JPH0632076B2 - Simple handwritten character recognition device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device, and more particularly to a simple handwritten character recognition device suitable for a device such as a facsimile which reads image information in a line-sequential manner.

[Conventional technology]

A conventional character recognition device of this type is disclosed in Japanese Patent Laid-Open No. 59-1230.
As described in Japanese Patent No. 0, the input image data is processed bit by bit. A conceptual configuration of such a bit processing system is shown in FIG.

In the system shown in FIG. 23, 1-byte (Byte) data is transferred from the image data 201 to the register 202, the data in the register 202 is shifted by 1 bit, and the carry flag 203 is referred to determine whether the pixel is a black pixel. , The number of black pixels of the required block in the character frame is determined from the output of the count block designating unit 204 for designating the range for counting the number of black pixels and the address of the data being processed. A recognition table 206 indicating whether or not there is a write in each block measured by the in-block black pixel number measuring unit 205 is created, and what character is used depending on which block has a write and which block has no write. It was performing recognition processing. However, this bit processing method has a slow processing speed because of the large amount of data to be processed, and no consideration has been given to this point. For example, Japanese Laid-Open Patent Publication No. 58-22476 discloses a technique for dividing a number entry frame into seven segments and detecting the entry line segment length in each segment to recognize a character. For the detection of, dot-by-dot processing was performed.

[Object of the Invention]

An object of the present invention is to accurately judge a handwritten character,
An object of the present invention is to provide a simple character recognition device capable of processing this at high speed.

[Outline of Invention]

The outline of the present invention is that when converted to run length data, the amount of data can be reduced to about 1/10 of that of the conventional one, and that addition / subtraction of this run length data can perform the same processing as bit processing. Focusing attention, the run length data and the range to be measured are compared to measure the number of black pixels, and the number of black pixels can be measured at high speed.

Example of Invention

 An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is an explanatory diagram showing an example of an input sheet used in the present invention. 302a and 302b are inclination detection markers, 30
5a to 305t are character entry frames, and 304a to 304t are bands above the character entry frame. The band above the character entry frame forms a character cutting marker. Tilt detection marker 302a
, 302b and bands 304a to 304t are printed in colors detected by the facsimile reading unit.

FIG. 2 shows bands 304a-304t and character frames 305a-3.
It is a detailed view of 05t. 305 is a character frame, 401A and 401B are write-protected areas in the character frame 305, and
2 is a character entry section. The character frame 305 and the write-protected areas 401A and 401B are printed in a color dropout color that is not detected by the facsimile reading unit.

The present embodiment relates to a device for recognizing a character (only one character is written in an 8-character frame) written in the character writing unit 402. In character recognition, the character entry unit 402 is divided into a plurality of blocks, for example, 150 as shown in FIG.
Blocks 1 to 1515 are detected, and blocks in which writing is performed are detected. What a are prepared recognition table defining whether the character if there is writing in advance which blocks and in which block, recognition table is compared with the detected write block, something that whether the character is determined It Conventionally, the written characters are read as image data consisting of white pixels and black pixels, the number of black pixels in each block is counted, and the number of black pixels for each block is used to enter the block. It was determined whether or not to write. As described above, according to the present invention, when counting the number of black pixels in each block, the read image data is converted into run length data, and the run length data is used for each block. The number of black primes of is calculated.

FIG. 3 is a diagram showing a schematic configuration of the present invention. Reference numeral 501 is a marker detection unit, 502 is a band detection unit, and 503 is a black pixel number measurement unit.

The operation of the above embodiment will be described below.

The image data input by facsimile is transferred line by line to the run length data creation unit 102.

The run length data creation unit 10 processes the data for one line bit by bit, and creates run length data indicating how many bits each of black pixels and white pixels are continuous.

The marker detection unit 501 detects the tilt detection marker 302 with reference to the run length data, and calculates the average bit shift on the main scanning side with respect to one sub-scanning line. As shown in FIG. 8 (a), the average bit shift on the main scanning side means that when the input data is inclined, the position of the intersection of the scanning line in the main scanning direction and the block boundary line is It shows how many bits are shifted per line when changing to the sub-scanning direction.

The band detection unit 502 refers to the position of the tilt detection marker 302 detected by the marker detection unit 501 and the run length data, detects the band 304 on each character frame 305, and calculates the center point of the band 304. As a result, the count area and the count line which are the reference of each character frame 305 are determined. The count area is an area for counting black pixels, and the count line is a reference line for counting black pixels.

