JPS5951027B2 - Character reading method - Google Patents

Description

TECHNICAL FIELD OF THE INVENTION The present invention relates to a character reading system, and particularly to a system for optically reading formalized characters such as handwritten, printed, typed, etc. characters.

Prior Art and Problems Conventional devices for reading typed characters or handwritten characters are
Most of them employ complicated scanning means using flying spot scanning tubes or use a large number of stages of registers, and are very expensive.

In addition, since pattern matching is used to identify characters, when the original character pattern is converted into an electrical signal and stored in a shift register etc., the character pattern may be slanted, vertically or horizontally If the characters are moved or are irregularly reduced or enlarged in the vertical or horizontal direction, the characters cannot be identified. requires complex preprocessing. However, even with such preprocessing, it becomes difficult to identify characters if the relative slope between the characters and the vertical scan or the horizontal scanning speed changes significantly. Therefore, a scanning device capable of precise scanning, such as a flying spot scanning tube, is required, and a paper feeding mechanism similarly requires a precise mechanism that keeps the feeding speed and traveling position constant. As a result, it becomes very expensive. OBJECTS OF THE INVENTION The present invention aims to improve the above-mentioned drawbacks, and its purpose is to improve the image quality even when there are large fluctuations in horizontal scanning speed and relative inclination between characters and vertical scanning. To make characters distinguishable and readable even when manually scanned.

Structure of the Invention First, characters printed or handwritten on a sheet of paper are scanned by a series of horizontally adjacent vertical scans (including scans tilted within a permissible range, and the same applies hereinafter); The vertical scan movement proceeds from right to left or vice versa across the character, and each vertical scan is sampled at regular time intervals.

Each time the sampled signal is scanned once in the vertical direction, the white and black pattern information in the scanned vertical column is divided into horizontal line segments and vertical line segments. The horizontal line segment is further divided into a plurality of horizontal line segment areas, such as the top, center, and bottom, and the number of appearances of each is counted, normalized by the number of vertical scans per character, and the peripheral distribution is determined.
In addition, for vertical line segments, the logical sum of the binary signals corresponding to the output of each photoelectric conversion element of vertical scanning white/black data in which vertical line segments are continuously included in adjacent vertical scanning data is calculated. After that, vertical line segments are detected, and characters are identified based on the order of occurrence of the vertical line segments and the peripheral distribution of the horizontal line segments. Horizontal line segments divided into multiple areas are each normalized by the number of scans per character to obtain the marginal distribution, and vertical line segments are scanned white/black data in which vertical line segments are consecutively included in adjacent vertical scan data. Since the detection is performed after calculating the logical sum of , there is little effect even if a black point is added or a part of a line segment is deleted due to noise.

This allows for the identification of scanned characters without the need for preprocessing, regardless of variations in horizontal scan speed, or even when the vertical scan and character are tilted relative to each other.
Scanned characters can be identified. Characters are identified based on the peripheral distribution of horizontal line segments, which is obtained by counting the number of times a horizontal line segment appears divided into multiple parts, and normalizing it by the number of vertical scans per character, and the order of occurrence of vertical line segments. , there is no need for a storage device to store a huge amount of pattern information as in the past.

Embodiment of the Invention FIGS. 1 and 2 are block diagrams of a character reading system showing an embodiment of the present invention.

In FIG. 1, 1 is a sheet of paper on which formatted characters such as 0CR-AFONT are written, and is moved, for example, in the direction of the arrow.

When light is emitted from the light source 2, the reflected light from the paper 1 enters the photoelectric conversion device 4 via an optical system such as a lens 3. The photoelectric conversion device 4 is a device in which photoelectric conversion elements such as photo diodes are arranged in the vertical direction, and the photoelectric conversion device 4 has photoelectric conversion elements such as photo diodes arranged in the vertical direction.
(arrow). The unit light receiving area of the photoelectric conversion device 4 that receives reflected light from the paper 1 will be referred to as a "basic area", and the signal obtained from this basic area will be referred to as a "basic area signal".

Further, since the light reflected from the background area of the paper 1 and the light reflected from the character area each have different amounts of light, the corresponding basic area signal levels will be different. The basic area signal obtained by the photoelectric conversion device 4 is applied to a black and white determination circuit 5, and a level determination is performed to determine whether it is white or black.

