JPS58146974A - Reader of character - Google Patents

Abstract

PURPOSE:To attain character discrimination processing highly accurately and simply, by performing character discrimination processing independently of character discrimination processing to be collectively applied to the whole categories. CONSTITUTION:After storing a character pattern 11 in a pattern memory 10, a feature group is extracted by a feature group forming part 20 and inputted to a character discriminating part 30 through a feature group register 21. The character discriminatig part 30 matches the contents of combination logics with that of the feature group register 21 in each subcategory on a dictionary memory 40 and outputs a candedate category to an individual character discriminating part 60 through a candidate category register 31. The individual character discriminating part 60 discriminates characters on the basis of an individual feature group extracted from an individual feature group forming part 50 and the elements of successive combination logics from an individual dictionary memory 90.

Description

DETAILED DESCRIPTION OF THE INVENTION (1) Description of the field to which the invention pertains The present invention provides a method for distinguishing individual similar characters in 0 character reading processing, particularly in reading characters with similar glyph shapes.
The present invention relates to a character reading device that performs highly accurate identification processing based on minute differences in character shape.

(Explanation of conventional technology) Documents written using a mixture of kana, alphabetic characters, numbers, etc. Especially in handwritten documents, characters with similar shapes such as kana lchi' and numeral 14', kana l ko l There are many examples such as the alphabetic character Ill, the number 171, and the kana character I91 wa I.

Conventional character reading devices that include these similar characters in the reading target divide a category into multiple subcategories according to the transformation of handwritten characters, and describe each subcategory using combinatorial logic using various identification features. By constructing a dictionary that includes in the above combinatorial logic the discriminative features that represent differences in character shapes that exist between other categories that are easily confused, it is possible to protect subcategories related to handwritten deformation between them and other categories. The range was slightly adjusted.

In such a device, in order to ensure highly accurate reading performance, and to distinguish between various categories including similar characters, a logical number is set for each subcategory.

The number of subcategories for handwritten transformations has increased.

The disadvantages include that the device is expensive, that it takes time to process nine characters, and that it is difficult to modify the dictionary as the number of characters that can be read increases.

On the other hand, when multiple candidate categories appear for an input character in character judgment processing for all categories at once, the glyph differences unique to these categories are extracted as new distinct features, and the There are character reading devices that distinguish between similar characters.

With such devices, it is difficult to perform simple character recognition processing because the characters are not uniform due to similar character slams or handwritten deformations in a certain combination of categories. This method has the disadvantage that it takes time to extract the differences in detail.

(1) Purpose of the Invention The present invention eliminates these drawbacks, and includes character determination processing for nine individual combinations of categories that are similar characters, in addition to character determination processing for all categories ν at once. It is characterized by performing detailed character judgment processing not on a pair-by-pair basis but on subcategory pairs according to handwriting deformation, and giving priority to subcategories ν that are close to standard glyph shapes over others.・Similar glyph shapes in character reading devices The purpose of this invention is to realize character judgment processing for character slams with high precision and simple processing.

(4) Description of structure and operation of the invention Figure 1 shows an actual example of the present invention, in which 10 is a button memory, 20 is a feature group creation section, 21 is a feature group, a register, 30
Reference numeral 40 indicates a character determination unit, and 40 a dictionary memory.

31 is a candidate category register, 50 is an individual feature group creation unit, 51 is an individual feature group register, 60 is an individual character determination unit, 70 is a category pair table, 80 is a subcategory pair table, 90 is an individual dictionary memory, ll is an input 61 is the character determination result.

To operate this, first the input character button 11 to be read is stored in the button memory 10.

The feature group creation unit 20 extracts a feature group that describes the character shape from the al character batan in the batan memo ulO,
The extraction result is stored in the feature group register 21. The feature group is made up of features that are effective in separating each of the nine subcategories from other categories in a character determination process for all categories to be read at once.

Regarding the category ν to be read, the dictionary memory 40 stores
Each category is divided into subcategories according to handwriting deformation, and combinatorial logic for 1 determination is stored. In this embodiment, the combinatorial logic elements of each subcategory on the dictionary memory 40 are It consists of an address indicating the feature content and the #V-valley range of the value of that feature, and the final address of each subcategory contains the category hood to which the subcategory belongs, the subcategory code, and the variation of the standard glyph for that category. A weight value representing the degree of smallness is stored.

