JPS6116383A - Word reading system - Google Patents

Abstract

PURPOSE:To improve the word recognition rate to shorten the processing time by selecting properly characters in accordance with relations between the number of characters having the similarities higher than a prescribed value and prescribed lower limit and upper limit numbers and using selected characters as recognition candidate characters. CONSTITUTION:A character on a form 1 is photoelectrically converted by a scanning means 2, and this photoelectrically converted character pattern is recognized by a character recognizing means 3, and plural recognition candidate characters to which similarities are added are selected and outputted. Plural recognition candidate characters are selected by a recognition candidate character selecting means 8. The recognition candidate character selecting means 8 uses similarities obtained from a similarity table 9 for recognition candidate character selection and upper and lower limit values obtained from an upper and lower limit table 10 for recognition candidate character selection to select recognition candidate characters. A word determining means 4 selects the word, which has a minimum sum of the order and exists in a word dictionary 5, from recognition candidate characters.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to a word reading method, and more specifically, a word reading method that recognizes each character of a word, and recognizes and determines a word based on the results. It is related to the method.

[Prior art]

Conventionally, the characters that make up a word are recognized character by character, the ranking is determined according to the degree of certainty (similarity) as characters, a certain number of recognition candidate characters are selected, and these recognition candidate characters are Words were selected from combinations by referring to a word dictionary.

For example, there is a word reading method according to the above-mentioned prior art as shown in FIG. The functions and processing flow of each component in FIG. 1 are as follows.

In other words, the characters on the paper/print are photoelectrically converted using a scanning means],
This photoelectrically converted input character pattern is recognized by the character recognition means 3.
A plurality of recognition candidate characters are selected with a degree of similarity representing the similarity to a reference pattern that has been recognized in the character recognition means 3 and stored in advance in the character recognition means 3, and the word determination means tei selects these recognition candidate characters A combination of recognition candidate characters is created using a certain number of recognition candidate characters selected in descending order of similarity, and a word is selected from this combination using the word dictionary S to determine a word name. was.

FIG. 2 is a diagram for explaining a word determination method in a conventional word reading system.

6 in the figure is an example of the input word "
This input word 6 is photoelectrically converted by the scanning means and sent to the character recognition means 3 as an input character pattern. Reference numeral 7 denotes recognition candidate characters obtained by the character recognition means 3, which are selected from the top 6 characters by numbering each character constituting the input word 6 in descending order of similarity. An example of a method for determining word names using a word dictionary is a method of first ranking words and then using the rankings themselves. In this method, it is checked whether a word consisting of a combination of each character of the recognition candidate characters 7 exists in the word dictionary in descending order of similarity, and if there are multiple words, the ranking given to each recognition candidate character is determined. The word whose sum is the minimum is determined to be the input word 6, that is, "Kawasumi".

In the combination of recognition candidate characters shown in Figure 2, if we limit the reading to Japanese surnames, the word stored in the word dictionary is "Kanwazumi".
Only. With this word ° determination method, if six recognition candidate characters 7 are selected for each character (up to the 6th place), it can be read correctly, but if there are 5 or less (within the 5th place), "shu" is Since it is not selected, it cannot be read correctly.

As described above, in the conventional word name determination method using rankings based on similarity, the reading rate improves as the number of target recognition candidate characters increases, but the disadvantage is that the processing time increases.

In order to suppress this increase in processing time, in other conventional word reading methods, instead of selecting a fixed number of recognition candidate characters, word names are determined only from recognition candidate characters with a predetermined degree of similarity or higher. method is adopted. but,
In this case, if the similarity is increased, no recognition candidate characters will be selected, and if the similarity is decreased, too many characters than necessary will be selected. The drawback was that it was extremely difficult.

[Summary of the invention]

This invention was made to eliminate these drawbacks, and aims to improve the word recognition rate and shorten the processing time.

In order to achieve this objective, the technical means adopted in the word reading method according to the present invention recognizes each character that makes up a word, and assigns a degree of similarity to each character to indicate its accuracy as a character. If the number of characters having a degree of similarity equal to or higher than a specified value among the recognition candidate characters is less than a predetermined lower limit number, the recognition candidate characters are selected by the predetermined lower limit number in order from the characters with the highest similarity. is selected, and if the number of characters having a degree of similarity equal to or greater than the specified value is greater than a predetermined upper limit value, select the predetermined upper limit number of characters in descending order of similarity, and if the number of characters having a degree of similarity equal to or greater than the specified value is If the number is greater than the lower limit number and less than the upper limit number, only characters having a degree of similarity equal to or greater than the specified value are selected, and word names are determined from a word dictionary for the selected recognition candidate characters. There is.

[Embodiments of the invention]

Preferred embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 3 is a block diagram showing the configuration of an embodiment of the present invention. The functions and processing of each component in FIG. 3 are as follows.

First, the characters written on the paper L are photoelectrically converted by the scanning means 1, and this photoelectrically converted character pattern is recognized by the character recognition means 3, and a plurality of recognition candidate characters with degrees of similarity are selected and output. .

Next, a plurality of recognition candidate characters outputted from the character recognition means 3 are selected by the recognition candidate character selection means g. This recognition candidate character selection means uses the similarity obtained from the recognition candidate character selection similarity table 9 as L, and the recognition candidate character selection upper and lower limit table i.

Let N be the upper limit value obtained from , and M be the lower limit value obtained from . In other words, among the recognition candidate characters with the similarity obtained from the character recognition means 3, if the number of characters having a similarity higher than the specified value is less than or equal to the lower limit M (M>0), the characters have a high similarity. Select M recognition candidate characters in order from , and if the number of characters having a similarity higher than a specified value is the upper limit N (N>M) or more, select N recognition candidate characters in the same way,
There are more than M characters with similarity equal to or greater than the specified value, N
If the number of similarities is less than a specified value, it has a function of selecting only characters having a degree of similarity equal to or higher than a predetermined value.