For example, the signal corresponding to the background area of paper 1, that is, white, is "o," and the signal corresponding to the character area, that is, black, is "1."
” is added to each of the following. Arithmetic circuits 8 and 9 are
The difference between the address registers 6 and 7 and the basic area number (hereinafter referred to as address) of the photoelectric conversion device 4 added from the black and white judgment circuit 5 is taken, and the difference is added to the height change generation circuit 10 and the division circuit 22, respectively. The signal obtained by the black and white judgment circuit 5 varies from "0" to "1- or '.
It is valid when it changes from 'l' to 'o''.

The height change generation circuit 10 outputs a signal SD when changing the reference position of the height of the data obtained by scanning according to the difference between the contents of the address register 6 and the address added from the arithmetic circuit 8. However, the signal SD is not output when the reference position is not changed.

When the signal SD is output, a shift down described later is performed and the contents of the address register 6 are set to the address at that time. That is, the contents of the address register 6 are set as the reference position of the height of the line segment (or the basic area number of the reference line segment). Address register 7 is 1
This indicates the basic area number where black is detected at the top end of the character, and the arithmetic circuit 9 calculates the difference between the address register 7 and the address at which data that appears to be a line segment appears, and then divides the area. A circuit 22 determines the position of the line segment in the character.

Assuming that the contents of address register 7 are MSB and the address where "°1" is detected at the top end in one vertical scan is SCCl, updating of the contents MSB of address register 7 is as follows:
When SCCl-MSB is 0, a signal is applied from the partitioning circuit 22 to the address register 7, and the current address SC
By setting Cl as the MSB, the MSB becomes the address of the upper black edge within one character.

Space flags 13 and 14 are used to detect the start and end of scanning for one character, and are used to detect “1” from the black and white determination circuit 5.
” is added, the contents of space flag 13 are set (“F”) and 1 drop! Check the contents of space flag 13 every time a direct scan is completed, and if it is 11'',
Apply the signal HSCK to the horizontal scanning counter 34, and then
Move the contents of space flag 13 to space flag 14 and clear space flag 13 (゜゜0゛2
).

When one vertical scan ends and the contents of the space flags 13 and 14 become "0" and "1", respectively, that is, when there is no black data in each basic area signal of the photoelectric conversion device 4, It is assumed that scanning of one character 5 has been completed, and a signal SRl is generated.

The black run length counter 11 counts the number of successive occurrences of "1" in the signal applied from the black and white judgment circuit 5. Given the contents of the long counter 11, the black long counter 11
Clear the contents of.

The line segment determination circuit 12 determines whether or not it is a horizontal line segment based on the content. For example, if the content added from the black run length counter 11 to the line segment determination circuit 12 is BNCl, then BNCl
When ≦a (a: constant), it is considered a horizontal line segment. When it is determined that it is a horizontal line segment, a signal HE is applied from the line segment determination circuit 12 to the dividing circuit 22 to divide the horizontal line segment into a plurality of positions, such as the top end, center, bottom end, etc.

To explain the classification according to FIG. 5, black and white judgment circuit 5
The address when the signal from ゛ changes from "゜1" to "O゛" is set as SCC2, and the contents of address register 7 are set as M
If SB, A when SCC2-MSB=A is added from the arithmetic circuit 9 to the dividing circuit 22, and when A<d, it is taken as the upper end of the line segment, and the signal HTCK is added, and when d≦A<e, the line is When e≦A<f, the lower end of the line segment is output, and the signal HBCK is output.

If it is determined that it is not a horizontal line segment, the signal E is added to the vertical line segment flag 15, the vertical line segment flag 15 is set (set to "1"), and when one vertical scan is completed, the contents of the vertical line segment flag 15 are changed. In addition to the vertical line segment flag 16, the vertical line segment flag 15 is cleared (set to "O"), and the logic of the black and white data between the contents of the shift register 19 and the contents added from the black and white determination circuit 5 is determined by the OR gate 18. The sum is taken and stored in the shift register 19.

