JPS61147385A - Recognizing device - Google Patents

Abstract

PURPOSE:To generate a sure feature dictionary by providing a recognizing device with a filter means which can eliminate an indefinite part due to quantization error. CONSTITUTION:A dictionary matching means 8 compares and collates sequentially with many dictionary patterns stored in a dictionary storage means 6 with one another with respect to all combinations of them and selects the combinations of patterns which resemble each other very much respectively and have the degree of discordance lower than a certain reference, and selected patterns resembling each other are compared and collated with each other by a feature dictionary generating means 9, and the characterized difference is detected as a feature pattern. The output of the feature dictionary generating means 9 is given to an indefinite feature part eliminating means 11, and only feature parts from which indefinite parts are eliminated are registered in the dictionary storage means 6 by a feature dictionary command from the indefinite feature part eliminating means 11.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a pattern recognition device, and in particular, extracts a mismatch pattern part between similar dictionaries as a special part, and extracts an uncertain part within a predetermined bit from the special part. The present invention relates to a recognition device that extracts a pattern from which .

[Conventional technology]

Conventional pattern recognition devices compare and match the pattern to be recognized with dictionary patterns registered in advance, and determine the pattern with the smallest degree of mismatch to be the pattern to be recognized. However.

Even if you compare and match very similar patterns.

Since there are few areas of mismatch, there is a risk of erroneous recognition due to the degree of mismatch caused by scratches in the recognized pattern. In order to eliminate such shortcomings of conventional recognition devices, the present applicant registers differences between similar patterns in a dictionary as feature patterns, and weights the feature dictionary patterns to improve recognition accuracy. He proposed a pattern recognition device (Japanese Patent Application No. 59-131717). An embodiment of this pattern recognition device will be explained with reference to the system diagram shown in FIG. The pattern to be recognized 1 is imaged by an imaging means 2, and the analog video signal imaged by the imaging means 2 is converted into a binary video signal by a binarization means 3 such as an analog-to-digital conversion circuit. is stored in the frame storage means 4. On the other hand, the plurality of dictionary patterns registered in the dictionary storage means 6 and the recognized patterns stored in the frame storage means 4 are superimposed and compared in the pattern matching means 5, and the pattern with the least degree of mismatch is recognized. Pattern 5a to be
The recognition result 7a is outputted through the control means 7.

The dictionary matching means 8 provided for dictionary creation instructs the pattern matching means 5 to compare and match the patterns for the combinations of dictionary patterns registered in the dictionary storage means 6. By selecting combinations of very similar dictionary patterns for which the degree of discrepancy between registered dictionary patterns is below a certain standard,
An inter-dictionary matching command 8b is executed. The mismatch pattern 5b between the dictionaries in this selected combination of similar patterns is sent to the feature dictionary creation means 9, and a feature dictionary is created as a result of superimposing the similar patterns. This feature dictionary is registered 9a in the dictionary storage means 6. Furthermore, when comparing and matching with the recognized pattern, weighting is performed by the matching means 10. When a mismatch occurs in the feature dictionary part, the number of mismatching bits is counted and 1 weighting coefficient 10a is used for comparison and matching via the control means 7. In addition to the number of mismatched bits, the overall judgment was also calculated.

[Problem that the invention seeks to solve]

According to the above-mentioned conventional structure, there is a drawback that an uncertain part similar to the special part caused by a quantization error during pattern matching between dictionaries is also created as a pattern for the special part.

[Means for solving problems]

In order to solve the above-mentioned problems, the present invention creates a reliable feature dictionary by providing a recognition device equipped with a filter means capable of removing uncertain portions caused by quantization errors.

Hereinafter, the principle system diagram of the recognition device of the present invention will be explained with reference to FIG.

In FIG. 1, the same parts as in the conventional FIG. 5 are given the same reference numerals, and redundant explanation will be omitted. The area within the broken line is the same as that in FIG. 5, and the output of the feature dictionary creation means 9 is given to the uncertain special feature part removing means 11, and the feature dictionary command 11a from the uncertain special feature part removing means 11 is used to store the dictionary in the dictionary storage means. 6, only the special part from which the uncertain part of the special part has been removed is registered.

[For production]

In the recognition device described above, the dictionary matching means 8 sequentially compares and matches a large number of dictionary patterns stored in the dictionary storage means 6 for all combinations, and selects combinations of very similar patterns whose degree of mismatch is below a certain standard. As an example of comparing and collating the selected similar patterns in the feature dictionary creation means 9 and detecting their distinctive differences as feature patterns, the original image of the character 0 (see FIG. 2 (8)) is stored in the dictionary storage means 6. ) and the original image of the character C (Fig. 2 (b)) are stored, and dictionary matching is performed to create the image shown in Fig. 2 (C).
When the images are superimposed and compared and verified as shown in .

In this way, combinations of very similar patterns whose degree of mismatch is less than a certain standard are provided to the feature dictionary creation means 9 via the pattern matching means 5. Here, the feature dictionary creation means 9 focuses only on the superimposed differences in FIG. 2(C), that is, the mismatched portions 12.

For the character 0, this part is used as the special part 13 of the character, as shown in Figure 2+d), and for the character C.