Next, word names are determined by the word determining means g from the recognition candidate characters obtained from the recognition candidate character selecting means g. For example, the word determining means DA selects a word that has the smallest sum of ranks and exists in the word dictionary S from the recognition candidate characters obtained from the recognition candidate character selection means using the method described in the prior art. .

Next, the operation of the embodiment of the present invention will be explained.

Figure 7 shows an example of candidate characters for recognition obtained by scanning the input word 6 written on form 1, ``Kawasumi'', with the scanning means 2, and recognizing each character with the recognition means 3. The value corresponds to the degree of similarity.

FIG. 5 shows that when the similarity is obtained from the recognition candidate character selection similarity table for the recognition candidate characters in FIG. This is an example of selecting recognition candidate characters. That is, for the input character "God jtt", the specified value L (L = / 20)
There are two recognition candidate characters with the above similarity, and the input character "
For the input character [Jt3, there are 5 recognition candidate characters with similarity equal to or higher than the specified value L (L=/20), and for the input character There are nine recognition candidate characters having a degree of similarity. Further, the ranking given to the recognition candidate characters is No based on the degree of similarity.

FIG. 6 shows the recognition candidate characters when the upper limit value N and lower limit value M obtained from the recognition candidate character selection upper and lower limit table 10 are set to N=& and M=j, respectively, for the recognition candidate characters in FIG. This is an example of recognition candidate characters selected by the character selection means. In other words, for the input character "God Jtt," as shown in Figure S, there are two recognition candidate characters with a degree of similarity greater than the specified value L (L = /, 2θ), but the lower limit value M is three. Therefore, the recognition candidate characters up to the third rank are selected, and the correct character "Kami" llI is also selected as a result. As for the input character "WA", the specified value L (L=/koO) is shown in Figure 3.
) There are five recognition candidate characters having a degree of similarity equal to or higher than the lower limit value M and the upper limit value N, so the recognition candidate characters up to the fifth rank are directly selected. Regarding the input character "Sumi" 13, as shown in Figure S, there are 9 recognition candidate characters that have a degree of similarity equal to or higher than the specified value, but since the upper limit N is g, recognition is limited to the 1st digit. A candidate character is selected, the number of recognition candidate characters is reduced by one, and the processing time is reduced.

The word name is determined by the word determination means tei using the word dictionary 9g for the recognition candidate characters selected by the recognition candidate character selection means t in this manner. Specifically, the word name is determined by checking whether or not the character combination is registered in the word dictionary, starting with the combination of characters with the lowest total rank of recognition candidate characters.

In addition, although the case of kanji has been described above, the present invention is not limited to this, and can also be applied to katakana and hiragana.

〔Effect of the invention〕

The present invention will be described in detail with reference to the above embodiments.

First, for the input character [God J//, if a recognition candidate character with a degree of similarity higher than the specified value is simply selected, the third candidate character "God"/R, which is out of the selection range, will be recognized according to the present invention. By applying this, it is possible to correct the character to 5 so that it is selected as a recognition candidate character, and it becomes possible to obtain correct recognition results. In addition, by applying the present invention, the number of recognition candidate characters for the input character [J/3]
It is possible to reduce the number from 9 to g, which has the effect of reducing the processing time. As described above, by using the present invention, there is an advantage that processing time can be improved without lowering the word recognition rate when reading words.

Furthermore, since the specified values are stored in the recognition candidate character selection similarity table, there is an advantage that the specified values can be optimally set for each type of word such as surname, first name, address, company name, etc.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of the configuration of a word reading method using a conventional method, FIG. 1 is a diagram for explaining a word determination method in the conventional word reading method, and FIG. A block diagram showing an example of the configuration of a word reading system according to an embodiment, Figure 5 is a diagram showing an example of output from the character recognition means, and Figures 5 and 6 are diagrams for explaining the recognition candidate character selection means. be. In the figure, 3 is a character recognition means, Hiroshi is a word determination means, S is a word dictionary, is a recognition candidate character selection means, is a recognition candidate character selection similarity table, and 10 is a recognition candidate character selection upper and lower limit table. be. Note that the same or equivalent parts in the figures are indicated by the same reference numerals. Coin 1 figure leather 2 figure flame 3 figure flame 4 figure leather 5 figure 6 figure procedural correction band (self-produced) 1986. It January 5th

Claims (3)

[Claims] (1) Recognize the characters that make up a word one by one, select a plurality of recognition candidate characters for each character with a degree of similarity indicating the degree of certainty that it is a character, and select a number of recognition candidate characters that have a similarity of more than a specified value among the recognition candidate characters. If the number of characters with a degree of similarity is less than the predetermined lower limit number, the recognition candidate characters are selected in order of the characters with the highest similarity by the predetermined lower limit number, and if the number of characters with a degree of similarity equal to or greater than the predetermined value is greater than the predetermined upper limit number. select the predetermined upper limit number of characters in descending order of similarity, and if the number of characters with similarity greater than the specified value is greater than the lower limit number and less than the upper limit number, select the predetermined number of characters with similarity greater than the specified value. 1. A word reading method characterized in that only characters having the same character are selected, and word names are determined from a word dictionary for the selected recognition candidate characters. (2) The word reading method according to claim 1, wherein characters having a degree of similarity equal to or higher than a specified value among the recognition candidate characters are selected from a recognition candidate character selection similarity table. (3) The word reading method according to claim 2, wherein the predetermined lower limit value and upper limit value are selected from a recognition candidate character selection upper and lower limit table.