When HE is added consecutively in multiple vertical scans, the data obtained by calculating the logical sum of all of them is transferred to shift register 1.
9 contents. If signal HE is vertical line segment flag 15
If the signal is not added to the shift register 19 during the next vertical scan, the contents of the shift register 19 are blocked by the AND gate 17, and the signal applied from the black/white determination circuit 5 is stored in the shift register 19.

When the vertical scanning is completed and the contents of the vertical line segment flags 15 and 16 are "0" and "1", respectively, it is assumed that the vertical line segment has passed, and a signal SR2, which will be described later, is generated.

The black run length counter 20 counts the number of consecutive signals “1” applied from the shift register 19.
When it changes from "1" to "0", the content of the black run length counter 20 is given to the line segment determination circuit 21, and the black run length counter 2
Clear 0.

The line segment determination circuit 21 divides the contents into a plurality of parts and applies a signal to the division circuit 22. For example, black rencho counter 20
As content BNC2, when h>BNC2≧g, SV is considered as a short vertical line segment, when i>BNC2≧h,
Each signal of LV is treated as a long vertical line segment and is divided into a dividing circuit 22.
(However, g, h, i = constants).

In the partitioning circuit 22, the contents of the shift register 19 are “1”.
” to “o”, the address is SCC2′,
If the contents of the address register 7 are MSB, then A when SCC2'-MSB=A is added from the arithmetic circuit 9, and the signals of VU2° and VL2° are output based on the signal from the line segment determination circuit 21 and A. do.

For example, as shown in FIG. 5, the value of the signal SV,A is d≦
When A<e, the upper short vertical line segment is VU2°, and when the value of the signal SV,A is e≦A<f, the lower vertical line segment is VL2≦, and the signal LV,A is The value is e≦A<
When f, VL2° and VU2° are output as long vertical line segments (d, e, f = constants). In Figure 2, 34
is a horizontal scanning counter, which counts the signal HSCK and
Detects the number of vertical scans of a character.

Signals HTCK, HMCK, output from the division circuit 22
HBCK, VU2O, VL2 horizontal line segment counters 31, 32, 33 and vertical line segment detection flag 41 (a
), added to 41(b).

Horizontal line segment counters 31, 32, 33 receive signals HTCK,
When HMCK and HBCK are added, they are added,
Find the number of occurrences of each.

When the scanning of one character is completed, the arithmetic circuit 35 outputs a signal SRI.
is added to normalize the horizontal line segment from the horizontal line segment counter 33 and horizontal scanning counter 34. If the contents of the horizontal line segment counter 33 and horizontal scanning counter 34 are respectively HBC and HSC, then the value of HBC/HSC is It is divided into several types and stored in the horizontal line segment flag 36. For example, H
When BC/HSC≦l, there is no horizontal line segment, 0, l
When <HBC/HSC≦m, as a short horizontal line segment,
1. When m<HBC/HSC≦n, the horizontal line segment is divided into half lengths; 2. When n<HBC/HSC, the horizontal line segments are classified as long.

When the signal SRI is applied, the horizontal line segment counter 31,
The contents of 32 are shifted to the horizontal line segment counters 32, 33, the contents of the horizontal line segment flags 36, 37 are shifted to the horizontal line segment flags 37, 38, this is repeated, and finally the contents of the horizontal line counters 31, 32, 33 are shifted. It is normalized by the content of the horizontal scanning counter 34, divided into a plurality of horizontal line segment flags 36, 3.
7,38.

When signals VU2° and VL2° are applied from the dividing circuit 22 to the vertical line segment detection flags 41(a) and 41(b), the vertical line segment detection flags 41(a) and 41(b) are set.

