JPS6036628B2 - Segment method in pattern recognition - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides a novel segmentation approach in pattern recognition.

In pattern recognition, the segment method is one of the effective and appropriate methods to prevent misreading.

However, the conventional segment method outputs a segment signal to cut out characters after detecting a collection of continuous parts of characters or after confirming a white background of a certain width or more after the end of the character section, and it is a method to cut out characters by detecting dirt or stains on the label. etc., and it was possible that a stain in a space between characters could be determined to be a character part and be erroneously segmented. In view of the above, the present invention provides a guard bar in front of the character, calculates a reference width signal from the bar, and calculates a value in the scanning direction (horizontal direction) by multiplying the width value of the reference width signal by several times. Set the scanning field of view corresponding to the column width, while in the scanning direction (vertical direction), use the number of scan lines as length information, calculate the length between the scan lines, and then tell the number of lines. The length is measured based on the reference width signal above, and the character part is
Set the length of the blank part, and determine what relationship the actually obtained continuous part of the scan signal that contains black information, and conversely the part of the scan signal that does not contain black information, have with the set length of the character part and blank part. In other words, the method performs segmentation by comparing the size relationships to determine character portions or blank portions, and has the feature that stains and stains of less than a certain width can be ignored and segment errors can be prevented.

This method will be specifically explained below with reference to embodiments shown in the drawings.

Figure 1 shows the label to be scanned according to the embodiment, and the number string 1
and a guard bar 2 provided in front of it. In the figure, scanning is performed from bottom to top, and when one of them (Q scan) is taken and the binarized video signal (scan signal) is shown, it looks like a stretched scan signal in the printed part. level, and other parts are at L level. A reference width signal R is calculated from the binary signal obtained by the guard bar 2, and the width of the character string 1 is set based on this reference width signal R, and the field of view for scanning is set from the start signal a to the above reference width. The character width setting value obtained from R is counted and determined by generating an end signal b. Further, the reference lengths CR and SR for the scanning progress method V of the character portion and the space portion are determined from the reference width R. That is, in the first scan, a reference width R is obtained from the guard bar 2, and an end signal b is counted from this R to determine the scanning field of view and the character range. Next, scanning is performed sequentially, but in the figure, the number of scans is counted from the scan line 8 at the tip of the first character 2, and if you count the number of scans to the end in combination with the spacing between the scan lines, it is character 2.
The width C of the frame is 0 length, and by comparing it with the previous character standard width CR, it becomes clear whether it is a character part or not, and as shown in the figure, C>CR
If so, an end signal for the character section is generated and becomes a start signal for counting the next blank section. Scanning is continued, and if the scan signal of a blank area that does not include black information exceeds the previously determined blank area reference width signal SR, as with the black area, that area S is determined to be a blank area, and the reference width signal SR is determined. A segment signal is issued at the time of the width signal SR. Further scanning is continued, and for character 5, the same operation as for character 2 is repeated again. In this way, the present invention obtains the reference width signals corresponding to the character parts and the blank parts between characters from the guard bar, and calculates the number of scans of the characters and the blank parts, respectively, which are converted into length information to be actually bound. It compares the reference width signals of the characters and blanks with the respective reference width signals, determines whether or not the width is a character or blank, and performs segmentation based on the determination result. To convert the number of scans into length information, all you need to do is calculate how many standard clock pulses the interval between scan lines corresponds to. explain. In a TV camera, the number of effective scan lines per frame is about 250, the effective horizontal scan time is about 53 French Sec, and the horizontal/vertical screen ratio is 4:3. Furthermore, if the frequency of the reference clock pulse is f, then the relationship between the horizontal day length clock pulse count and the vertical scanning number is 53 x 10-6 x f = 2, where the conversion coefficient is K.
50×XK K〒0.16×f×10-6, which indicates that the interval between each scanning line corresponds to the length of K clock pulses in the horizontal direction.

Therefore, how many scanning lines the reference width signal R set in the scanning direction corresponds to can be realized by, for example, the circuit shown in FIG.

That is, the count value RD based on the reference clock with the reference width R is input to one side of the comparator 3, and the K clocks counted for each scan are input to the other side, and the clock pulses input for each scan are the count value. When it matches RD, a match output is output, and it becomes clear how many scans the reference width R corresponds to.

Note that in order to output K clocks for each scan, a closed loop can be constructed using a flip-flop 7, an AND gate 5, and a coefficient multiplier 8 that integrates the K clocks and outputs the final output, as shown in Figure A. For example, insert a K times coefficient multiplier 9 between the counter 4 and the comparator 5 as shown in (c), or insert a 1/K times multiplier 10 on the input side of the count signal RD of the reference width R as shown in C. This can be easily realized. In this way, the reference width signal R obtained from the guard bar 2 is set to the date in the scanning direction of the time axis, but once the format of the character string is determined, the reference width signal R for the character part and blank part is set accordingly. CR
, SR are found, and the conversion coefficient K corresponding to this CR, SR is determined.
By setting , segments can be automatically performed regardless of the size of the enemy label.

A specific example of a circuit for realizing the above-mentioned method shown in FIG. 3 will be shown.

In the figure, block 1 is a circuit that determines whether or not the binary video signal (scan signal) obtained as a result of scanning contains a character part, that is, black information. , into the flip-flop 12, set a start signal a indicating the beginning of a character string, and an end signal b indicating the end, respectively, based on the flip-flop 13 which serves as a reset input, that is, AND with its Q output.
The gate 11 is opened and closed, and the end signal is sent to the timer 14.
as a reset signal for the destination flip-flop 12 through
If the scan signal V contains black information, the AND gate 11 is opened by the start signal a, the Q output of the flip-flop 12 becomes H level, and the black information is input to the block D which performs the next character section discrimination. do.

On the other hand, if the scan signal does not contain black information and is blank information, it is reset by the information sent from the end signal b through the timer 14 based on the first scanning, and the 9 output goes to H level. The flip-flop 12, which is , is not inverted, but it is ANDed with the output of this Q and the end signal b indicating the end of the current scan signal, and the signal containing black information is detected as a segment by determining the blank area between characters. bu for giving signals. Send it to Kukku. This is a circuit for determining whether or not it is a block character part, and the setting of the coefficient unit 80 corresponds to the character width in the circuit for converting the reference width R to the scanning direction as explained in FIG. The flip-flop 70 receives the black information scan signal, the AND gate 50, the counter 80, the counter 40, the comparator 30, and the output of the coefficient multiplier 80 which resets the flip-flop 70 by a coincidence output. An OR gate 160, whose function will be explained later, is inserted into the reset side of the comparator 40.
It consists of a gate 170.

That is, if the number of scan signals containing black information is greater than the reference signal CR for determining whether it is a character portion, a match output is output, the counter 40 is cleared, and the next signal for determining a blank portion is output. It is sent to block m. Flip-flop 70 is also reset. Therefore, when black information continues to arrive, the counter 40' starts counting again from the beginning. In this way, when it is determined that these series of scan signals contain black information and are character parts, and a matching output is generated, the signals are sent to the next block m, and the flip-flop 1 in this block m is
The Q output of No. 9 becomes H level. Therefore, when the character part ends and the scan signal changes to a "signal without black information", the AN
The D gate 18 opens, and it is determined that it is a blank area, which consists of 7 m of flip-flops, 5 m of AND gates, 8 m of coefficient multipliers, 16 m and 17 m of OR gates, 4 m of force counters, and 3 m of con/gulators, and performs segmentation. If the number of scan signals that do not include black background information is greater than the reference blank area SR, a matching output is output and it is determined that the blank area is between characters. , a segment signal is output and segmentation is performed. This corresponds to 20 in FIG.