In this embodiment, the character determination unit 30 matches one combinational logical content for each subcategory on the dictionary memory with the content of the feature group register 21, and selects the category to which the subcategory that satisfies the logical condition belongs as a candidate category. As the storage order of each subcategory in the dictionary memory 40, the characters are arranged in descending order of the standard shape for each category, and only the subcategories for which the feature group of the character pattern satisfies the combinatorial logic for the first time in each category are selected during the single character determination process. It is configured so that there are N choices. Therefore, one set of categories/categories for each candidate category.
The food, subcategory code, and weight value are stored in the candidate category register 31.

Next, the individual feature group creation unit 50 extracts individual features that describe various differences in character shape that exist between categories of similar characters. Here, various similar characters that are valid for category pairs are extracted all at once and stored in the individual feature group register 51 as nine individual feature groups.

Next, the individual character determination section 60 receives the number of candidate categories for the input character pattern from the candidate category register 31, and when the number is less than one category, the category code is set as the character determination output 61.

When the number of candidate categories is more than 2 categories, first, the weight value for each category in the candidate category register 31 is determined to be small (corresponding to subcategories close to the subcategories II).
As a result, the number of candidate categories is 1.
If it is a category, output that category ν code for character judgment 61
Then, if the power category is 2 or above, the following individual character judgment processing is performed.

Do this.

The purpose of individual character determination processing is to simplify the process.

In this embodiment, when two candidate categories are obtained.

It is assumed that a choice is made based on the combinatorial logic in the individual dictionary memory 90. The category pair table 70.subcategory pair table 80 is a mapping table for quickly searching one individual dictionary memory 90 for combinational logic corresponding to any category pair. However, category pairs and subcategory pairs that do not correspond to similar characters are not included because this can significantly reduce the amount of table memory.

First, the category pair table 70 is searched based on the content 62 of the category pair that required processing in the 0 individual character determination unit 60. If there is a table element storing the contents of the same category pair, the addressing information 71 of the subcategory pair table 80 included in that table element is output.

Next, based on the content 63 of the subcategory pair obtained from the candidate category register 31 through the individual character determination unit 60, the subcategory pair table 80 is searched from the table element designated by the addressing information 71. If there is a table element storing the contents of the same subcategory pair, outputting the addressing information 81 of the individual dictionary memory 90 included in that table element.6 Storing the combinational logic regarding the subcategory pair in the individual dictionary memory 90. The start address is configured to be obtained from the address designation information 81 (first, the 0 combination logical elements 91 are sequentially sent from the individual dictionary memory 90 to the individual character determination unit 60 and the 0 individual feature group register 51 The last combinational logic element 91 of the subcategory pair includes an end flag, and the 1 individual character determination unit 60 repeats the matching process until it receives this flag.As a result, any of the subcategory pairs When a category with a certain power can be selected, the category code is outputted as the character determination result 61, and when not, the eject code is outputted as the character determination result 61.

Because of this structure, when multiple categories of similar character shapes are candidates, it is necessary to associate each category with a subcategory according to the content of the handwritten transformation of the input character pattern, and to assign each subcategory pair to the character A fast search for combinatorial logic to determine the proximity of each subcategory to a standard glyph.

This makes it possible to prioritize and select similar characters depending on the character, and it is possible to realize detailed processing when determining similar characters individually.

FIG. 2 shows the structure of the dictionary memory in this embodiment, and based on this, the contents of one subcategory will be specifically explained. 40 is a dictionary memo! j, 401°402.40
3 is address information indicating a specific feature on the feature group register 21. A%, A%, 411°412.413
is the permissible condition Bj1. of the feature value whose address is specified by AL A inertia, A%. Bm, Bs, 420 is the category code-Ct regarding the above combinational logic.

430 is the jt-th subcategory code SB in the category Cm related to the above combinatorial logic, and 440 is the weight value W(j
It is.

The dictionary memory 40 has a sequential memory structure,
For each category, information regarding all Pcategos ν belonging to that category is continuously stored. The combinatorial logic element (AIB) in FIG.

' Ctj, Wj/) is dictionary information regarding the subcategory Stj.

These operations are performed by access from the single character determination unit 30 in the process of repeating the matching process for each subcategory.

First, the combinational logic element (ALBIL) is input to the single character determination unit 30.
is taken in, and the contents of address 82 of the feature group register are compared with the allowable condition Bffi. If the condition is satisfied in this comparison, the next combinatorial logic element (As,
Repeat the same process for Bs).

If not satisfied, the matching process for the subcategory is ended and the matching process for the next subcategory is performed.If all the conditions are satisfied up to the final combinational logical element (A%, Bm), it is stored at the 0th address (C (.

Sij, W(j) is sent to the candidate category register 31. Therefore, the contents of subcategories that satisfy the 8 combinatorial logics can be easily searched for by simple operations.

Figure 3 is an example of a subcategory related to similar characters. The number 141 is centered around the kana 'chi'. This shows the relationship between kana rkel and kana 'te'.

There are two possible types between Ichil and I41, depending on the presence or absence of a loop structure. A subcategory is created using combinatorial logic that describes the character shape for each of these handwritten transformations, and is stored in the dictionary meso 40. The figure shows ``Chi I to Chi 1.'' Chi atmosphere, Chi1
．． 4 types of Chi 4 141 shows examples of subcategories of 4t, 4mノ21[, 'KeI of KE+'TE of Te. These honorable categories are an example of what is processed by the separate character determination section 60 in this embodiment, and are not just category codes.

By leading to the content of the subcategory food.

The point of view regarding the difference in glyph shape between the two categories becomes clear.

If it falls down (chi 1.4t), the vertical relationship between endpoint l and endpoint 2,
The bending of stroke l, (chi, 4g) is the direction near the end point 3, the bending of stroke 2, (chi 3. keri).

In (H4.TE1), the appearance of the intersection point is 0. In this embodiment, the 0 individual feature group creation unit 50 extracts these features and uses them as collation data in the individual character determination process.

FIG. 4 shows the configuration of the candidate category register in this embodiment. 31 is a candidate category register, 421,
431.441 are register strings that store the category code, subcategory food, and weight value obtained from the dictionary memory, respectively, and are 311°, 312.313, and 314.3.
15 is a register line that stores up to five types of candidate Kanakori information. In the storage example in Figure 1, there are three combinational logics of subcategories S l * S * eS3 for each of the categories ')ClaCm and Cm on the dictionary memory. If the feature group of the input character slam satisfies the 9 weight value witW! , Ws as well as three types of candidate category information are stored.

FIG. 5 is an explanatory diagram of the weight values of subcategories, and shows the subcategories of standard glyphs as 0. Set the category/category with the lowest probability of confusion to 1. This is an example in which a five-value scale value is used, such that the subcategory with a high probability of confusion is set to 2. The above candidate category information becomes input information for the individual character determination process.

FIG. 6 is a concrete explanatory diagram of the search procedure for the individual dictionary menu ν in this embodiment. 62 is category pair information (C=, Cj) regarding character determination unity; 63 is subcategory pair information (Sjm) regarding character determination results.

5js), 70 is a category pair table, 710 is a storage column for category pairs, 720 is a storage column for addressing information, 8
0 is a subcategory pair table, 81O is a storage column for an end flag indicating a boundary, 820 is a storage column for subcategory pairs, 8
30 is a storage column for addressing information, and 90 is an individual dictionary memo! J, 910 is a storage column of end flags indicating boundaries, 920
930 is the address storage column for extracting a specific element of the individual feature group register, and 930 is the former name in the category pair information! 940 is the permissible condition storage column for the latter category CC1) in the category pair information, 71 is the search start address designation information of the subcategory pair table 80, 81 is the search start for the individual dictionary memo ν90 Addressing information 91 is a combinational logic element sent to the individual character determination section 60.

To operate this, first, the category pair table 70 is searched for a table element that matches the category pair information 62. If there is no match, no individual character determination processing is performed. If there is a match, the A(/ stored in the same line as shown in Figure 6) is output as address designation information 71.Next Table 80 vs. subcategories
In.

A table element matching the address A t j based on the address designation information 71 and the pair information 63 is searched. If there is no match after searching up to address A (j+%), which is the end flag Elf, the individual character determination process ends.

In Figure 6, there is a match at address A<7+z.

l stored on the same line! , tj* is stored in the individual dictionary memory 90.
is output as the search start address 81. After this, a 1m separate dictionary memory 90 is provided for each category of the category pair.
Verification is performed based on combinatorial logic.

The combinatorial logic of the category pair is from the address A i j 2 of the individual dictionary memory 90 to the 7 address A ij2 + arithmetic' where the end flag is Ill. Each combination between these
゛The zero combination logic elements that sequentially send the logic to the individual character determination unit 60 are CAL, Bib, Bjk, where h=0.
Bth based on the value of the individual feature at address A& of one individual feature group register.

The individual character determination unit 60 performs a matching process for the two allowable conditions of Blh.