When the vertical scanning is completed and the signal HE is not added to the first vertical line segment flag 15 and the contents of the vertical line segment flags 15 and 16 become '0-'1', the scanning of the vertical line segment ends. A signal SR2 is generated to indicate that the process has been completed. When SR2 occurs, the vertical line segment flags 41(a), 41(b) are changed to the vertical line segment detection flags 42(a), 42(b), 42(a),
The contents of 42(b) were changed to 43(a) and 43(b), and 43(
The contents of a) and 43(b) are shifted to 44(a) and 44(b), and the vertical line segment flags 41(a) and 41(b) are cleared. On the other hand, if the signal SR2 is not generated, it is assumed that scanning of the vertical line segment is still continuing, the contents of the vertical line segment flags 41(a) and 41(b) are cleared, and the vertical line segment is detected. The contents of the vertical line segment detection flags 41(a) and 41(b) are made valid only when the vertical line segment detection flags 41(a) and 41(b) are completed. Since the line segments are divided based on the content of the address register 7, that is, the position of the topmost black of one character, there is no inclination relative to the photoelectric conversion device 4 and the character to be read. Even if a line segment at the bottom of the character is detected during scanning, it is treated as a line segment at the top of the character, so what was treated as the top horizontal line segment is changed to the center horizontal line segment. What was treated as a horizontal line segment becomes the bottom horizontal line segment,
It is also necessary to shift what was treated as the upper vertical line segment to the lower vertical line segment (hereinafter referred to as "shifting down").

The shift down is performed by the signal SD from the height change generation circuit 10.
is output, the horizontal line segment counters 31 and 32 are set to the horizontal line segment counters 32 and 33, and the vertical line segment detection flags 41(a), 42(a), 43(a), and 44(a) are set to The contents are vertical line segment detection flags 41(b), 42(b),
Shift to 43(b) and 44(b). The signal SD is output as follows.

The arithmetic circuit 8 calculates the difference between the content TAR of the address register 6 and the address error SCCl when the black and white data of the black and white judgment circuit 5 changes from "0" to "1", and when the difference is within the range, the height A change generation circuit 10 generates a signal SD. For example, if TAR-SCCl=A, when A<b, SD is not output, when A≧b, SD is output, and when A≧C, SD is output again (B.C= constant).

When the signal SD is output, the address SCCl is set as the content TAR of the address register 6.

A specific example is shown in FIG.

Assuming that the contents of address registers 6 and 7 are TAR and MSB, in the scan at position t1-1, the data in basic area 6 is regarded as the upper horizontal line segment and is counted in the horizontal line segment counter 31; When scanned at the position of tl,
It becomes necessary to move the data in the basic area 61 to the data at the lower end.

Let SCCl be the address where the uppermost black color occurs at the position t1, the signal SD is generated twice from A≧C with TAR-SCCl=A, and the contents of the horizontal line counter 31 are shifted to the horizontal line counter 33, Hereinafter, SCCl will be referred to as TAR. When "1" is not added to the space flag 13 and the content of the space flag 14 is "1", it is assumed that the scanning of one character is completed, and when the signal CKl is added to the storage device 52, the horizontal line segment flag is 36, 37, 38, and vertical line segment detection flags 42(a), 42 (b included 43(a),
Signals HT, HM, HB and VU2l, VL2l, VU22 from 43(b), 44(a), 44(b), respectively.
, VL22, VU23, and VL23 are added to the read-only storage device 51, the read contents D, are written to the storage device 52, and DOut is output.

The contents of DOut correspond to the character identification results, and the characters are identified based on the contents. Storage device 5
2, when the signal CKl is applied, first initializes it with the signal CK2 and then writes D.

Output ut. After identifying l characters, each counter and flag are cleared and set to the initial state.

FIG. 4 shows an example of the identification method according to the present invention.

While moving the paper with the letter "1" of 0CA-AFONTSIZEI written on it from left to right, 0.3 mm x 0.3
Scanning is performed with a photoelectric conversion element array having a basic area of size m, and the data of white ("・" in Fig. 4) and black ("3゛" in Fig. 4) are sequentially 62 (1) to (n). can get.

Vertical scanning is scanning in the direction of arrow 63, where 64 corresponds to the upper part of the character and 65 corresponds to the lower part of the character. At that time, horizontal scanning counter 34, horizontal line segment counters 31, 32, 3
3 contents as HSC, HTC, HMC, HBC, horizontal line segment flags 36, 37, 38, and vertical line segment detection flag 41.
(A, b), 42 (A, b), 43 (A, b), 44 (
A, b), and D output from the storage device 52. The contents of ut are HT, HM, HB, V2O, V2l, V22 respectively.
, V23, DOut. The numbers are HSC, HTC,
The count numbers of HMC and HBC are HT, HM, HB, and V.
2O, V2l, V22, V23 are 4 of 0, 1, 2, 3
It is divided into species and shown. V2l is VU2l and VL2
1 when only VL2l is set, 2 when only VU2l is set, and 2 when only VU2l is set.
When both V2l and VL2l are set, it is 3, and when both are not set, it is 0.
The same applies to V22 and V23.

