JPS594066B2 - Character extraction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character extraction method for a character recognition device, and more particularly to a character extraction method that can effectively remove noise other than characters.

In general, characters printed in one line of 35 rows are scanned in the vertical direction by a sensor having photoelectric conversion elements arranged vertically in a row, and the sensor or the characters are moved horizontally to read them one after another. FIG. 1 shows a situation where the field of view of the sensor 2 is moving over a sheet of paper 1 on which characters are printed. Paper 1 has the numbers ``1'', ``2'', and ``3'' printed on it as characters 3 to be read, and there are stains 4 and 5 around it. These stains are other than character information to be recognized, and become noise, causing misreading or unreadability. Conventionally, such noise has been removed by changing the reading level, by installing a special filter in front of the sensor, or by reading the noise once and then processing it using software to remove it. was.

However, the former method has unreliable operation, and the latter method requires a large amount of program capacity and processing time, and still cannot effectively remove noise. Furthermore, if the characters printed in one horizontal line are uneven vertically, or if the sensor is manually moved and scanned, the characters may protrude from the sensor's field of view 2, resulting in erroneous identification. SUMMARY OF THE INVENTION An object of the present invention is to eliminate the above-mentioned drawbacks of the prior art and to provide a cutting method that is highly effective in removing noise. Another object of the present invention is to provide a method for cutting out characters so that they can be reliably distinguished from adjacent characters. Still another object of the present invention is to provide an extraction method that reduces the burden on the computer of the recognition unit and improves the reading speed by realizing extraction using hardware. Still another object of the present invention is to provide a method for cutting out images that is not significantly affected by the scanning speed even if the moving speed of the sensor is different. Still another object of the present invention is to detect when a part of the character to be read protrudes from the field of view of the sensor, and to prevent erroneous identification.

The present invention calculates the appearance frequency distribution of character line segments projected onto the vertical axis of the read pattern every time one scan in the vertical direction is completed,
This method is characterized by normalizing the frequency distribution, finding vertically continuous line segments of a certain length from the normalized data, and cutting out the height of the character from the length range of the line segments. The present invention will be explained below with reference to the drawings. FIG. 2 shows, as an embodiment of the present invention, a character recognition apparatus in which a scanner 12 is held in the hand and moved in the horizontal direction to perform manual scanning.

A sheet of paper 11 is illuminated with lamps 13a and 13b, and a character pattern on the sheet of paper 11 is imaged through a lens system 14 onto a sensor 15 having photoelectric conversion elements arranged in a row.

Each photoelectric conversion element of the sensor 15 will be referred to as a cell, and the area where data is compressed to the required resolution will be referred to as a unit area. Since the light reflected from the background area of the paper 11 and the light reflected from the character area are different, the signal levels from the corresponding valley cells are different. The signals obtained from each cell of the sensor 15 are applied to the control and binarization circuit 16, and the level is determined to determine whether the signal is white or black. For example, a signal corresponding to the background area, ie, white, of the paper 11 is set to "0," and a signal corresponding to the character area, ie, black, is set to ``11,'' which will be output to each of the following. When the counter control circuit 25 sequentially receives signals corresponding to each unit area from the control and binarization circuit 16, when the signal is black, that is, 61'',
The output of the shift register 27 is preset in the preset counter 26, and 1 is added to the contents of the preset counter 26 and the result is stored in the shift register 27.

Also, when the signal is white, that is, "01", the preset counter 26
The output of the shift register 27 is preset, and the contents of the preset counter 26 are directly transferred to the shift register 27.
It is controlled so that the data is stored in the . Further, when the one-character scanning end signal GAP (described later) is input from the gear flag 21, the counter control circuit 25 clears the contents of the preset counter 26, and when the signal is 61'', it adds 1 and resets the signal to 601. Sometimes, the contents of the preset counter 26 are controlled to be stored as they are in the shift register 27.The shift register 27 has the number of stages equal to the number of cells of the sensor 15 or the number of unit areas corresponding to the number of data compressed in the vertical direction. The number of appearances of black in each unit area is counted by the pulse signal CK for each unit area scanned in the vertical direction, and the contents of the shift register 27 are the number of character lines in the horizontal direction for each unit area. It represents the number of occurrences. Also, when the contents of the preset counter 26 are stored in the shift register 27, the contents of the preset counter 26 are sent to the comparison circuit 28. The comparison circuit 28 compares the contents of the storage device 29 (denoted as A). ) and the content input from the preset counter 26 (assumed to be B) Store B in the storage device 29 only when B>A. As soon as one vertical scan is completed, the signal SEND from the control and binarization circuit 16 The contents of the storage device 29 are transferred to the maximum value storage device 30 at the timing of , and then the contents of the storage device 2
Clear the contents of 9. That is, the maximum value storage device 30 stores the maximum value of the counted number of black appearances for each unit area stored in the shift register 27. The normalization circuit 31 calculates the ratio between the contents of the shift register 27 and the contents of the maximum value storage device 30, and divides the ratio into a plurality of data (hereinafter referred to as normalized data).

When moving the sensor manually, the speed of movement of the sensor varies greatly depending on individual differences or each character scanned by the sensor. Since the number of appearances is normalized, the normalized value is not affected greatly even if the horizontal scanning speed differs. The normalized and divided contents are sent to the binarization circuit 32 or the identification circuit 35. The identification circuit 35 identifies characters from the normalized data, for example, based on the arrangement of the normalized data. Further, the binarization circuit 32 compares the normalized data with a certain threshold value SHL, and divides the binary values, that is, the pattern information projected on the vertical axis, into a character area and a background area of the paper 11. For example, in FIG. 1, sensor 2 is t1
FIG. 5 shows the normalized data at the T2 position when moving from the position to the T2 position, and FIG. 6 shows the binarized normalized data. The normalized data corresponding to character 3 and stains 4 and 5 are 51, 50, and 52.

In FIG. 6 after binarization, the background area and character area are expressed as W (white) and B (black), respectively. A read-only storage device 33 stores the output signal of the binarization circuit 32 and the storage device 3.
The read contents are composed of the output contents of 4 and are stored in the storage device 3.
4 and the signals SET, POUT, ER, O
Output MIT.

The memory device 34 is cleared by the 1 vertical scan end signal SEND.

FIG. 4 shows a state transition diagram of the contents read to the read-only storage device 33. In the figure, NB, nW (n: number or p) are W, B, which add the type of state to the arrow.
indicates w (white) and B (black) of the binary signal, and SE, T, ER, POUT, OMIT added to the valley state are the signals output when each state is reached, (0MIT ) is 0M when one column of vertical scanning is completed and the corresponding state is
It indicates that the IT signal is output.The signal SET indicates the start of character cutting, the signal ER indicates noise because the line segment is longer than the character height, and the signal POUT indicates the end of character cutting, and the signal 0.
MIT means that a part of the character line segment protrudes from the sensor field of view 2, and priority is given to the signal output later. Assume that a signal as shown in FIG. 6 is input from the binarization circuit 32.

SO and Sl5 represent the top and bottom cells of the sensor field of view 2. The binary signals from the top of the cell are B, W, W,
W, W, W, B, B, B, B, B, B, W, W, W, B
If this is output, the state will be 0B, 1 in Figure 4.
1B, 1W, 1B, 2B, 3B, 4B, 5B, 6B, P
It becomes B. Once the signal SET is output in the state of 11B, it becomes 1.
In the state of B, the signal SET is output again, and then in the state of PB, the signal END is output, indicating that cutting out of characters in the vertical direction has been completed.

Misrecognition may occur if the sensor is manually moved and scanned, or if the printed characters vary widely vertically and some of the character lines protrude from the sensor's field of view. A case will be described in which it is detected that a part of a character line segment has protruded from the sensor field of view.

FIG. 7 shows the normalized data and the binarized data when the sensor visual field moves from the position T3 to the position T4 in FIG. 1.

From now on, the binary signal will be B, B, B, B, B, W, W, W
, W, W, W, W, W, W, W, W, the states are 0B, 11B, 12B, 13B, 1 from Figure 4.
4B, 15B, 11W, and 1W, and when the state is 11W, the signal 0MIT is output, which means that the sensor has gone beyond the upper end of the field of view. Also, in FIG. 4, state 5B
, 6B, and 7B, when one vertical scan is completed and the signal SEND is output, it is assumed that the character has protruded from the lower end of the sensor field of view, and a signal 0MIT is output.

The control and binarization circuit 16 has a counter to determine which unit area is currently selected during vertical scanning, and the content SCC of this counter indicates the address SCC of the unit area.

In FIG. 3, when a signal SET is input from the read-only storage device 33, the storage device 37 selects the current address SCC.
is written. Since priority is given to the signal output later, in the binary data shown in FIG. 6, the address SCC at that time is written again at the second SET signal.
After cutting out the characters and moving to the PB state, SE
The state of PB is held and the POUT signal continues to be output until the ND signal arrives.

When the POUT signal is output when one vertical scan is completed, by writing the contents of the storage device 37 to the storage device 38 at the timing of CKl immediately before the arrival of the SEND signal, the storage device 38 is filled with characters in the vertical direction. is cut out, and the address CTA of the unit area at the top end of the character is stored. The signals SET and POUT are ANDed by AND gates 39 and 40, respectively, with a signal CK output for each unit area scan and a clock signal CKl generated every time a vertical scan is completed, and input to the storage devices 37 and 38. Furthermore, when the output of the AND circuit 40 becomes logic 61'', the cutting flag 42 is set to indicate that the character is cut out in the vertical direction.The signal ER output due to noise during cutting, etc., extends beyond the sensor field of view. When the signal 0MIT or the space signal SP described later is output, the OR gate 45 clears the cutout flag 42 and assumes that the character has not been cut out.If the character has not been cut out, an erroneous recognition occurs. In another embodiment, when a character or the like protrudes outside the field of view of the sensor and a signal 0MIT is output, this signal is processed as if it were an unidentifiable character.

When detecting the end of scanning for one character, the arithmetic circuit 44 adds a constant (FIG. 6 d) to the address of the unit area and compares it with a certain margin above the vertically cut out range of the character. Output to circuit 41. The comparison circuit 41 compares the unit area address CTA of the uppermost end of the cut out character with the content sent from the arithmetic circuit 44, and when the two are equal (the upper end cell is set to O, the lower the cell, the more the address A signal is output to the character height detection flag 43 when the character height detection flag 43 is turned upside down). The character height detection flag 43 is set when the cutout flag 42 is set and a signal is input from the comparison circuit 41, and the counter 47 that outputs the signal POF outputs the signal POF.
A character height detection flag that counts the clock signal CK until a unit area corresponding to the character height CHH (see Figure 6) is scanned after it becomes logic "17", and outputs an overflow signal when CHH is counted. 43 is cleared by the logical sum signal (logical sum circuit 48) of this overflow signal and the 1-column vertical scanning end signal SEND.For example, the content of SCC during vertical scanning is Sn, and the height of the character is h.
Since we know that, in Figure 6, CHH=h+2
d When CTA+h+d≧Sn≧CTA-d, it is a character area and a signal POF is output.

Counter 47 is also cleared by signal SEND.

With the above operations, the character position in the vertical direction is cut out as shown as POF in FIG. 6. Next, we will discuss cutting out in the horizontal direction. Gap detection flags 18 and 19 are set by the SEND signal generated after vertical scanning is completed, and gap detection flag 18 is cleared when black (character line segment) is detected between the vertical character cutting signals POF. The detection flag 19 is a flip-flop that is cleared when black (character line segment) is detected within one vertical scan. Immediately before generating the SEND signal after one vertical scan, the gap detection flag is set to 1 at the timing of the block CKl.
Take the logical sum of the output signals of 8 and 19 (logical sum circuit 20)
This logic signal is stored in the gap flag 21 and the contents of the gap flag 21 are transferred to the gap flag 22. In other words, the gap flag 21 is activated when black (character line segment) is not detected in the vertical character extraction section (POF).
Alternatively, it is a flag that is set when black (character line segment) is not detected in one vertical scan, and the gap flag 22 stores the contents of the gap flag 21 during the previous vertical scan. The output signal GAP of the gap flag 21 is output to the counter control circuit 25, and is ANDed with the signal SEND by the AND gate 23 to add the contents of the space counter 24. When the content of the space counter 24 exceeds a preset constant, it is assumed that one line of characters has been scanned, and a space signal SP is output, setting it to an initial state. The space counter 24 is black (
Cleared when a character line segment) is detected. The gap flag 22 outputs the signal BFGAP and stores the signal GAP. At this time, BFGAP is 617 and GAP is "O"
When this happens, it is assumed that one character has been scanned, and the character is identified. When the sensor field of view 2 moves to position T2 in FIG. 1, if it is cut out as shown in FIG. 6, the gap detection flag 19 is "01" but the gap detection flag 18 is "1" and the gap flag 21 is 1F. is set to .

If the gap flag 21 is set to "0" at the position before T2, then at T2 GAP="1" and BFGAP="O", which means 1.
It assumes that character scanning is completed and then performs identification operations.

The space counter 24 determines the end of one line when the number of vertical scans is much greater than the number of vertical scans between characters. As explained above, in the present invention, the appearance frequency distribution of character line segments is determined for each valley unit area projected onto the vertical axis for each vertical scan, the number of appearances for each unit area is normalized by the maximum number of appearances, and then the character It is characterized by being unaffected by the moving speed of the sensor.

In addition, if the characters printed in one horizontal line are uneven vertically, or if the sensor is manually moved and scanned, it will be detected that a part of the character line protrudes from the sensor field of view.
It is characterized by preventing misidentification. In the present invention, since binarization is performed by normalizing and appropriately selecting the threshold value SHL, noise around characters can also be removed.

In the explanation, the shift register 27 has the same number of stages as the number of cells of the sensor 15, but it is also possible to reduce the number of stages of the shift register 27 by compressing data or the like. Furthermore, the state transition shown in FIG. 4 differs depending on the characters to be read and the dimensions of the sensor. The above explanation deals with the case where the scanning speed in the horizontal direction is variable, but for example, when scanning is performed mechanically at a constant speed, the count value of the number of appearance of character line segments for each unit area and the maximum value of the number of appearance It is possible to similarly extract only the count value without finding the ratio.

[Brief explanation of the drawing]

Figure 1 is a diagram of the relationship between read characters and read heads, Figures 2 and 3 are block diagrams of the present invention, Figure 4 is an explanatory diagram of a read-only storage device, and Figure 5 is the frequency of appearance of character line segments. A distribution diagram, FIG. 6 is a binarization change diagram, and FIG. 7 is an explanatory diagram of an example in which characters protrude from the sensor field of view. Explanation of symbols, 1...Paper, 2...Sensor, 3...Character, 4,5...Noise,
11...paper, 12...scanner, 1
3a, b... Lamp, 14... Lens system, 15... Sensor, 16... Control and binarization circuit, 17, 23, 39, 40, 46...
...AND gate, 20,45,48...OR gate, 18,19...Gap detection flag, 2
1, 22... Gap flag, 24...
・Space counter, 25... Counter control circuit, 26... Preset counter, 27...
...Shift register, 28...Comparison circuit, 2
9... Storage device, 30... Maximum value storage device, 31... Normalization circuit, 32...
Binarization circuit, 33...Read-only storage device, 34
...Storage device, 35...Identification circuit, 3
7, 38... Storage device, 41... Comparison circuit, 42... Extraction flag, 43...
・Character height detection flag, 44... Arithmetic circuit, 4
7...Character height counter, 50, 51, 52.
...Normalized data.

Claims (1)

[Scope of Claims] 1. A sensor consisting of a plurality of photoelectric conversion elements arranged in a line that is scanned at a constant period is moved in a direction substantially perpendicular to the arrangement direction of the photoelectric conversion element rows to detect characters and symbols. In an optical character recognition device that scans and identifies character lines, the number of appearance of character line segments is counted for each unit area of the sensor consisting of one or more photoelectric conversion elements, and the appearance frequency distribution is determined. A character or symbol is scanned according to the frequency distribution, and it is detected whether or not a part of the scanned character or symbol protrudes from the field of view of the sensor. If the part of the scanned character or symbol protrudes from the field of view of the sensor, no identification processing is performed. A character extraction method characterized by processing characters as unidentifiable or unidentifiable characters. 2. Scanning and identifying characters, symbols, etc. by moving a sensor consisting of a plurality of photoelectric conversion elements arranged in a line that is scanned at a constant period in a direction substantially perpendicular to the arrangement direction of the photoelectric conversion element row. In an optical character recognition device, the number of occurrences of character line segments is counted for each unit area consisting of one or more photoelectric conversion elements of the sensor, and the appearance of character line segments for each unit area in the arrangement direction of the photoelectric conversion element array is calculated. The maximum value of the number of occurrences is detected, the appearance frequency distribution is normalized by calculating the ratio between the maximum value and the number of occurrences detected for each unit area, and the characters or characters scanned by this normalized appearance frequency distribution are A character extraction method that detects whether or not a part of a symbol protrudes from the field of view of a sensor, and if it protrudes from the field of view of the sensor, performs no identification processing or processes it as an unidentifiable character. 3. Scanning and identifying characters, symbols, etc. by moving a sensor consisting of a plurality of photoelectric conversion elements arranged in a line that is scanned at a constant period in a direction substantially perpendicular to the arrangement direction of the photoelectric conversion element row. In an optical character recognition device, the number of times a character line segment appears is counted for each unit area of one or more photoelectric conversion elements of the sensor, and character line segments are calculated for each unit area in the arrangement direction of the photoelectric conversion element array. Detect the maximum number of occurrences of , find the ratio between the maximum value and the number of occurrences detected for each unit area, and obtain a normalized appearance frequency distribution;
After comparing the distribution with a constant value and binarizing it, it is detected whether a part of the scanned character or symbol has protruded from the sensor's field of view, and if it has protruded from the sensor's field of view, an identification process is performed. A character extraction method characterized in that characters are not processed or are treated as unidentifiable characters.