The black pixel measuring unit 503 uses the basic count area determined by the band detection unit 502 and the start address of the count area of each line due to the average bit shift on the main scanning side with respect to one sub-scanning line calculated by the marker detection unit. The end address is determined, the start address of the count area and the start address of the black run length, the start address and the end address of the black run length in the count area are determined by the end address, and the number of black pixels in the area is measured The number of black pixels in each block in each character frame 305 by the count line is measured.

The recognizing unit 109 determines the presence / absence of writing in each block based on the number of black pixels in each block measured by the black pixel measuring unit 503, and outputs a character code according to the combination.

The details of each element of the embodiment shown in FIG. 3 will be described below.

FIG. 4 is a conceptual diagram of the run length data creation unit 102. 6
Reference numeral 01 is a raw data table, 602 is a register, 603 is a carry flag, 604 is a counter, 605 is a change point detection unit, and 606 is a run length data table. To create run length data from image data, first, 1-byte data from the data for one line stored in the raw data table 601 is transferred to the register 602. The register 602 shifts the data by 1 bit and refers to the carry flag 603 to determine whether the shifted 1-bit data is a black pixel or a white pixel depending on whether it is "1" or "0". Change point detection unit 6
Transfer to 05. The change point detection unit 605 stores whether the data of the previous bit is a white pixel or a black pixel, and if it does not match the data of the current bit, the value counted as the change point by the counter 604 is used as the run length. As data,
It is stored in the run length data table 606. In addition, run length data of white pixels is stored in the first row of the run length data table 606, and subsequently, black pixels and white pixels are sequentially stored in this order.

The detailed operation of the circuit will be described with reference to the flowchart shown in FIG.

FIG. 5 is a flowchart of the operation of the run length data creation unit 102. In FIG. 5, 701 is an initial value setting step, 702 is a 1-byte data transfer command step, 703.
Is a bit shift command step, 704 is a carry flag reference step, 705 and 706 are change point detection steps, 707 and 708, 709, 710 are counter instruction steps, 711 and 712 are counting steps, 713 is a shift end determination step, 714. Is a judgment step for judging whether or not the processing of the data for one line is completed.

In the initial value setting step 701, the counter 604, the run length data table 606, and the change point detection unit 605 are initialized. Since it is created from the run length of the white pixel as described above, “0” is stored as the initial value in the change point detection unit 605, and 0 is set in the counter 604.
The run length data table 606 is initialized. Then 1
By the Byte data transfer command step 702, 1 Byte of raw data (image data) in the raw data table 601 is
It is transferred to the register 602, and bit shift is performed in the bit shift command step 703. In the carry flag reference step 704, it is determined whether the data in the carry flag is “0” or “1”. If the obtained result is “0” indicating white, the process proceeds to step 705.
If it is “1” indicating black, the process proceeds to step 706. In step 705, the value stored in the change point detection unit 605 is compared with "0" to determine whether it is the first white, that is, whether the value stored in the change point detection unit 605 is "1" or "0". To be judged. If it is the first white (that is, if the value currently stored in the change point detection unit 605 is “1”), the process proceeds to step 707, and the current value in the counter 604 is stored in the run length data table 606. Next step 70
Go to 8. In step 708, the counter 604 is set to "1".
Is set and "0" indicating white is set in the change point detection unit 605, and the process proceeds to step 713. If it is not the first white (that is, if the value currently stored in the change point detection unit 605 is “0”), the process proceeds to step 711, 1 is added to the value of the counter 604, and then the process proceeds to step 713. move on.

In step 706, the value stored in the change point detection unit 605 is compared with "1" to determine whether it is the first black, that is, whether the value stored in the change point detection unit 605 is "1" or "0". Is judged. When it is the first black (that is, when the value currently stored in the change point detection unit 605 is “0”), the process proceeds to step 709, and the current counter 604.
Is stored in the run length data table 606 and the process proceeds to the next step 710. Counter 6 in step 710
"1" is set in 04 and the change point detection unit 60
“1” indicating black is set in 5, and the process proceeds to step 713. If it is not the first black (that is, if the value currently stored in the change point detection unit 605 is “1”), the process proceeds to step 712, 1 is added to the value of the counter 604, and then the process proceeds to step 713. move on.