Special equipment part 1 of the background (blank area) as shown in Figure 2 (Q)
In step 4, feature patterns are created only for portions that differ between similar patterns. Conventionally, as shown in FIG. 2(e), an uncertain element portion 15 occurs due to quantization error, and such uncertain element portions 15a, 15b are generated as shown in FIG. 2(d), (e).
), they will be mixed with the character and background features 13 and 14 created by the feature dictionary creation means. Therefore, in the present invention, this is removed by the above-mentioned uncertain special part removing means 11.

〔Example〕

An embodiment of the pattern recognition device of the present invention will be described below in detail with reference to FIG. 3 and pattern 4. In Fig. 2, the original image was directly compared and verified as a dictionary, but in the present invention, Fig. 3 (al~
(As shown in kl, the feature dictionary creation means 9
Created by That is, Fig. 3(al) and (e) are, for example, original images of similar characters 0 and C, Fig. 3(b), (f)
is stored in the dictionary storage means 6 as a thick dictionary with thick characters made from the original image, and similarly, FIG. 3 (C1, (g) is stored as a thin dictionary with thin characters made from the original image) It is stored in 6.

To explain the procedure for creating a feature dictionary using such a thinning and thickening dictionary, Figure 3 (11 shows a pattern when the letter O of the thinning dictionary is superimposed on the letter C of the thickening dictionary;
The thin dictionary portions 15c and 16 of 0 that protrude from the thick dictionary are added to the uncertain feature portion removing means as candidates for the feature dictionary. Similarly, as shown in FIG. 3(j), by superimposing the thick dictionary of O and the thin dictionary of C, the thin dictionary part 15d of C that protrudes from the thick dictionary of 0 is added to the uncertain special part removing means. . Here, 16 is a portion that should become the original feature dictionary, and 15C and 15d are uncertain special feature portions.

As shown in FIG. 4, the uncertain special part removing means is composed of, for example, an nxm bit filter for removing the uncertain part of the characteristic pattern. The one shown in Figure 4 is m x n
The filter has a matrix configuration of 3 x 3 bits, that is, the uncertain special part removing means 11, and the mismatch pattern data stored in the storage means of the feature dictionary creation means 9 is processed through the first shift register 23. Arranged Ml
, M2° is added to the matrix section 26 of M3, and is also arranged in the second row through the second shift register 24.
a, Ma, and Ma are added to the matrix section 26.

Further, row array matrix units MA, M5 . Ms is M in the matrix section 26 arranged in rows in the next third column? .

'Mal and M9 are connected to perform 3×3-bit matrix scanning, and the outputs of each matrix M1 to M9 are applied to the AND gate circuit 27 for each row array group.

The AND gate circuit output is given to the dictionary 6.

Of course, an nxm bit filter configuration requires n shift registers and nxm matrices.

By passing the information through the filter as described above, the uncertain portion due to the quantization error that occurs when creating the special machine partial dictionary is completely removed. In other words, as shown in FIG. 3 (1), 0), the characteristic patterns 15c, 15c, 15c, 15c, 15c, 15b, 15c, 10, 12, 12, 12, 12, 22,
5d is completely removed and corresponds to special equipment part 16 20.2
Since only 1 can be taken out as shown in Figure 3 +d) and (h), this special part is registered in the dictionary storage means 6.
1a.

In the above embodiment, similar characters such as the letter O and the letter C have been described, but the same comparison and matching as described above is performed for other similar characters.

〔Effect of the invention〕

Since the present invention is configured and operated as described above, patterns other than special differentials due to quantization errors are completely removed, and special part patterns can be automatically created.

[Brief explanation of the drawing]

FIG. 1 is a system diagram showing the principle configuration of the recognition device of the present invention. Fig. 2 (al to e) are explanatory diagrams of the operation of the feature dictionary creation means and the uncertain special part removal means shown in Fig. 1. Fig. 3 is a second embodiment showing another embodiment of the recognition device of the present invention. An explanatory diagram similar to the figure. Fig. 4 is a specific circuit diagram of the uncertain special feature section removal means of the present invention. Fig. 5 is a system diagram of a conventional recognition device. 1... Pattern to be recognized. 2 ...Imaging means. 3.. Binarization means. 4.. Frame storage means. 5.. Pattern matching means. 6.. Dictionary storage means. 7.. Control means. 8..・Dictionary matching means. 9: Feature dictionary creation means. 10: Weighting is done by matching means. 11: Uncertain special feature section removal means.

Claims (2)

[Claims] (1) An imaging means for imaging the pattern to be recognized, a binarization means for digitizing the analog video signal from the imaging means, a frame storage means for storing the binarized output, and a frame storage means for storing the binarized output. A pattern matching means for superimposing a stored recognized pattern and a dictionary pattern in a dictionary storage means created in advance and counting the number of mismatched bits between the two, and pattern matching between the dictionaries at the time of dictionary creation to achieve a high degree of consistency. It has dictionary matching means for detecting similar dictionaries, and feature dictionary creation means for creating a feature dictionary from mismatched parts between similar dictionaries, and the feature dictionary creation means removes uncertain feature parts having a size within a predetermined bit. A recognition device characterized by comprising an uncertain special part removing means. (2) In the feature dictionary creation means, one of the two similar dictionaries is a thick dictionary with thick patterns, and the other is a thin dictionary with thin patterns, and the two dictionaries are superimposed, and one An indeterminate special part removing means is provided for creating a mismatching pattern part of the other dictionary that protrudes from the other dictionary as a feature dictionary, and removing an indeterminate feature part having a size within a predetermined bit from the pattern part of the feature dictionary. A recognition device according to claim 1, characterized in that: