JPS60230279A - Recognition system of pattern - Google Patents

Abstract

PURPOSE:To attain the pattern recognition stably at a high speed even if the line width of an input pattern is changed by deciding the input pattern from the result of matching by the 1st and 2nd polarities. CONSTITUTION:After picture information from a camera 10 is binary-coded in a binarization circuit 11, it is inputted to the 1st comparator 13. Simultaneously its polarity is inverted in an inversion circuit 15 and inputted to the 2nd comparator 17. The 1st and the 2nd comparators 13 and 17 calculate matching degrees with the aid of the 1st and the 2nd dictionaries 12 and 16. A decision circuit 18 executes the pattern recognition with the aid of these matching degrees.

Description

[Detailed description of the invention] [Technical field of invention] The present invention relates to a pattern recognition method.

[Background technology of the invention and its problems]

There are various methods for recognizing printed characters, such as a pattern matching method, a stroke analysis method, a feature analysis method, and a composite similarity method. Generally, the pattern matching method is widely used.

Character recognition using this pattern matching method involves capturing an image of the character along with its background using a camera, converting the image information obtained from this camera into binary quantized information using a binarization circuit, and converting the quantized information into binary quantized information. This is a method in which information is successively compared with a plurality of dictionary patterns stored in a dictionary, and the dictionary pattern that most closely matches is used as the recognition result. However, this method is sensitive to changes in the # width of the character, and has the drawback of often causing recognition errors. In other words, the line width of printed characters generally changes depending on the printing conditions, so for example, if the line width is
If the character (1) shown in Figure A is thick, when this is superimposed on the dictionary pattern (2) K stored in the dictionary shown in Figure 1B, a large mismatched part ( 3), and when superimposed with other dictionary patterns, a clear distinction cannot be obtained, leading to recognition errors.

In order to prevent such recognition errors, it is possible to store a large number of dictionary patterns with different line widths for the same character in the dictionary, but increasing the number of dictionary patterns in this way requires a large memory capacity. Not only that, but also the time required for recognition becomes longer.

Another method is to weight characteristic parts of characters when comparing quantized information obtained by converting image information with dictionary patterns, but this method
Compared to the above method, the device is more complicated and larger, and its manufacturing cost is higher. Additionally, published by Kyoritsu Publishing Co., Ltd.
9. There is also a pattern recognition method disclosed in pages 44 to 49 of "Zr Character/Graphic Recognition Machine", but with this method,
There is a problem in which it is mistakenly recognized that the line width of the reference mask stored in advance is not equal to the line width of the input image.

[Purpose of the invention]

An object of the present invention is to provide a turn recognition method that can stably and quickly recognize patterns even if the line width of patterns such as characters and figures changes.

[Summary of the invention]

Regarding the binarized pattern information, a first polarity signal and a second polarity signal obtained by inverting the first polarity signal are obtained, and the first polarity signal is stored in advance. For matching with the memory output of the same first polarity, the second
The polarity signal is matched with a pre-stored memory output of the same second polarity, and a pattern recognition method is obtained for determining the input cover pattern from the matching result of each of the first and second polarities. It is.

[Embodiments of the invention]

Hereinafter, an example in which this 8A is applied to character g recognition will be described based on the drawings.

Figure 2 is a circuit configuration diagram of the character recognition device, where 4 is a camera that images the characters printed on the substrate surface together with the substrate surface that is the background of the characters, for example, an industrial TV camera, and OD is the output of this ITV camera. This camera 2 responds to the captured image by including the signal of a predetermined character area among the signals.
The image information sent from the α groan is A/D converted and 2
A is a binarization circuit that converts the input quantization information into quantization information consisting of a value. 0 is the background part of the character of the quantized information sent from the binarization circuit circle and the first word *a' is stored in the memory in advance.
This is a first comparator that sequentially compares, for example, the background portions of characters in a plurality of dictionary patterns stored in the arc, and calculates the degree of matching between the two for each dictionary pattern. Kasumi is an inversion circuit that inverts the polarity of the quantized information sent from the above binarization circuit (III to the second polarity), and uQ is this inversion circuit (identifies the inverted quantized information (negative) sent out from 1G) A η is a second turn in which a plurality of (negative) turns are stored in memory in advance, and aη is the inversion circuit (
A character portion signal (negative image) of quantized information of second polarity obtained by inverting the polarity of the input signal sent from 1G and the second word 4
A second step of matching, for example, sequentially comparing character parts (negative masks) of a plurality of dictionary patterns of the second polarity stored in ae, and calculating the degree of matching between the two for each dictionary turn.
It is a comparator. 9 and 08 are for determining input characters, and are based on the degree of matching calculated by the first comparator and the second comparator.
The logical product with the matching degree calculated by the comparator aη is calculated for each term 41A.
This is a judgment circuit that calculates for each same character in a turn.

Suppose that a symbol consisting of multiple characters whose reflectance is lower than that of the substrate surface is printed on the substrate surface, and the character information signal is taken in from the camera lens by sampling or other means.
The transmitted image information also consists of a character pattern (sha) in which the character part (21a) has a lower reflectance than the background part (20a), as shown in the third figure. Binarization circuit aυ
Based on the image information sent from the camera (1
, the text part (21b) is rOJ, the background part (2ob)
It is converted into a binarized pattern (c) (positive image) with f r I J. This is used to obtain first polarity pattern information. This binary pattern I23 is converted into information of the second polarity by an inverting circuit, that is, as shown in FIG. 3C, the character part (21
C) is set to "1" and the background portion (20c) is set to "0".

1st polarity -* (Ia is the binarization output from the binarization circuit 0υ) In order to compare with the turn @, the first polarity o
A number of dictionary patterns (positive masks) are memorized, and as shown in FIG.
, the background portion (20d) is set to "1", and the line width of the character is thicker than the line width of the character in the binarized pattern, for example, the maximum thickness that can be recognized as a character. The first comparator is a circuit for checking the characters in the image taken by the camera 4, that is, the characters included in the image information sent from the camera (II) corresponding to this image. , a background part (20
b) Calculate the matching, that is, the degree of coincidence for (20d). That is, as shown in FIG. 3F, the binarization pattern (c) is sequentially superimposed on a plurality of dictionary patterns (251), and the background part (20b) is created for each dictionary pattern (c).
) and (2oa) are both "1". In this case, the degree of matching is defined as the number of pixels in the background part (20b) of the binarized pattern (c).

If the number of pixels in the background part (20d) of the dictionary pattern (■) is N1, it is expressed as n4/NH, and if both patterns 1J3 (c) completely match, rB/N,
= 1, the value becomes smaller as the number of unmatched parts increases.

The second dictionary area stores a plurality of dictionary patterns to be compared with the inverted and nine-binarized pattern (2) sent from the inversion circuit (1), and as shown in FIG. 3E, the moxibustion dictionary pattern ( 5) is the same as the above inverted nine-binarization pattern (2), the character part (21e) is set to "1", and the background part (21e, )
is set to "0" and the line width of the character is thinner than the line width of the character of the inverted binary pattern information, for example, the minimum thickness that constitutes a clear character. The second comparator a'7) is a circuit that checks the skeleton of a character included in the image information sent from the camera 4 in response to the calculation of the degree of matching in the first comparator (first floor). , the inverted binary pattern (2) sent from the inverting circuit (19).
4 and the second dictionary (a plurality of dictionary patterns (5) stored in the IF5), the degree of coincidence of the character parts (21c) and (21e) is calculated. That is, as shown in FIG. The value pattern (to) is sequentially superimposed on a plurality of dictionary patterns (27), and the character parts (21c) and (21e) are both "1" for each dictionary pattern (27).
Find the number of pixels. The degree of matching is determined by the number of pixels in the character part (21c) of the inverted binary pattern (2) being 02, and the number of pixels in the character part (21e) of the dictionary pattern (5) being N2.
Then, it is expressed as nz / Np, and both/zo turnsci 4. If @ perfectly matches, nz / N2 = 1
The value increases as the number of parts that do not match the character part (21e) of the dictionary pattern 127) increases.

The degree of coincidence calculated by the first and second comparators [3] and 0.71 is sent to the sequential determination circuit Q8. That is, the degrees of matching obtained by comparing with the dictionary patterns (c) and Q7) regarding the same character are sequentially sent out. The determination circuit 08 sequentially accepts the degree of coincidence and sequentially calculates the logical product x sentence. Then, it is assumed that the character imaged has the dictionary pattern C251 or (27) from which the maximum logical product value is obtained.

For symbols consisting of multiple characters, the entire symbol can be read by repeating the above operation for each character.

As mentioned above, the image information sent from the camera OQ is binarized in the same way as the character A-tar (binarization pattern (c) corresponding to the 2nd turn (c) and this binarization/ξ turn (c), and A plurality of dictionary patterns (to) in which the line width of the characters is thicker than the line width of the characters of the binarization pattern (c) are sequentially compared, and the degree of matching of the background parts (20b) and (20d) is determined. At the same time, the binarization pattern ξ is reversed with respect to the above binarization pattern.
Turn (to) and this inverted binary pattern (24)
The character is binarized in the same manner as above, and the line width of the character is reversed, and the character is Part (2
1c) and (21e), these background parts (2
0b), (20d) and character part (21c) - (
21e) If you judge the character by logical AND with the matching degree, you can check the character y from a different point of view),
Even if the line width of characters changes, there is no risk of misidentification, and stable recognition can be performed.

Also, even if you double check one character,
The degree of correspondence between the background portions (20b) (20d) of the characters and the degree of correspondence between the character portions (21c) (21e) can be processed by binarization and simultaneously sent to the judgment circuit H in real time, so character recognition is possible. can be processed at high speed.

Furthermore, since the recognition process is performed on a single sheet of paper, the apparatus can be made compact.

An embodiment of the present invention has been described above, but when capturing an image of a subject with a camera, a symbol consisting of a plurality of characters may be simultaneously imaged, and each character may be extracted and processed one by one from the image information obtained as a result. However, the characters constituting the symbol may be imaged one by one with a camera and processed.

Further, the degree of coincidence sent from the first and second comparators may be calculated by logical product in real time as shown in the above embodiment, but it may also be calculated after being stored in the memory. Furthermore, in the above embodiment, a case has been described in which the background verification means is performed first and then character verification is performed for recognition; however, conversely, characters can be recognized by first performing character verification and then background verification. It's okay.

Furthermore, in the above embodiment, an example was explained in which the present invention was applied to character recognition, but it goes without saying that the present invention can be applied to any signal that can be treated as pattern information, such as graphic information or audio information.

〔Effect of the invention〕

According to this invention, the pattern information based on the first polarity and the first
Since the pattern information line width is recognized as the matching with the pre-stored memory contents according to the polarity of the first polarity, and the matching between the pattern information according to the second polarity and the pre-stored memory contents according to the second polarity is recognized as a pattern information line width. It has the effect of being able to be recognized stably even if the values change.

[Brief explanation of drawings]

Figures 1 (4) to (c7) are explanatory diagrams of the conventional character recognition method using the pattern matching method, Figure 2 is a circuit diagram of the character recognition device of the present invention, and Figure 3 is a diagram of the character recognition device of the present invention. It is an explanatory diagram of the character recognition method.Ql...Camera (11)...Binarization circuit a success...
1st dictionary 03... 1st comparator (19 inversion circuit tte
...Second dictionary 0η...Second comparator 0 groan...Judgment circuit @...Character pattern (to)...Binarization/ξ turn @...
...Reversed binarization pattern (25, (2'θ・Dictionary pattern agent Patent attorney Inoue-Male 1st Figure 21! El lift let me 21e

Claims (2)

[Claims] (1) Means for fully binarizing the input pattern information, and means for obtaining a first polarity signal and a second polarity signal obtained by inverting the first polarity signal for the binarized pattern information. , means for matching the signal of the first polarity with a memory output of the same first polarity stored in advance, and means for matching the signal of the second polarity with a memory output of the same second polarity stored in advance. 1. A pattern recognition system comprising: means for performing matching with the above-described first and second polarities; and means for determining an input cover pattern from the results of matching according to each of the first and second polarities. (2) The pattern recognition method according to claim 1, wherein the matching based on the first polarity is matching regarding a background portion, and the matching based on the second polarity is matching related to pattern information. .