In the shift end judgment step 713, it is judged whether or not the data in the register 602 has been shifted eight times, and if the shift has not ended, the procedure returns to step 703 and the above procedure is repeated. If eight shifts have been completed, the routine proceeds to step 714, where it is judged whether the processing of the data for one line has been completed. If the processing of the data for one line is not completed, the next 1-byte data is transferred to the register 602, and the above procedure is repeated until the data processing for one line is completed.

FIG. 6 is a block diagram showing an example of an LSI for performing the processing of the run length data creation unit 102 at high speed.

As for the outline of this LSI, the change point detection unit and the image signal restoration unit that require high-speed bit processing are made into hardware, and the reference and calculation of the code table are made flexible and faster by a microprogram. For details of this LSI, see Japanese Patent Application Nos. 57-174623 and 58-10.
No. 71.

Next, the tilt detection method performed by the marker detection unit 501 in FIG. 3 will be described.

FIG. 7 is a diagram showing the concept of tilt correction.

FIG. 7 (a) shows input data. When such data is input, if the inclination is not corrected, the characters actually cut out are as shown in FIG. 7 (b), and the blocks to be referenced (divided into 15 blocks in the drawing). Will be misaligned. Therefore, if only the sub-scanning direction is corrected,
As shown in FIG. 7C, there is a block shift in the main scanning direction. Further, if only the main scanning direction is corrected, the result is as shown in FIG. 7D. However, although there is a shift in the sub scanning direction, the number of black pixels in the reference block does not change so much. Therefore, if the main scanning direction and the sub scanning direction are corrected, it becomes appropriate as shown in FIG. 7 (e). From this, it is understood that it is preferable to correct the main scanning direction and the sub scanning direction. Only the main scanning direction is corrected based on cost, program capacity, processing speed, etc.

Next, marker detection will be described with reference to FIG.

When the image data is read in a state where the horizontal line of the input paper is parallel to the main scanning direction, the vertical line of the character entry frame 305 is vertical to the main scanning direction, and The position of the vertical boundary line of the character entry frame 305, which is divided into blocks for character recognition described with reference to FIG. 13, does not change even if the scanning line advances in the sub-scanning direction. However, as shown in FIG. 8, when the horizontal line of the input paper is not parallel to the main scanning direction, the position of the boundary line moves left or right as the scanning line advances in the sub scanning direction. It will be. In the case of FIG. 8, the position of the boundary line moves to the left as the scanning line advances in the sub-scanning direction. Left and right inclination detection markers 30 on the input sheet 301
The average bit shift ΔX on the main scanning side per line from the difference ΔY in the number of lines 2a and 302b that appear and the difference L in the start addresses of the left and right tilt detection markers 302a and 302b is ΔX = ΔY / L.

The operation of the marker detector 501 will be described with reference to the flowchart of FIG.

First, the name of each step will be described. 1101 is a line number counting step, 1102 is a marker width determining step, 1103 and 1109 are position determining steps, and 110.
4 and 1110 are initial marker determination steps, 1105 and 1112 are marker start detection steps, 1106 and 1111 are marker length counting steps, 1107 and 1113 are marker length determination steps, 1108 and 11
Reference numerals 14 and 1115 are marker determination steps. The marker length here is the number of lines counted in the sub-scanning direction, and the marker width is the dimension of the marker in the main scanning direction.

In the marker width determination step 1102, the data in the run length data table 606 for one line is referred to, the black run length in the data is read to determine whether or not there is a marker width,
If there is, it is determined in position determination steps 1103 and 1109 whether the marker is the right marker, the left marker, or noise. If neither the right marker nor the left marker on the sheet is determined to be noise, the process returns to step 1101.

If it is the left tilt detection marker 302a, it is judged in the first marker judgment step 1104 whether it is the first marker width. If it is the first marker, in the marker start detection step 1105 the start address in the main scanning direction and the sub-scanning side. A table of the number of appearance lines is created, the marker length (the number of appearance lines) is set to 1, and the process proceeds to step 1107. If it is not the first marker width, the marker length is counted in the marker length counting step 1106 (marker length = the number of appearance lines is incremented by 1), and the process proceeds to step 1107. In step 1107, it is judged whether or not the marker length data has a specified value, and if it has the specified value, the process proceeds to the marker judgment step 1108, and the marker on the left side is made to exist. If the marker length data does not have the specified value, the process returns to step 1101 to repeat the procedure so far.

