JPS63208182A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To improve the recognition precision of an unknown pattern of low quality by measuring a pattern in a resembling area to check the degree of variance of the unknown pattern and changing the unknown pattern based on the degree of variance. CONSTITUTION:If there are a set of resembling categories as the decision result of a deciding means 2, a collating means 10 checks a table 14 stored in a storage means 3 and measures the degree of variance of the unknown pattern in a resembling area 14c from some standard pattern in the same resembling area 14c. A changing means 11 changes the line thickness of the whole of the unknown pattern based on the degree of variance measured by the collating means 10, and such correction is performed in the resembling are that the unknown pattern and the standard pattern coincide with each other. A determining means 4 compares the unknown patterns after change with standard patterns to determine a category name corresponding to the unknown pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pattern recognition device, and particularly to a pattern recognition device for recognition of printed kanji characters and the like.

[Conventional technology]

As a prior art in this field, there is Japanese Patent Publication No. 59-49630 (hereinafter referred to as the document). FIG. 3 is a block diagram showing a schematic configuration of parts related to the present invention of the apparatus disclosed in this document, in which (1) is a selection means, (2) is a determination means, and (3) is a selection means. is a storage means, and (41 is a determination means.

- In the 2-turn recognition device, standard patterns for all types of image sono-turns to be recognized are stored in a recognition standard/setan storage device (not shown in Figure 3) and input. Determine which standard/setan among the stored standard/sequence turns the unknown image nonturn corresponds to. The above documents are related to printed kanji recognition, and since they are printed characters, excluding differences in thickness due to printing pressure or the condition of the printing ribbon, quantization errors in character reading devices, etc.
An unknown character/setan should match a standard pattern of the same character. Therefore, the pattern recognition device shown in the literature compares the logic of each bit at the same position in the input unknown pattern and the standard pattern, and finds the standard/setan character that has the maximum number of bits with matching logic. Determine that it is a pattern category (character type). for example,
Assuming that one character is represented by a bit pattern of x40 pits, if the number of bits in which the bit logics of the unknown pattern and the standard pattern match is 1,600 bits, then the unknown pattern and the standard pattern are completely matched. It turns out.

That's what I did.

Since there are many character types in kanji, ranging from 2,000 to 4,000, it is necessary to memorize that many types of standard patterns, and it is necessary to compare unknown/setans for each of these memorized standard patterns. Since this would require a long processing time, methods have been implemented to shorten this processing time, but this part is not directly related to the present invention and will therefore be omitted. The operation after the stage where the selection means (1) selects the category name of the standard pattern as a candidate for the unknown pattern using the method will be explained.

FIG. 4 is an operation explanatory diagram for explaining the operation of the apparatus shown in FIG. 3. In the figure, (5) is an unknown pattern, and (6) is a selection means ( 1
) of the standard pattern selected by
17+ indicates a table stored in the storage means (3); all but one row in the table is omitted in the figure) (
7a) and (7b) are each in the category “between”.

This row of table (7) stores data regarding a set of similar category sets (i.e., the set of categories "between" and "between" that have mutually similar standard patterns). (7C) is a similar category set (``ma'',
This is a difference area mask pattern that stores a difference area for the difference area (between).

As explained earlier, the standard patterns of (7a) and (7b) are stored in the recognition standard/9-turn storage device, and are stored in (7α) and (76) of table (7) in X7. The code representing the category name is stored,
Using this code as an address, each standard pattern is read out from the recognition standard pattern storage device and used.
7α) and (7b) show the standard/setan itself. Furthermore, according to the description in the above literature, the difference region mask/5 turns shown in (7C) can be used in common for other similar category sets (for example, closed, open, open, closed, etc.), so the (7) in table (7) 7C) The difference area mask pattern number is stored in the noran, the difference area mask pattern is stored in a separately provided mask 9 turn storage device, and the difference area mask pattern number is stored in the table (71
The number stored in the (7C) column is used as an address to output from the mask pattern storage device, but in order to make the drawing easier to understand, the mask pattern itself is temporarily shown in the (7C) column. In the following drawings, such representations will be used where there is no risk of confusion.

(8) in FIG. 4 shows each pattern after masking, and (8α
) is the result of masking unknown patterns, and (8b) is the standard ! for category (7a). As a result of masking the turn,
(8C) shows the result of masking category (7b).

(9) indicates the category name determined by the determining means (4) by comparing (8(Z), (8b), and (8C)).

Next, the operation will be explained. Depending on the character reading device (not shown) that reads printed characters and converts them into character patterns, the dimensions of the character pattern as read may differ from the dimensions of the standard pattern.

Since it is difficult to compare bit patterns with different dimensions, if the overall dimensions of the unknown pattern (5) are different from the dimensions of the standard nosetan, the dimensions of the unknown pattern (5) are normalized to match the dimensions of the standard pattern. must be transformed into In the following explanation, it is assumed that unknown pattern 5) is next to the normal ratio.

The selection means (1) selects the standard nog turn having the bit pattern that most closely matches the unknown pattern (5);
Suppose we get category 澗J +61.

Next, the determining means determines whether or not a similar category set including the category "ma" (6) exists by examining the table (7) belonging to the storage means (3).

In the device shown in FIG. 3, a plurality of category names whose standard patterns are similar to each other are collected to form a similar category group.
This similar category set is stored in table (7) in advance. If you look at table (7), you will find that the category "ma" (
For 6), it can be determined that there are similar category sets such as (7α), (7b), and (7C).

The determining means (4) selects the difference area mask from the table (7).
Extract that turn (7C), and extract this masktu turn (7C).
C) to find the unknown pattern (5) and the category “between” (
7α).

"Mask the standard pattern for interval j (76) (
The bits in the black part of the mask noso turn (7C) are logic “
1'', the bits in the white part are logical ``0'', and outputs the entire logical product of the bits at the same position in the unknown pattern and the standard pattern) and the result (8(L), (8b)).

Obtain the bit pattern (8C). Bit pattern (8
Comparing α) with (8b) and (8C), the bit patterns (8α) and (8b) match better than (8α) and (8C), so the unknown pattern (5) becomes (7b).
The category name corresponding to the unknown pattern (5) is "between" (9) instead of "between" (6). It is determined that there is.

The process performed when the determination means (2) determines that a similar category set does not exist is not related to the present invention, and therefore its description will be omitted.

[Problem that the invention seeks to solve]

Conventional pattern recognition devices are configured as described above and operate as described above, but even characters with the same category name printed by the same printing device may differ depending on the printing pressure or the condition of the printing ribbon. The lines may be thicker or thinner than those of the standard pattern. 8 If a bit no turn of a character printed with such low quality is input as an unknown pattern, the unknown pattern and the unknown pattern The bits between the standard pattern corresponding to
Setan concordance k. Because of this, the bit pattern matching degree between the similar standard pattern similar to the standard pattern and the unknown pattern becomes thicker than ko, and the unknown pattern is different from the above similar standard pattern. It may be mistakenly recognized as having the same category name.

For example, in Fig. 4, if the line thickness of the unknown pattern (5) is larger than the line thickness of the standard pattern, the line thickness between (8a) and (8C) is larger than that between (8α) and (8b). The degree of matching of the bit patterns is greater than that between the two, and the determining means (4)
The problem is that the unknown pattern (5) is erroneously recognized in the category [J (7b)], and the provision of the storage means (3) and determination means (4) has no effect.

This invention was made in order to solve the above-mentioned problems, and is an e-turn recognition device with high recognition accuracy that can recognize the correct category name even when a low-quality unknown turn is input. The purpose is to obtain.

[Means for solving problems]

In the system of the present invention, a storage means is provided for storing a similar area outside the different area for each similar category set, and the amount of variation of the unknown/mother turn from the standard pattern in the similar area is measured. Based on the result, modify the above variation rJJ in the unknown turn by a modification means,
The category name of the unknown pattern was determined by comparing the difference area of the revised unknown pattern with the difference area of all standard patterns in the similar category table.

[Effect]

Similar categories Bit patterns of similar areas in the desk and unknown/
(If the quality of the unknown nosetan is good, it should match the bit pattern in the similar area of the turn. The difference between the two nosetans is due to the influence of printing pressure, the condition of the printing ribbon, etc. This is because the quality of the pattern has deteriorated. Therefore, the variation of both putters in similar areas is measured, and from this variation the degree to which the line thickness of the unknown pattern should be changed is determined, and the thickness of the line of the unknown pattern is determined according to this determined degree. If you change the line thickness of the entire unknown pattern, the bit pattern in the difference area of the turn (unknown) will be corrected, so you can find out which standard pattern in the similar category corresponds to the bit pattern in the difference area. It can be determined correctly whether

〔Example〕

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

FIG. 1 is a block diagram showing one embodiment of the present invention.
In the figure, the same reference numerals as in FIG. 3 indicate the same or equivalent parts, (10) is a collating means, and (11) is a changing means. Further, (30) is a storage means corresponding to (31 in FIG. 3), and is provided with a table for storing information on all similar category sets.
Table (7) in Figure 4 is composed of category names (7n), Cab) and difference area mask patterns (7C), whereas table (14) in Figure 2 is configured as shown in Figure 4). It consists of category names (14a), (14b), difference region mask/4' turf (14d), and similar range mask pattern (14c).

If a similar category set exists as a result of the determination by the determination means (2), the collation means (10) checks the table (14) stored in the storage means (3) and identifies similar areas (146).
Any standard nosoturn in the similar region (14c) that is the same as the unknown pattern in the table (14c)
The bit patterns of all standard patterns in the same row of the standard pattern match).゛The changing means (11) changes the line thickness of the entire unknown pattern based on the degree of variation measured by the matching means (10), and makes corrections so that the unknown pattern and the standard pattern match within the similar area. . Other parts in FIG. 1 having the same reference numerals as those in FIG. 3 operate in the same manner, so a duplicate explanation will be omitted.

FIG. 2 is an explanatory diagram for explaining the operation of the apparatus shown in FIG.
This is the standard pattern category "Ba" selected by 1). Table (14) K has already been explained in connection with storage means (30) in FIG.

(15) shows the pattern after masking with the similar area mask pattern (14i1j), (15α) shows the result of masking the unknown pattern (12), and (15b) shows the category name (14α) (or (14b) ) shows the results of standard pattern gold masking. (16) measures the relative degree of change between patterns (15α) and (15b), and based on the measurement results unknown pattern (12)
(17) shows the bit setturn after being masked by the difference region mask (turn (14d)); 17a) Mother turn (16)
(17b) and (17C) show the results of masking standard patterns (14a) and (14b), respectively.

Also, (18) is (17α), (176), (17c)
The category name determined by the determination means 14) is shown for comparison.

Next, the operation will be explained.

The operations up to the selection means (1) and the determination means (2) are similar to those described with reference to FIG. Assume that the category name J (13) is selected for the unknown turn (12) in FIG. 2, and it is determined that a similar category set exists for the category name J (13). Information regarding the similar category set is represented as one row of data in table (14) (14α), (14b), and (14C).

(14d) is stored.

Next, the matching means (10) masks the unknown pattern (12) and the standard pattern corresponding to the category name (14α) using a similar region mask pattern (14C) to obtain bit patterns (15α) and (15b). , the variation between each bit turn (15α) and (15b) is measured.

The simplest variation measurement is, for example, the number of bits B! of logic "1" (assuming that logic "1" corresponds to a black pixel) in the bit set turn (15a). and bit pattern (
15b) Calculate the number of logic “1” bits Bs in BI
When /BS is calculated, the value of B□/Bs represents the ratio of the line thickness between unknown 9 turns (12) and standard non-turns (14α).

Therefore, the changing means (11) changes the line thickness of the unknown pattern (12) to Bs/B! The unknown pattern (1
Change the line thickness in 2) until it is almost the same as the line thickness of the standard pattern. With this change, for example, Bt
/Bs > 1, the logic of unknown pattern (12) “
The logic “1” bit adjacent to “0” is (1-Bs /B
l) It is sufficient to change the logic to "0" at a ratio of . For example, pattern (16) is obtained by such a change.

The operation in the determining means (4) is as explained in FIG. 4, and (17), (17α), (17b), (
17c) is +81. (8α) , (86) , (8
C), but (17α) is a turn (16
) is the result of masking.

Pattern (16) has been changed so that the line thickness is the same as the standard 9th turn line thickness, so the determining means (4)
1, the pattern (17α) is the same as the pattern (17C). It is determined that it is the mother turn and it is unknown. The mother turn (12) is identified by the name "ba" (14b).
) is determined to be the pattern. In the difference area mask pattern (17), the similar pattern parts are removed, so the difference between the patterns (17b) and (176) is emphasized, and the pattern (17α) is changed from (17b) to
It is easy to determine which of (17c) this corresponds to.

