JPS6136680B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a positioning method for determining positions for separating a plurality of characters into individual characters in character recognition.

In character recognition, a medium on which characters are printed is scanned to obtain a video signal, which is temporarily stored in a memory, and then characters are cut out one by one and character recognition is performed according to various logics. To cut out characters, assuming that line-by-line separation has been completed, it is sufficient to insert a break within the line to separate each character, and the present invention relates to a method for determining the position to insert this break.

When setting the reading area, the host device provides the reading start point (field start point FSP) address, reading end point (field end point FEP) address, and the number of characters to be read between these two points. FSP and FEP are usually slightly different from the start and end points on the image. If they match, FSP and FEP are given by the number of bits from the base point (this corresponds to the memory cell), so the character spacing, or pitch P, is determined as (FFP - FSP)/n (where n
is the number of characters), so from FSP or FEP
You can separate each character by separating them with . but
If FSP and FEP are misaligned, this method may end up separating each character with a dividing line.

FIG. 1 is a diagram for explaining such a character extraction method. In the figure, some characters in the line may be missing, but if the number of characters between FEP and FSP is Cm, the number of characters input here is the maximum number that can fit within the interval (FEP - FSP), Accordingly
Cm=(FEP−FSP)/P. Printed by machines such as line printers and typewriters, the pitch is accurate, so P ₁ in Figure 1,
P ₂ and P ₃ are all equal. Figure 2 shows the deviation between FSP etc. and the image, 1 is the medium on which the characters are marked, 2 to 4 are the images, 2 is the one where E is 0, 3 and 4 are the E
>0. When E becomes large and the characters are packed together and the play becomes small, the cutting line l will divide the characters.

The present invention attempts to perform accurate character segmentation by correcting FSP and FEP using actual measurement results, and its feature is that each character in a line is extracted from character image information temporarily stored in the memory of a character recognition device. In the positioning method for cutting out the line, the character pitch is calculated from the given reading start point information, reading end point information, and the maximum number of printed characters included between these end points, and the 2 Each bit of value information is added in the line width direction, the obtained numerical group in the line direction is separated from the end point by the character pitch to obtain a plurality of numerical blocks, and each numerical value in each numerical block is added to that block. The bit numbers with the same inner bit number are accumulated, the bit number with the smallest cumulative value is found, the end point information is corrected with this bit number, the true end point is found, and the character pitch interval position from the true end point is calculated. The purpose is to set the cutting position for cutting out each character. This will be explained in detail below using examples.

First, the positioning method of the present invention will be explained with reference to FIG. In this figure, 10 is a memory, into which a black and white binary image signal obtained by scanning characters on a medium is written.
In the drawing, this memory state is shown as a real image. The small rectangular frame 12 is one memory cell.
Assuming that FSP and FEP are given as shown in the figure (in this example, they are slightly shifted to the right), taking a margin and setting a point FSP' outside P/2 from them as a temporary starting point, we can calculate the vertical projection from there. Find (one extra character). The result is sequence 14. In other words, there are no characters around FSP' in this example, so 0 (white) continues, and black is 1 at the 5th position.
A bit appears, then 6 black bits appear, and then 0 continues for a while. This numerical sequence is divided into blocks of pitches P (in this example, P=6), and the numerical values of the numerical blocks are accumulated among those having the same bit number within the block.
Here, the bit number within a block refers to the order of each numerical value counted from the start of the block. The result is as follows.

000016 000033 400032 500033 600330 330000 1830031314 Of course there are no characters between characters, so it should be 0. In a character, if the characters are consecutive, 0 will not occur, but the kana "ha",
In separated characters such as "ri", 0 appears within the character. However, the width in which 0 appears is narrower than the character spacing, and the probability that all characters in a line are separate characters is small, so when added up, they tend to be erased. In this example as well, the rightmost character is a separate character, and although 0 appears, it is canceled out by other characters, and in total, 0 appears only between characters (third from the left in the above cumulative value). Thus, it can be seen that the character spacing, that is, the separation point is a point marked with a .lambda., which is shifted by 2 bits to the right from the point separated by the pitch P from the tentative starting point FEP'.

Note that the large cumulative value indicates the central part of the character. Further, in order to obtain such character centers and character spacings, it is not necessary to total up all the P number groups, but it is sufficient to total up a suitable number, for example, 5 to 6 of them. Next, a circuit for performing such positioning is shown in FIG.

In Figure 4, 20 is the subtracter, starting point FSP, ending point
Inputs FEP and outputs the difference. 22 is a divider which inputs the output of the subtracter 20 and the number of characters Cm, performs division, and outputs pitch P. 24 is a register that stores this. 26 is a storage device into which image information as shown in FIG. 3 is input.
28 is a column (vertical direction) counter, and 30 is a row (horizontal direction) counter. Counter 28 is a clock
Count the CLL, the bit corresponding to the width of the line, the third
In the example shown in the figure, it is an octal counter that returns to the original value after counting 8 bits. The counter 30 receives the divided-by-8 output of the counter 28 and specifies the address of the storage device 26. In this way, in the storage device 26, bits equivalent to the width of a row (in this example, 8 bits) are read out once, converted into a serial signal by the parallel/serial converter 32 to which the clock CLL is input, and then sent to the counter through the AND gate 34.
Applied to the enable terminals of C ₁ to Cn. The output of row counter 30 is also applied to comparators 36,38. FSP-1/2P becomes the other input to the comparator 36, and when both inputs match, an output (this corresponds to the temporary starting point FSP') is generated, and the flip-flop 40
Set. Therefore, from this point on, ANDGATE 3
4 is opened, and the output of the converter 32 is sent to the counters _C1 to Cn.
join. Comparator 38 has FEP'=FEP+1/2P as its other input and produces an output when both inputs match, resetting flip-flop 40. Therefore, the output of the converter 32 is added to the counters C ₁ to Cn until the output of the comparator 38 (which corresponds to a provisional end point extended with a margin) is received. Note 4
2 is a divider which receives pitch P and outputs P/2. 44 is a subtracter, and 46 is an adder. 48 is a pitch counter which counts the clock CLH generated in each column (bit) created by the AND gate 54, and starts counting by receiving a temporary starting point pulse obtained from the comparator 36 through the OR gate 52; The count value is input to the decoder 50. This count value is also input to the comparator 56, and when it becomes equal to the pitch P, the comparator 56 produces a coincidence output EQ.
This is input to the counter 48 through the OR gate 52, causing the counter to be cleared and restarted. As a result, the count value of the counter 48 is 1, 2, 3...
P, 1, 2, . . . are repeated, and the decoder 50 receives this and sets the counters C ₁ , C ₂ , C ₃ . . . Cn, C ₁ , C ₂
. . . are selected in sequence (where n=P).

Counters C ₁ -Cn are also controlled by the output of parallel to serial converter 32 and only count clock CLL when converter 12 is a "1" (black) signal. Therefore, when counters C ₁ to Cn are selected once by the decoder 50, the count value of each counter becomes the third
In the example shown in the figure, it becomes a number group. The second time, it becomes a number group, and since it is added to the previous count value, the result is +. The same goes for the following, and thus the counter
C ₁ to Cn are the aforementioned cumulative values 18, 3, 0, 3, 13, 14
become. The arithmetic unit 60 takes in the counted values of the counters C ₁ -Cn through the multiplexer 58 and outputs the smallest bit number. In this example, this is "2" as described above. This is the adders 62 and 64.
It is added to FSP'FEP'' and becomes the true starting point TESP and the true ending point TFEP. Note that FEP''=FEP-P/2. If a pitch P is used to separate the characters from this true start or end point, that becomes a separation point, and thus each character can be correctly cut out.

As described above in detail, according to the present invention, each character in a line can be accurately cut out, and the character recognition rate can be improved.

[Brief explanation of the drawing]

1 and 2 are explanatory diagrams for specifying a character reading range, FIG. 3 is an explanatory diagram for the positioning system of the present invention, and FIG. 4 is a block diagram showing an embodiment of the present invention. In the drawing, 26 is the memory, FSP is the reading start point,
FEP is the reading end point, Cm is the maximum number of characters, P is the character pitch, and TESP and TFEP are the true reading start and end points.

Claims (1)

[Claims] 1. In a positioning method for cutting out each character in a line from character image information temporarily stored in the memory of a character recognition device, given reading start point information, reading end point information, and maximum printing included between these end points The character pitch is calculated from the number of characters, and each pitch of the binary information representing each character in the line is added in the line width direction, and the obtained numerical value group in the line direction is divided from the end point by the character pitch to form a plurality of values. Obtain numeric blocks, add up each numeric value of each numeric block among those with the same bit number within the block, find the bit number that minimizes the cumulative value, and correct the end point information using the bit number. A positioning method characterized in that a true end point is determined, and a position at the character pitch interval from the true end point is set as a cutting position for cutting out each character.