When it is determined in step 1109 that the marker is the right marker, the same procedure is repeated in steps 1110 to 1114. If it is determined in step 1114 that there is a right-side marker, the process proceeds to step 1115, and it is determined whether or not there are both left and right markers. When both markers are present, the average bit shift ΔX on the main scanning side per line
A table of the start address and sub-scanning side appearance line number of the left marker in the main scanning direction created in step 1105, and the start address and sub-scanning line appearance line in the main scanning direction of the right marker created in step 1112 It is calculated by referring to a table of numbers. The difference between the start address of the left marker in the main scanning direction and the start address of the right marker in the main scanning direction is L, and the difference between the number of appearance lines of the left marker on the sub-scanning side and the number of appearance lines of the right marker on the sub-scanning side is As ΔY, the average bit shift ΔX is obtained using the formula described above. This calculation of ΔX is executed in step 1301 described later.

Next, the operation of the band detector will be described with reference to the flowchart shown in FIG. First, the name of each step will be described.
2 is an initial belt determination step, 1203 is an initial value setting step, 1204 and 1205 are belt start determination steps,
1206 and 1207 are band end determination steps, 120
8 is a band width counting step, 1209 is a band width determining step, 1211 is a band center calculating step, and 1210 is a line counting step. The band length referred to here is the length of the band in the main scanning direction, and the band width is the number of lines in the band portion counted in the sub scanning direction.

The band length determination step 1201 is the run length data table 6
The data of No. 06 is read, and the presence or absence of the band length is determined. If the read run length data has a black run length corresponding to the band length, which band length above the character entry frames 305a to 305t is the band length depending on the start address of the left tilt detection marker 302a. The first belt judgment step 120
Judge whether it is the first obi according to 2. If it is the first band, the process proceeds to an initial value setting step 1203, where the start address of the black run length is set to the start address of the band, and the end address of the black run length is set to the end address of the band. 4 is added to the number of lines of the black run length to create a table of the number of central lines of the sub-scan band, and the band width is set to 1. The band width here is the number of appearance lines of the band on the sub-scanning side.

If it is determined in step 1202 that the band is not the first band, the band start determination step 1204 compares the value of the band start address table with the value of the black run length, and the smaller value is stored in the start address table. Also,
Similarly, in the band end determination steps 1206 and 1207, a large value is stored in the end address table. In other words, the start address on the leftmost side is the start address of the belt and the end address on the farthest right is the end address of the belt among the several black run lengths indicating the belt. Detect the length (range).

In the band width counting step 1208, the band width is counted (the number of appearance lines is incremented by 1). In the band width determination step 1209, it is judged whether or not the band width has a prescribed value based on the value counted in the band width counting step 1208, and if there is (that is, it is considered that the check of the line corresponding to the band is completed). (If performed), the center of the band on the main scanning side is calculated from the band start address and end address tables in the band center calculation step 1211, and if not, step 1210 is executed. Step 121
At 0, the line count is incremented by 1, and this line count indicates the number of lanes of the data read from the run length data table 606. When the line count is incremented by 1 in step 1210, the process returns to step 1201 again, the data in the run length data table 606 is read, and the above procedure is repeated.

Next, the operation of tilt correction will be described with reference to the flowchart of FIG. This inclination correction is performed by the marker detection unit 5
The position of the boundary line of each block of the character entry frame 402 is corrected using the average bit shift ΔX calculated in 01 to prevent the inclination of the input paper from affecting the number of black pixels in each block.

First, the name of each step will be described. 1301 is an average bit shift calculating step, 1302 is a line number counting step, 1303 is a line feed judging step, 1304.
Is a bit shift calculation step, 1305 is a bit shift determination step, 1306 and 1307 are correction steps, 13
08 and 1309 are post-processing steps, 1310 is a black pixel measurement step, and 1311 is a character box end determination step.