In addition, when distinguishing multiple subcategories belonging to one category from other categories, there is a common difference in the 0 character shape and 1
When the combinational logic based on the individual feature groups has the same content, the following operations are possible in this embodiment in order to increase the memory size for different features. In the table element of the subcategory table 80, when one or both of the subcategory codes stored in the subcategory pair storage column 820 is used to collectively handle all subcategories belonging to the corresponding category.

For example, in the 8-individual character determination process for N s It 41 food, when matching the subcategory pair information 63 with the subcategory code of each table element of the subcategory table 80, if N%ro food is stored, a match is made. Then, all subcategories belonging to the category corresponding to the N and Qn code columns are compared with the individual feature group using one common combinational logic on one individual dictionary memory.

Through the series of operations described above, when an arbitrary category pair is generated as a candidate category pair, a mapping table is created based on the information of the subcategory pair regarding the contents of the handwritten transformation and a small mapping table consisting only of the necessary information. 1g
Relevant portions of the A-separate dictionary memory '90 can be searched quickly and easily.

(5) Description of effects As explained above, in the present invention, when multiple categories are output as candidates as a reading result, the deformation of the input character, that is, the difference in character shape depending on the content of subcategories between categories. You can quickly and easily search for combinational logic written in
Since it is possible to perform matching processing, it becomes clear to focus on differences in glyph shapes between categories when identifying similar characters individually. 9 Consists of combinatorial logic search and matching processing It has advantages such as short processing time for individual character determination. In addition, since we set a weight value for the closeness of the glyph to the standard glyph for each subcategory, we calculated a weight value of 1!1
By comparing the values, there is an advantage that categories close to the #I4I4 type can be easily selected.

In particular, in character reading devices that read a mixture of multiple character types such as kana, alphabets, and numbers, it is possible to perform individual character determination processing for similar characters with a simple configuration and without significantly increasing processing time. It has the advantage of guiding the reading results.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a character reading device according to an embodiment of the present invention, FIG. Iv2 is a block diagram of a dictionary memory in this embodiment, FIG. FIG. 5 is an explanatory diagram of the weight values of subcategories, and FIG. 6 is an explanatory diagram of the individual dictionary memory search procedure in this embodiment. In the figure, IO is a button memory, 20 is a feature group creation unit, 3
0 is a character determination unit, 40 is a dictionary memory, 50 is an individual feature group creation unit, 60 is an individual character determination unit, 70 is a category pair table, 80 is a subcategory pair table, 90 is an individual dictionary memory, 11 is an input character button, 21 is a feature group register, 31 is a candidate category register, 51 is an individual feature group register, 61 is a character judgment output, 62 is category pair information,
63 indicates subcategory pair information, and 71.81 indicates addressing information. Patent applicant Hiroshi Mori, patent attorney representing Nippon Telegraph and Telephone Public Corporation

Claims (1)

[Scope of Claims] (Li) A character memory for storing an input character pattern; a feature group creation unit for extracting a feature group describing a character shape from the input characters stored in the pattern memory; A character comprising a dictionary memory that collects and stores subcategories made up of feature combination logic, and an individual feature group creation unit that extracts an individual feature group that describes the differences between similar categories of the I (tan memorica) shape. The reading device selects a subcategory ν whose feature group of the character pattern satisfies a certain combinatorial logic, and outputs the category code and subcategory code to which the subcategory belongs. A category pair table that collects and stores combinations of similar categories/foods with parts and 1 glyphs, and category pairs/foods with similar glyphs included in different categories.
A subcategory pair table that collects and stores code combinations, an individual dictionary memory that collects and stores combinatorial logic created for subcategory pair separation based on the individual feature group, and information on the subcategory pair table listed in #I. An individual character determination unit is provided for comparing the combinational logic obtained from the individual dictionary memory with the individual feature group of the character pattern. When there is a plurality of category codes obtained from the character determination section, the category codes extracted from the character pattern are extracted from the character pattern based on the information in the category pair table and subcategory pair table and the information obtained from the character determination section. A character reading device characterized in that a character is determined by separating subcategories between categories whose feature groups are satisfied by combinatorial logic in the individual dictionary memory. (Phantom) In the character reading device recited in claim E (1), the one-character determination unit stores a weight value representing the small amount of deformation from the S-subscript form at the final address of each subcategory on the dictionary memory. By doing so, the combinational logic in the dictionary memory is configured to select a subcategory νf that satisfies the feature group of the input character pattern, and simultaneously output the category code, subcategory code, and the weight value. A character reading device characterized in that the character determination unit is configured to output a category code having a smaller 0 weight value with priority based on the size difference of *power distribution values.