For example, if a stain 21 exists between 5 and 4 in FIG. 1, it is determined in block D that the character "5" satisfies the standard width CR of the character part, a match output is issued, and the gate 18 is opened. After the character part is completed and the white information is input, the counter 4m of block m starts counting. In this state, if the stain 21 is hit before the white background standard width SR has been taught, the white background scan signal disappears and counting by the counter 4m is temporarily stopped, and conversely, the counter 40 on the block side starts counting.
Of course, the size of the stain 21 is smaller than the reference width CR of the black background, and the white background information comes again before the black background matching output is generated.
As a result, the counter 4m on the white background side of the block m starts counting again, and is integrated with the count value that was previously stopped at the beginning of the blot 21, and conversely, the counter 40' of the block 4 on the character side also stops counting. The white background continues, but eventually counter 4
The count value of m reaches the white background reference width SR, a coincidence output 22 is generated, and segmentation is performed. That is, even if there is a slight stain or the like on a white background, it is ignored and determined as a white background, and a segment signal is generated. Note that the black background information due to this stain 21 remains in the counter 401 as a character area, and the count for the next character area is accumulated on top of this, and although it will be somewhat shorter than the actual value, it will depend on the size of the stain 21. The width is smaller than the reference width CR of the black background, and there is almost no problem. The same thing can be said for black background information. For example, if there is a "miss" in the character part and a part of "4" in Figure 1 is not continuous and the scan signal contains white background information, the white background scan Counting by the block counter 40 is temporarily stopped during the signal, and counting is restarted again when switching to the black ground scan signal, and the setting of the black ground width is eventually extended by the missing portion. In short, segmentation is performed while ignoring stains and stains in the label so as to minimize the influence of stains and stains on the label. Contrary to the above, stains and dirt may also be treated as regular characters and blank areas to require greater precision in the subsequent pattern recognition process. The counter 40 due to chipping, etc.
It is conceivable to clear 4m, return to zero, and start counting again from the beginning.

In other words, the segment signal is not output unless black background information or white background information is received continuously, and the black background scan signal is used as a reset input for the counter 4m, as shown by the dashed line in FIG. 3, or vice versa. This can be realized by using the white background scan signal as the reset input of the counter 40. Further, as shown by the broken line in FIG. 3, the coincidence output of block 0 can be used as the reset input of the counter 4m of block m. This is because the scan signal for the character section is input while counting the white background scan signal, that is, before the white background width reaches the standard set width SR, and the scan signal width reaches the normal reference width CR.
If it has, a matching output with a black background will be emitted from the Prokuro, but the matching output will clear the counter 4m,
After the character part is finished, the white background scan signal is counted again, and the white background information counted up to the middle of the previous part is reset.
This prevents incorrect segments due to integration. In other words, if the space between two characters is less than the specified white width, it is of course considered to be one character and segmentation is not performed, but since the white space between them is counted, the latter This white background count value is cleared by the matching output of character information, and the white background area after the end of the character area is counted from zero to confirm that this white background area is larger than a prescribed reference width SR before segmentation. As described above, the present invention provides a guard bar in front of the character string, calculates the reference width using the guard bar, determines the standard for determining whether it is a character part or a blank part from the above reference width, and also calculates the reference width in the scanning direction. The feature is that the setting is made by the number of scan lines, and if the actual scan signal is larger than the character or blank area, it is determined that it is a character area or a blank area, and a segment signal is issued. This makes it possible to improve reliability compared to the conventional segmentation method, which simply determines whether it is a character portion or a blank portion from the actually obtained scan signal and performs segmentation.

[Brief explanation of the drawing]

Fig. 1 shows an example of a label according to the present invention in which a guard bar is provided in front of a character string and its scan signal waveform, Fig. 2 shows a circuit for converting the reference width obtained from the guard bar into the scanning direction, and Fig. 3 1 and 2 respectively represent an embodiment of a segment method assembled based on the conversion circuit shown in the second figure. 1: A circuit for determining whether or not a scan signal contains an H level;・・・
. . . Similarly, a circuit that counts the number of L level scan signals and determines whether it is a blank area or not. Spear figure 2 figure na 3 figure

Claims (1)

[Claims] 1 For example, a guard bar is provided on the label to be scanned, and from this guard bar, a reference width signal corresponding to characters, figures, etc. on the label is obtained, and in the scanning direction, the number of scan lines is used as length information, that is, the scanning Determine the spacing between lines and check how many of them match the reference width signal above, and set each reference value for the character part and the blank space between characters in advance based on the reference width signal above. Then, determine whether the actually obtained scan signal contains the H level or not. If it contains the H level, count the number of consecutive signals and compare it with the reference value of the previous character part. If it is larger than that, the character is The gate is opened by this signal, and the scan signal of the blank area that does not include the H level after the end of the character area is counted, and the counted value is higher than the set standard value of the blank area. This segment method for pattern recognition is characterized in that when the area becomes large, it is determined that it is a blank area and a segment signal is output.