This operation will be described. The average bit shift ΔX is calculated in the average bit shift calculation step 1301, and the number of lines currently scanned is counted in the line number counting step 1302. Line feed judgment step 1303
After the band 304 is detected, the line is sent to the character entry section 402, so that judgment is made (the number of lines to be sent). The bit deviation amount A of the line currently scanned in the bit deviation calculation step 1304 is the ΔX and the step 1302.
It is calculated based on the number of lines counted in. Also,
In the bit shift determination step 1305, if the value A calculated in the bit shift calculation step 1304 is 1 or more or -1 or less, an instruction to feed one dot is issued. If A ≧ 1, in the correction step 1306, one dot is moved to the right, −1.
If it is below, one dot is sent to the left in the correction step 1307. After the correction, the bit shift amount is decremented by -1 in the post-processing step 1308 or + in the post-processing step 1309.
1, the number of black pixels is measured in the black pixel number measurement step 1310, and the end is determined (step 1311).

Next, an example of actual correction using the above flow chart will be described with reference to FIG.

FIG. 12 shows an example in which the inclination correction method is actually applied to the calculation of the bit deviation of the line. FIG. 12 (a) is a correction example when ΔX per line is 0.25,
FIG. 13B is an example of correction when ΔX is 0.3. Lines 1401 and 1402 in the figure are boundaries of the count area.

When ΔX is 0.25, 0.25 bit shifts occur every scanning of one line, so the bit shift A becomes 0.25 × 4 = 1 every four scanning lines. There is a bit shift A of 0.25 between the line of the top row and the second row, A = 0.5 in the third row, A = 0.75 in the fourth row, and sequentially accumulated. Since A = 1 in the column of, the boundary 1401 is shifted to the right by one dot. With this dot shift, the bit shift of the fifth column becomes A = 1−1 = 0, and the bit shift A of the sixth column again increases by one line and becomes 0.25. 0.5 in 7th row, 0.75 in 8th row, A again in 9th row
= 1 and the boundary 1401 is again shifted to the right by one dot. In this way, the boundary 1401 is sequentially shifted to the right by one dot in the line where the accumulated bit deviation value A is 1 or more, and at the same time, the accumulated value of the bit deviation is changed from A to A-1. Then, ΔX (0.25 in this example) is added to A-1 for each line from the next line, and when the accumulated value becomes 1 or more, the boundary 1
That is, 401 is shifted to the right by one dot.

If ΔX is 0.3, A = 0 for the first line, A = 0.3 for the second line, A = 0.6 for the third line, A = 0.9 for the fourth line, A = 1.2 for the fifth line. Becomes 1
That is all. Therefore, at the fifth line, the boundary 1402
Is shifted to the right by one dot, and A is set to 1.2-1 = 0.2 by the processing of the post-processing unit 1308. Add 0.3 again on the 6th line, A = 0.5, A = 0.8 on the 7th line, A = 1.1 on the 8th line and shift 1 dot to the right again, A =
Set to 0.1. In this way, the count area correction is performed.

Next, the operation of the black pixel number measuring unit 503 will be described.

FIG. 13 is an example in which the character entry frame 305 when measuring the number of black pixels is divided into blocks. This division is performed in order to extract the portion that becomes the point of character recognition from the read image data. The name of each block will be described. 1501
Is a main block 1,1502 is a main block 2,1
503 is a main block 3, 1504 is a main block 4, 1505 is a main block 5, 1506 is a main block 6, and 1507 is a main block 7. 150
8 is sub-block 1,1509 is sub-block 2,15
10 is a sub block 3, 1511 is a sub block 4, 1
512 is a sub-block 5, 1513 is a sub-block 6,
1514 is the sub-block 7, and 1515 is the sub-block 8.
Is. 1516, 1517, 1518, 1519, 1
Reference numerals 520 and 1521 denote boundaries of counting areas, and 1522 and 1522.
1523, 1524, 1525, 1526, 1527,
1528 and 1529 are boundaries of the count lines. Count area boundaries 1516, 1517, 1518, 1
Tilt correction is performed on the positions of 519, 1520, and 1521. The boundary of the count area represents the boundary of the block in the main scanning direction, and the boundary of the count line represents the boundary of the block in the sub scanning direction.

A method of measuring the number of black pixels in the count area after the inclination correction will be described with reference to FIGS.

FIG. 14 is an example of block division of the character entry frame 305 with writing. In FIG. 14, a vertical line 151
Of the sections (blocks) divided by 6 to 1521, the range of addresses 1 to 21 in the main scanning direction from the left is 1, the range of addresses 22 to 25 is 2, the range of addresses 26 to 31 is 3, and the address 32 is 32. 35 ranges 4, addresses 36-5
The range of 6 is numbered as 5, and the horizontal section divided by the horizontal line is from the top, the range of the number of lines Y in the sub-scanning direction is 1 to 18, and the number of lines Y is the range of 19 to 24. Numbers are sequentially numbered up to 7, such as 2. 1601 is the run length of the written black pixel, and 1602 is the block reference point.

From the start address Xs and the end address Xe on the main scanning side from the block reference point 1602 of the run length 1601 of the black pixel, the start block number on the main scanning side (block number with Xs, In the example of FIG. 14, 3) Is and end block number (block number with Xe, 5) I in the example of FIG.
_E is calculated. Further, the sub-scanning side block number JJ is calculated from the position Y on the sub-scanning side from the block reference point 1602 of the run length 1601 of the black pixel and the line conversion table shown in FIG.

Regarding the run length 1601, the block in which the run length 1601 exists (in this example, the sub-scanning side block JJ is 3 and the main scanning side number is Is (3) to _IE (5))
Is used as a count area, and black pixels are counted for each block. The block to be counted is represented by symbol II.

The calculated Is to _IE blocks are sequentially counted blocks. Is becomes the counting target block first, and at this time, boundaries 1518 and 1519 of block II (= Is)
Is assigned to the start address X (II) and end address X (II + 1) of block II (block 3 in the example of FIG. 14), and the run length 1601 of the black pixel is
Counted range in block II of the black run length as compared with the start address Xs and the end address X _E is set, the number of black pixels counted in the area is measured.

FIG. 15 is a detailed diagram of counting black pixels. Reference numeral 1701 indicates a white pixel, and 1702 indicates a black pixel. First, the start address X (II) of the count area (block II) is compared with the run length start address Xs of the black pixel 1702, and the larger one is set as the run length start address of the black pixel 1702 in the count area. In FIG. 14, it is Xs. Next, the count area end address X (II +
1) is compared with the end address X _E of the run length of the black pixel 1702, and the smaller one is the black pixel 170 in the count area.
The end address has a run length of 2. In Figure 15, X
(II + 1). As a result, the number of black pixels in the count area becomes X (II + 1) -Xs, which is counted as the number of black pixels in the blocks on the main scanning side II and the sub scanning side JJ.

FIG. 16 is an address conversion table. 170 black pixels
2 of the main scanning side block numbers by entering the run length of the main scanning side address Xs and X _E Is and I _E
Is output.

FIG. 17 is a line conversion table. Black pixel 1702
The block number JJ on the sub-scanning side is output by inputting the number of lines Y on the sub-scanning side of the run length.

FIG. 18 is a flow chart showing the procedure described with reference to FIGS. 15 and 16 for measuring the number of black pixels in the count area of the black pixel number measuring unit 503. To explain each name of each step, 2001 is a block number determining step, 2002
Is a scan block designation step, 2003 is a count area start address determination step, 2004 and 20
Reference numeral 05 is a count area start address specifying step, 2006 is a count area end address determining step, 2007 and 2008 are count area end address specifying step, 2009 is an area black pixel number calculating step, 2010 is a scanning block changing step,
Reference numeral 2011 denotes a scan block end determination step.

In the block number designation step 2001, a start block Is of black run length data, an end block Ie of black run length data, and a block JJ of a line of black run length data are created from a table. Next, block designation step 200
At 2, the first block Is is given to the counter II. Start address judgment step 2 in the count area
In 003, the beginning of the block given to the counter II (start address X (II) is the beginning X of the run length data).
If X (II)> Xs, it is determined in step 2004 in the count area start address designation that the black in the block is the beginning X of the block.
(II) is designated, and this is designated as the start address Zs. When X (II) <Xs, in the start address designating step 2005 in the count area, the beginning Xs of the run length data is set at the beginning of black in the block and designated as the start address Zs.

Then, in the end address determination step in the count area, it is determined whether the end X (II + 1) of the block is larger than the end Xe of the run length data, and X (II + 1) is determined.
> Xe, in the count area end address designation step 2007, the end Xe of the run length data is set at the end of black in the block and designated as the end address Ze, and if X (II + 1) <Xe, the count area In the inner end address designation step 2008, the end X of the block is set to the end of the black in the block X (II + 1), and this is designated as the end address Ze.

Then, in the area black pixel number counting step 2009, the area black pixel number = Ze−Zs is calculated. Next, in the scan block change step 2010, the counter II is incremented by 1 to shift the count target block to the right by one, and then in the scan block end determination step 2011, whether all blocks have ended for the black run length (3 in the example of FIG. 14). , 4, 5), and if not completed, the process returns to step 2003.

When the above steps are repeated once, the processing of one line is completed, and by repeating this for the required number of lines, the number of black pixels in each block in the character frame 305 can be measured. Further, in order to speed up the above processing, hardware may be used.

FIG. 19 shows a configuration example in which the black pixel number measuring unit 503 has a hardware configuration. 2101 is a run length data storage register, 2102 is an address generator, 2103 is a black run length start address storage register that stores the black run length start address Xs, and 2104 is a black run length end address storage that stores the black run length end address Xe. A register 2105 is an address conversion table shown in FIG.
2106 is a start block number storage register for storing the start block number Is, 2107 is an end block number storage register for storing the end block number Ie, 2108 is a signal when there is a difference between the output of 2107 and the output of 2106. Output comparator, 2
109 receives the signal output of 2108 and increments the output of 2106 by 1, and 2110 receives the output of 2106 to increase the area start address X (I
I) and the division table for outputting the area end address X (II + 1), 2111 for the area start address storage register, 2112 for the area end address storage register, 2113 for the in-area black pixel start address generator, and 2114 for Ze. The black pixel end address generator for outputting the area, 2115 is a black pixel number measuring device for calculating the area Ze-Zs, and 2116 is a 17th for outputting the sub-scanning block number JJ by inputting the line number.
The line exchange table shown in the figure, 2117 is a block pointer for designating the number of the main block or the sub block formed by the block number JJ output from the line exchange table and the block number II output from the counter II,
Reference numeral 2118 denotes an in-block black pixel number measuring device 2119 which adds the output of the in-area black pixel number measuring device 2115 and the output of the in-block black pixel number table 2119 and stores again at the corresponding position of the in-block black pixel number table 2119. The number of black pixels in each main block and sub-block is stored, and the number of black pixels in the main block or sub-block designated by the block pointer is calculated as the number of black pixels in block 2118.
It is a black pixel number table in the block to be output to.

The incrementer 2109 sets the counting target block to Is.
It plays the role of moving to the right in order. Comparator 21
08 plays a role of determining that the counting of the black run length is completed when the count target block becomes Ie.

The operation of the recognition unit 109 will be described.

FIG. 20 is a diagram shown for explaining the concept of the recognition unit 109. 2201 is a main block register, 2202
Is a sub-block register, and 2203 is a recognition table. The recognition unit 109 determines the presence or absence of writing in each block based on the number of black pixels in each block in the character entry frame 305 measured by the black pixel number measurement unit 503 and the determination value, and the main block register 2201 and the sub block. The register 2202 is created. The determination value referred to here is a threshold value for judging that the number of black pixels in each block is the number of black pixels to be written, and the main block register 2201 and the sub block register 2
In 202, only the presence / absence of writing to each block is recorded, and the black image pixel number itself is not recorded. The recognition unit 109 determines what the character is by referring to the created main block register 2201, the sub block register 2202 and the created recognition table 2203, and outputs the character code. The recognition table 2203 is composed of a character code of a character, a sub block table and a main block table corresponding to the character, and a memory for three lines is given to one character.

FIG. 21 is an explanatory diagram shown for explaining an example of the recognition table 22203. Main block register 2201 created depending on the presence / absence of writing in each block
, Sub-block register 2202, and recognition table 2
The recognition main block table value and the recognition sub block table value in 203 are compared, and when both match, the character code is output.

The special work of the recognition unit 109 will be described in detail with reference to FIG. FIG. 22 is a flowchart shown to explain the operation of the recognition unit 109. To describe the name of each step, 2401 is a table creation step, 2402 is a recognition table search offset initial setting step, 240
3 is a recognition table search step, 2404 is a reject determination step, 2405 is a table comparison step, 2
406 is a character code output step, 2407 is an offset change step, and 2408 is a table search end determination step.

In the table creation step 2401, the main block register 2201 (TB (J)) and the sub block register 22
02 (STB (J)) is created. In the recognition table 2203, a combination of a recognition main block table value, a recognition sub block table value, and a character code represented by these values is calculated from a certain address (start address) on the memory as a value of the recognition main block table value. (Second
The values are stored in ascending order of the value in the hexadecimal column in FIG.

Recognition table search offset initial setting step 240
In 2, the offset value from the start address of the recognition table 2203 is set to 0, and the difference between the final address and the start address is set as the reference value. In the recognition table search step 2403, the recognition main block table value and the recognition sub block table value of the address indicated by the offset value are set in the registers PTB (J) and SPTB (J),
At the reject determination step 2404, the value TB (J) of the main block register 2201 is compared with the recognition main block table value, and if the recognition main block table value becomes large, it is rejected as unrecognizable.

In table comparison step 2405, main block register 2201 (TB (J)), recognition main block table value PTB (J) and sub block register 2202 (STB).
(J)) and the recognition sub-block table value STB (J) are respectively compared. If they match, the character code is output in the character code output step 2406. If they do not match, the data of the next character is used. In order to compare and contrast, in the offset change step 2407, the offset is +3 (since the data of one character occupies three lines, that is, three offsets, the offset is shifted by three each time the character to be compared and contrasted is changed. ), And then step 2 of determining the end of the table search
At 408, it is confirmed whether the offset does not exceed the standard value. If the offset exceeds the standard value, it is rejected as no corresponding character in the recognition table. If the offset does not exceed the reference value, the value of the address indicated by the offset value is reset in the table and the comparison operation is repeated.

〔The invention's effect〕

According to the present invention, a character handwritten on an input sheet can be recognized quickly and accurately.

According to the present invention, since black pixel counting is performed using run length data, the amount of data to be referred to is reduced to about 1/10 as compared with the method of processing bit by bit, and the processing time is reduced to about 1
It can be reduced to 1/0.

[Brief description of drawings]

FIG. 1 is an explanatory view showing an example of an input sheet, FIG. 2 is a detailed explanatory view of a character frame, FIG. 3 is a block diagram showing an embodiment of the present invention, and FIG. 4 is a concept of a run length data creating section. FIG. 5 is an explanatory diagram shown for explanation, FIG. 5 is a flow chart showing the details of the operation of the run length data generation unit, FIG. 6 is a block diagram showing an example of the configuration of the run length data generation LSI, and FIG. FIG. 8 is a diagram shown for explaining the concept of FIG. 8, FIG. 8 is a diagram shown for explaining the marker detection method, FIG. 9 is a flowchart shown for explaining the operation of the marker detection unit, and FIG. 10 is a band detection unit. FIG. 11 is a flow chart for explaining the operation of FIG. 11, FIG. 11 is a flow chart for explaining the operation of tilt correction, FIG. 12 is a view for explaining an example of tilt correction, and FIG. , FIG. 14 shows a writing example FIG. 15 is an explanatory diagram showing details of the writing example, FIG. 16 is a diagram showing an address exchange table, FIG. 17 is an explanatory diagram showing a line exchange table, and FIG. 18 is an operation of the black pixel number measuring unit. A flow chart showing
FIG. 19 is a block diagram showing a hardware configuration example for realizing the operation of FIG. 18, FIG. 20 is a diagram showing the concept of a recognition unit, FIG. 21 is an explanatory diagram showing a recognition table, and FIG.
FIG. 23 is a flowchart showing the operation of the recognizing unit, and FIG. 23 is an explanatory diagram shown for explaining the concept of the bit processing method. 102 ... Run length data creation unit, 501 ... Marker detection unit, 502 ... Band detection unit, 503 ... Black pixel number measurement unit, 109 ... Recognition unit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yoshiharu Tatari 4026 Kuji Town, Hitachi City, Hitachi, Ibaraki Hitachi Research Laboratory, Hitachi, Ltd. Ceremony company Hitachi's Totsuka factory (56) Reference JP-A-58-22476 (JP, A)

Claims (1)

[Claims] 1. A reading means for reading, using a facsimile, simple character information such as alphanumeric characters and symbols written by hand on a sheet on which a character cutting marker is independently provided for each character, and the cutting means. Counting means that divides the character area designated by the marker into a plurality of count blocks and counts the number of black pixels in each count block, and recognition means that recognizes characters by determining whether or not writing is performed in each count block. In a simple handwritten character recognition device including the above, the counting means converts the image data input by facsimile into run length data on a line-by-line basis, and a band for detecting the character cutting marker. A detection unit and a black pixel number measurement unit that sets a count area and counts the number of black pixels in the count area using the run length data. Simple handwriting recognition apparatus characterized by comprising Nde.