The content of the black leader counter 11 BNCl is 1≦BNCl≦2
When , the line segment determination circuit 12 outputs the signal HE, and the BNC
It is assumed that the signal HE is output when l=0 and BNC≧3. In 61(1), space flag 13 is set to “O”.
is set, and no characters have appeared yet.

In the vertical scanning of 62(8), a black character line appears, the space flag 13 is set to ``1'', and scanning of one character is started.

At this time, the topmost address y1 is stored in address register 6.
,7. Since the horizontal scanning counter 34 is incremented by 1, HSC=1. Since BNCl=3, the signal HE is output and the vertical line segment flag 15 is set. 6
From 2(9) to 62(11), HE is similarly output, and the black logical sum for each cell is calculated.

Furthermore, HSC is added one by one. Next, at 62 (12), signal HE is output and set to V2l as a vertical line segment. At this time, the content BN of the black rencho counter 20
C2 is 4 and outputs SV as a short vertical line segment. The MSB at this point is Y, and since this vertical line segment is separated from the short line segment above, it is set to VU2'', and 2 is shown in V2l in FIG. 4. Also, 62 (12
), the signal HE is output to indicate the existence of a horizontal line segment, the position of this line segment is SCC2-MSB=Y, the center of the line segment is output, the signal HMCK is output, and the horizontal line segment counter 32 is added. In FIG. 4, HMC=1 is shown. 62 (13), the topmost address is Y.

and TAR-SCCI=Y,-Y,=≧b(b=3
SD is generated and a downshift is performed.

This moves the contents of VU2'' to VL2'' and the contents of HMC to HBC, so V2l=1, HBC=1
becomes. Furthermore, since horizontal line segments at the top and bottom edges appear, 1 is added to HTC and HBC, and HTC=
1, HBC=2.

At 62 (19), the next vertical line segment is detected and the contents of V2l are set to V22 and the contents of V2O are set to V2l.

The vertical line segment detected in 62 (19) is a long line segment, and V
Since it is set to both U2'' and VL2'', V2l is 3.
Therefore, FIG. 4 shows V2l=3 and V22=1.

Since no character line segment is detected in the vertical scan at 62 (25), the space flag 13 is set to ``0'',
Space fladas 13 and 14 become “0” and “1”,
This means that scanning of one character has been completed.

At this time, HSC=17, HTC=7, HMC=0
, HBC = 9, HSC, HTC, HMC
, HBC are normalized, and the normal values HT, HM, and HB are as follows: HT = 2 (half-long horizontal line segment) HM = 0 (no horizontal line segment) HB = 3 (long horizontal line segment).

Also, Taru. The straight line segments are V2l=3, V22=eng, and V232o, and the result of identifying characters from HT, HM, HB, V2l, V22, and V23 is D in FIG.

The number 1 is shown in ut. As another example, if a sheet of paper with the number 2 written on it is scanned moving from right to left (not shown), HT=3, HM=3, HB=3, V2l=1
, V22=2, V23=0 are obtained, and the identification result is number 2.
becomes.

In the embodiments of the present invention, each constant is determined by the size of the target character, the basic area of the photoelectric conversion device, the number of vertical line segments, horizontal line segments, and the like.

In addition, in the example, the basic area is divided by photoelectric conversion element rows, and the case is shown in which they are arranged in one row in the vertical direction, but it is also possible to use multiple rows, a two-dimensional arrangement, or It can also be configured in combination with optical fibers.

Of course, it is also possible to scan using a flying point scanning tube or a vidicon. Further, the target characters can be read and identified if they are formalized, such as numbers, symbols, or alphabetic characters.

Effects of the Invention I As explained above, the present invention divides the '0' and '1' information of the basic area signal into horizontal and vertical line segments for each vertical scan, and the horizontal line segment is further divided into a plurality of vertical line segments. The method divides the characters into regions, counts the number of times they appear, normalizes it by the number of vertical scans per character to find the marginal distribution, and identifies characters based on the marginal distribution and the order of occurrence of vertical line segments.
There is no need to memorize character patterns as electrical signals,
Therefore, since a large-scale storage device is not required, an inexpensive device can be provided.

In detecting a horizontal line segment, normalization is performed by determining the ratio between the number of times a horizontal line segment appears and the number of vertical scans of one character, that is, the number of horizontal scans.

Whether the horizontal movement speed of the paper is fast or slow, the ratio of the number of times a horizontal line segment appears, the number of square scans per character, and the number of horizontal scans per character are almost the same, so the photoelectric conversion element There is less influence from fluctuations in the horizontal relative speed between the column and the paper. Furthermore, if the photoelectric conversion element array and the characters are relatively inclined, it is difficult to detect vertical line segments, but the present invention
In this method, a vertical line segment is detected after calculating the logical sum of adjacent vertical scanning data that includes continuous vertical line segments, so there is little influence from the slope.

In manual scanning reading in which a photoelectric conversion element array is manually held and moved over characters, the horizontal scanning speed may be large or small, and the photoelectric conversion element array may be tilted against the paper, but in the present invention, these effects are small due to the above. Therefore, manual scanning and reading is possible.

[Brief explanation of the drawing]

Figures 1 and 2 are graphical diagrams of a character reading device showing one embodiment of the present invention, Figure 3 is an example of data obtained by vertical scanning for the number "3", and Figure 4 is a bookmark diagram of a character reading device according to an embodiment of the invention. FIG. 5, which is an example of the identification method according to the invention, is an explanatory diagram of division of line segments. 1: paper, 2: light source, 3: optical system, 4: photoelectric conversion device,
5: Black and white judgment circuit, 6, 7: Address register, 8,
9: Arithmetic circuit, 10: Height change generation circuit, 11: Kuro Rencho Count Hisa 12: Line segment judgment circuit, 13, 14: Space flag, 15, 16: Vertical line segment flag,] 7 Reand gate, 18: OR gate, 19: Shift register, 20: Black run length counter, 21: Line segment judgment circuit, 22
: Division circuit, 31, 32, 33: Horizontal line segment counter, 3
4: Horizontal scanning counter, 35: Arithmetic circuit, 36, 37,
38: Horizontal line segment flag, 41(a), (b included 42(
a), (b), 43(a), (b), 44(a), (b
): Vertical line segment detection flag, 51: Read-only storage device,
52: Storage device, 63: Scanning direction of vertical scanning, 64
: Above the photoelectric conversion element row, 65: Below the photoelectric conversion element row, 66: Identification result.

Claims (1)

[Scope of Claims] 1. In a character reading method in which a formatted character is scanned by a line of vertical scanning and the scanned characters, etc. are identified, binary data for each vertical scanning is divided into horizontal lines or vertical lines. means for dividing the horizontal line segment into line segments; means for dividing the horizontal line segment into a plurality of regions and counting the number of appearances for each section; means for dividing the vertical line segment into a plurality of regions; and counting the number of vertical scans of one character. means for normalizing the count value for each division by the number of vertical scans of one character, and character identification means, dividing into horizontal line segments or vertical line segments for each vertical scan,
For horizontal line segments, divide them into multiple areas, count the number of times the line segment appears in each divided area, normalize the counted value by the number of vertical scans of one character, calculate the marginal distribution, and calculate the vertical line segment. A character reading method characterized in that characters are divided into a plurality of regions, the order of appearance is determined, and characters are identified from the peripheral distribution of horizontal line segments and the order of appearance of vertical line segments. 2. In the character reading method recited in claim 1, when vertical line segments are included in adjacent vertical scanning data consecutively, each photoelectric conversion element of the vertical scanning data is A character reading method characterized by detecting a vertical line segment after calculating the logical sum of binary codes corresponding to . 3. In the character reading method described in claim 1, the basic area number that serves as a reference for a line segment in the height direction and the basic area number at the lower end or upper end of the character area obtained by each vertical scan. A character reading method characterized by changing the basic area number and changing the classification of horizontal line segment data and vertical line segment data detected in the past when the difference is greater than a certain value.