JPS59128681A - Character reader - Google Patents

Abstract

PURPOSE:To select a candidate character string from a word dictionary while remarking a character with high similarity and to recognize the character string for a short time by using the 1st grade or the 2nd grade of a recognition dictionary in accordance with the using frequency of character. CONSTITUTION:A character entered in a business form 1 is scanned by a scanner in a character reader and the feature of the character is extracted by a feature extracting means 2. Subsequently, the similarity is calculated from the feature variable of the input character extracted by the 1st similarity calculating means 3 and the feature variable of a reference pattern stored in the 1st recognition dictionary 4. The reference pattern of a character with high using frequency is stored in the dictionary 4, and while remarking the recognized character with high similarity by a deciding means 5 inputting the operation result of the means 3, a character string including a candidate character to be recognized is selected from the word dictionary 6. If no reference pattern in the dictionary 4 exists in the character string selected from the dictionary 6, the similarity is found with the 2nd similarity calculating means 8 by using the 2nd recognition dictionary 7. Thus, the character string can be recognized for a short time.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention provides a character reading device that uses a word dictionary to recognize and read a character string composed of, for example, a plurality of characters written on a form or the like. It is related to.

[Prior art]

Conventionally, this type of character reading device has a reference pattern for all characters to be recognized written on a form, etc., and determines the character by calculating the similarity between the input character and the above reference pattern. was. For example, when reading Japanese text that includes kanji, since common kanji are widely used in Japan, recognizing about 1,000 to zooo character types can cover most of the character types used. However, it cannot be said that there are no kanji other than common kanji that are used less frequently, and if these kanji are included, there are about 5,000 to 10 kanji.
,000 character types need to be recognized. Therefore, in this case, it is necessary to calculate the degree of similarity between the input character and the reference pattern of all recognition target characters, regardless of how frequently they are used, so as the number of recognition target characters increases, There was a drawback that processing time increased.

[Summary of the invention]

This invention was made for the purpose of eliminating the above-mentioned drawbacks of the conventional methods.In order to recognize characters recorded on forms, etc., this invention uses a recognition dictionary for frequently used characters to recognize input characters. Find the similarity, select a string from a word dictionary by focusing on characters with high similarity, and if this string contains characters that are used less frequently, select a recognition dictionary for characters that are used less frequently. The object of the present invention is to provide a character reading device that can recognize character strings in a short time by using a method to determine similarity.

[Embodiments of the invention]

Examples of the present invention will be described below. FIG. 1 shows the internal components of a character reading device which is an embodiment of the present invention! It is a lock diagram. In the figure, 1 is a form, 2 is a feature extraction means, 3 is a first similarity calculation means, 4 is a first recognition dictionary, 5 is a determination means, 6 is a word dictionary, 7 is a second recognition dictionary, B is the second similarity calculation means.

Next, the operation of the character reading device configured as shown in FIG. 1 will be explained. First, the ledger! The characters written above are scanned by a scanning device (not shown), and the features of the characters are extracted by the feature extraction means 2. Next, the first similarity calculation means 3 calculates the similarity from the feature amount of the input character extracted by the feature extraction means 2 and the feature amount of the reference pattern stored in the first recognition dictionary 4. In addition, the first
The recognition dictionary 4 stores standards and base turns of frequently used characters. The determining means 5 focuses on recognition candidate characters with a high degree of similarity, and selects a character string including the recognition candidate character from the word dictionary 6. If all the characters in the selected character string exist in the reference pattern in the first recognition dictionary 4, the sum of the similarities for each character calculated by the first similarity calculating means 3 for each character string is calculated. The character string with the maximum value of this sum is the recognition result. Furthermore, if there is a character in the character string selected from the word dictionary 6 that does not exist in the standard gloss turn in the first recognition dictionary 4, that is, a character with a standard z4 turn in the second recognition dictionary 7, this The second similarity calculating means 8 is used to calculate the similarity only for the characters, and the character string having the maximum sum of the similarities for each character is taken as the recognition result.

FIGS. 2, 3, and 4 show one character string for explaining the operation of the character reading device shown in FIG. 1.

FIG. 3 is a diagram showing a specific example of similarity and the like.

In Figure 2, 9 is a "wedge-shaped" input character, and 10 is the first
This is an example of the standard Gutanme character in the recognition dictionary 4, and the character example 10 of jin is IJ, r form.

4- It is indicated by each character 13, 14, and 15 of "state".

11 is a character example of the standard mother turn in the second recognition dictionary (F), and this character example 11 is represented by 16.1 characters each of ``rnlJ'' and ``r fang''. Reference numeral 12 is an example of a character string in the word dictionary 6, and this example character string 12 is "elephant".

Each character string 1B is "cuneiform", "cuneiform", and "toga".

19, ao, and 21.

In FIG. 3, as a result of calculating the input character 9 of "cuneiform" by the first similarity calculation means 3, candidate characters with high similarity are "
In the candidate character 22 of ``shape'' (jl), there is no candidate character that has a high degree of similarity to the character 16 of r model J. Among the 12 character string examples in the word dictionary 6, "geograph" is the second word with "shape".

Each "cuneiform" character string 18 and 19 is selected as a candidate character string.

In FIG. 4, the degree of similarity of each character string 18 and 19 of "elephant j r wedge shape" with respect to the input character 9 of "cuneiform" is shown. In other words, the similarity of "elephant" to the character string IB is 0.50, indicated by the code z3, for "elephant" determined by the first similarity calculating means 3, and 0, indicated by the code 24 for "shape". It can be found from .95. Further, the degree of similarity to the character string 19 of "cuneiform" is 0, which is indicated by the symbol 25 to rmJ found by the second similarity calculation means 8.
．． 90 and a similarity of 0.95 indicated by the reference numeral 26 for the "shape" obtained from the first similarity calculation means 3. In other words, the degree of similarity of "Glyph" to the character string 18 is 0.5.
0 +0.95 = 1.45, and the "cuneiform" character string 1
The similarity to 9 is 0.90 + 0.95 = 1.85, so the input character 9 of "cuneiform" is "cuneiform" with high similarity.
is recognized as character string 19.

Therefore, compared to a conventional character reading device of this type that calculates the degree of similarity between an input character and all reference patterns in a recognition dictionary, the character reading device of the present invention described above Narrow down candidate character strings by focusing on characters that are frequently used and have a high degree of similarity to the input character, and then calculate the similarity of these characters only when there are characters that are used less frequently in this candidate character string. Therefore, similarity calculation need only be performed for characters in the first recognition dictionary 1 and characters in the second recognition dictionary 7 that are present in the candidate character string. Therefore, character strings can be recognized at a very high speed. Moreover, specifically, the first recognition dictionary 4
As the second recognition dictionary 7, for example, a high-speed ROM (read-only memory) is used, and as the second recognition dictionary 7, for example, a low-speed,
This can be easily achieved by using a large-capacity magnetic disk.

Although the above embodiment describes the case where the number of characters in the input character string is two, the present invention is not limited to this, and it is sufficient that the number of input characters matches the number of characters in the word dictionary; Not limited.

〔Effect of the invention〕

As explained above, this invention uses a hierarchical (two-stage) recognition dictionary for characters with high or low frequency of use, and uses a recognition dictionary of characters with high frequency of use as the first stage. Find the similarity of the input characters, focus on the characters with high similarity, and select candidate sentence 7 from the word dictionary. Since the system is configured to calculate similarity using a recognition dictionary for characters that are used infrequently in each step, it is possible to reduce the number of similarity calculations as much as possible, and to recognize input character strings at extremely high speed. This has excellent effects.

[Brief explanation of drawings]

FIG. 1 shows the internal components of a character reading device which is an embodiment of the present invention. Lock diagram, Figure 2. 3 and 4 are diagrams showing specific examples of two character strings, similarities, etc., for explaining the operation of the character reading device shown in FIG. 1. In the figure, l... form, 2... feature extraction means, 3
... first similarity calculation means, number ... first recognition dictionary, 5 ... determination means, 6 ... word dictionary, 7 ... second
recognition dictionary, 8... second similarity calculation means. Agent Makoto Kuzuno- 8-

Claims (1)

[Claims] In a character reading device that reads characters recorded on a form or the like by &[, the character extraction means scans the characters and extracts their characteristics, and the first recognition target character type standard and main turn are stored. a recognition dictionary; a first similarity calculation means for calculating similarity from the characteristics of the input character and the characteristics of the base turn (reference in the first recognition dictionary); The second one that stores the standard t4 turn of the Song Dynasty characters
a second similarity calculation means for calculating similarity from the features of the input characters and the features of the reference pattern in the second recognition 1-leaf dictionary; A word dictionary storing character strings composed of many types of characters to be recognized, and a determining means for recognizing and determining the character strings, the character having a high degree of similarity determined by the first similarity calculating means. is selected from the word dictionary, and if the selected character string is composed only of the first recognition target character type, the character string for each character calculated by the first similarity calculation means is The similarity of the entire character string is calculated from the similarity, and if the selected character string includes a second recognition target character type, the similarity with this second recognition target character type is calculated from the second recognition target character type. The similarity of each character with the first recognition target character type obtained by the first similarity calculation means and the second similarity calculation means obtained by the second similarity calculation means. A character reading device that recognizes a character string by determining the similarity of the entire character string from the similarity of each character with a character type to be recognized.