In the above embodiment, when changing the unknown number 9 turn (12) in the changing means (11), the bit of logic "1" adjacent to the bit of logic "0" is changed to logic "0" K. , unknown setan (12) is converted into an electric signal by a photoelectric conversion device, the electric signal corresponding to each pixel is converted into a multi-value digital signal, and this multi-value digital signal is converted into a binary signal of logic [JrOj However, these two
The multi-value digital signal before being converted into a value signal may be stored separately, and the threshold value for converting this multi-value digital signal into a binary signal may be changed based on the value of B/BS.

Further, in the above embodiment, the degree of variation in the matching means (10) is expressed by the ratio of the number of logic "1" bits Bl/B, but it can also be expressed by finding the line width of the pattern and expressing it by the ratio of the line widths. good.

Further, in the above embodiment, the case where one category is selected by the selection means (1) has been explained, but the selection means (1)
11, a plurality of categories may be selected in descending order of similarity, and the determining means (2) may determine whether or not these multiple categories match a similar category set.

Further, in the above embodiment, the similar area mask pattern and the different area mask pattern are stored in the storage means (30).
Similar regions and different regions can also be extracted by overlapping standard patterns that are similar to each other.

Further, in the above embodiment, the unknown/mother turn is changed in the changing means (11), but this may change all the standard/second turns in the similar category table.

Further, in the above embodiment, the selection means (1) and the determination means (4)
(beep) A? Although the degree of similarity was determined by matching the pattern bit logic, it would be better to determine the degree of similarity using various features extracted from the pattern.

Furthermore, in the above example, A? Although the case where the thickness of the pattern line fluctuates has been described, the apparatus of the present invention can also be applied to the case where the thickness of the dA pattern fluctuates due to chipping, collapse, noise, etc.

〔Effect of the invention〕

As described above, according to the present invention, the degree of variation in the unknown 9 turns is determined by measuring patterns in similar regions, the unknown arm turn is changed based on the degree of variation, and the unknown pattern after the change is determined to be different. Since the category name for the unknown turn is determined by comparison with the standard pattern in the region, it is possible to accurately recognize a low-quality unknown pattern with similar categories and large fluctuations without misrecognizing it.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is an operation explanatory diagram explaining the operation of the device shown in FIG. 1, FIG. 3 is a block diagram showing a conventional device, and FIG. 4 is a block diagram showing the conventional device. FIG. 4 is an operation explanatory diagram illustrating the operation of the apparatus shown in FIG. 3; (1) is a selection means, 2) is a judgment means, (30) is a storage means, (10) is a collation means, (11) is a change means, (4)
is the determining means, (12) is the unknown pattern, (13) is the category name selected by the selecting means for the unknown pattern (12), and (14) is provided in the storage means (30) and stores information on similar category sets. A table for storing (14a). (14b) are category names, and (14c) are similar area masks! (14d) is the difference area mask/turn. Note that the same reference numerals in each figure indicate the same or corresponding parts.

Claims (2)

[Claims] (1) In a pattern recognition device that stores standard patterns for all types of image patterns to be recognized in advance and determines standard patterns corresponding to inputted unknown image patterns, Selection means for selecting category names of standard patterns as candidates, collecting a plurality of category names whose standard patterns are similar to each other to create similar category sets, and selecting similar areas where patterns are similar to each other for each similar category set. and a different region where the patterns are different from each other; a determining means for determining whether or not a similar category set including the category name exists for the category name selected by the selecting means; If it is determined that a similar category set exists, a similar region corresponding to the similar category set is extracted from the storage means, and the degree of relative variation between the unknown pattern and the standard pattern is calculated in this similar region. Based on the verification means to be measured and the degree of variation measured by this verification means,
A changing means for changing the unknown pattern or the standard pattern to reduce the degree of variation, extracting a difference region corresponding to the similar category set from the storage means, and extracting the difference region corresponding to the similar category group from the storage means, and extracting the difference region corresponding to the similar category set from the storage means, A pattern recognition device comprising: determining means for comparing an unknown pattern with all standard patterns in the similar category set and determining a category name for the unknown pattern from the comparison result. (2) The storage means for storing similar regions where patterns are similar to each other and different regions where patterns are different from each other for each similar category group stores them in the form of a similar region mask pattern and a different region mask pattern. A pattern recognition device according to claim 1, characterized in that: