal 5ETY is a signal used to latch the initial value in the register 404.

FIG. 9 shows the latch signal 463 of the register 404 described above in more detail.

As shown in FIG. 9, when the 5LCT signal is 1, the SgT
The Y signal becomes 1, and the value Y selected by MUX416
. is set (latched) in the register 404. signal 5
YNQ is a 2048 pulse train for adding 1 to the value of X.

Figure 10 shows register (sxgw) 405, register (Y
) 404, register (Yo) 403, and register (Xo) 402. register 405
consists of 15 bits, and the lower 6 bits contain the signal 252 (
θ, ) in FIG. 5 is set, and the upper 9 bits are set to 0.

The operation of the RAM 401 will be briefly explained with reference to FIG. R.A.
M401 is called a 2-bot RAM, and when the synchronization signal (SYN2) 483 is 1, it is in the write mode, and when it is 0, it is in the read mode. Writing is performed at the time of falling from 1 to 0. In Figure 11, 501
and 502 are the write data! FIG. 4 is a waveform diagram showing the time when the value of read data is correctly defined.

The hatched portion is the time when the value is undefined.

The output of the slope normalization circuit 60 described above is a corrected video signal (CP) 472 (signal 61 in FIG. 3). A feature of the apparatus according to the present invention is that, by simply adding the circuits 40, 50 and 60 shown in FIG. 3, even when the paper is tilted, the same video signal as when it is not tilted can be obtained as the signal 61.

〔Effect of the invention〕

As explained above, in the apparatus according to the present invention, by providing a processing section that performs tilt correction between the character pattern observation section and the recognition processing section, the paper is greatly tilted with respect to the line sensor (pattern observation section). Even if the paper is fed in the same way as in 9 when the paper is not tilted, it can be processed in exactly the same way.

In principle, it is possible to correct the tilt by including it in the cutting processing in the cutting processing section, but in that case, the processing content becomes complicated and the amount of calculation increases, so high speed is required. This method is not suitable for devices. On the other hand, in the present invention, since the tilt is corrected as a pre-processing independent of the cutting process, the conventional cutting method that allows simple and high-speed processing can be used as is.

Therefore, in the apparatus according to the present invention, since the permissible range of inclination is wide especially when performing high-speed paper feeding, a relatively inexpensive paper feeding mechanism can be adopted, and high-speed cutting processing can be performed.
Advantages include the ability to use line sensors with superior structure.

In this embodiment, one line sensor covers the entire screen, but if the paper size is larger or the number of bits of the line sensor is smaller, multiple lines/sensors may be used. It is conceivable to take images by combining sensors, but even in such a case, the method of the present invention can be applied to each line sensor. However, in this case, only the tilt is corrected without performing position correction in the X-axis direction, and the position information is not corrected. is used in the cutting section to correct one position with equal constraints.

Additionally, in this embodiment, four measurement points are used to detect the inclination of the paper, but this is to reduce errors when the leading edge of the paper is not straight but uneven. be. This number is not limited to four, and the present invention also includes methods using other numbers of measurement points.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of paper that is tilted to a conventionally allowable degree, FIG. 2 is a diagram showing an example of paper that is tilted to a greater extent than that, and FIG. 3 is a diagram showing the functions of an embodiment of the apparatus according to the present invention. Figure 4 is a diagram for explaining the principle of tilt detection, Figure 5 is a block diagram of the tilt normalization circuit, Figure 6 is a block diagram of the X-axis direction position detection circuit, and Figure 7 is a diagram of the tilt detection circuit. A block diagram of the correction circuit is shown in Figure 8. Figure 9 is a waveform diagram of the main signals of the circuit in Figure 7. Figure 10 is a diagram showing the bit configuration of each register in the circuit. Figure 11 is the same. 2-bot RA used in circuit
The operating waveform diagram of M, FIG. 12, is a diagram for explaining the principle of tilt correction. Figure 11 - 441 - (^) ) (?) > ε run θ in 5

Claims (1)

[Claims] 1. In a pattern recognition device using a one-dimensional photoelectric conversion element as a photoelectric conversion part, means for detecting the relative inclination of a pattern entry sheet with respect to the -dimensional photoelectric conversion element, The present invention is characterized by comprising means for detecting the relative position of the pattern entry sheet with respect to the photoelectric conversion element, and means for correcting the relative inclination using the detected relative inclination and relative position. Pattern recognition device.