JPS60147888A - Postprocessing of character recognition and device - Google Patents

Abstract

PURPOSE:To shorten retrieving time of a word dictionary by setting preliminarily an optimum index character position in accordance with the number of characters of a word to be registered in a word dictionary, using a character in correspondence to an optimum index character position of a candidate word string as an index character, and retrieving a word of the same number of characters as that of a candidate word from a word dictionary. CONSTITUTION:It is assumed that an input word is ''KEISANKI'' having the number of letters of an input word n=3. candidate characters ''tei'', ''kei'', ''tou'', ''kyo'' and ''so'' for ''kei'', those of ''sen'', ''toku'', ''tou'', ''ga'' and ''bo'' for ''san'' and those of ''ki'', ''so'', ''ou'', ''ran'' and ''sei'' for ''ki'' are supplied. In this condition the value in accordance with the number of characters of a candidate word (n=3) is stored in a number of characters of a word register 7a. When it is transmitted to an index character reading circuit 7c, it reads out sequentially respective candidate characters of a word ''sen'', ''toku'', ''tou'', ''ga'', ''bo'', in correspondence to an optimum index character position, m=2 as an index character from a candidate word buffer 5 refering to an index character table 7b.

Description

Detailed Description of the Invention (1) Technical Field of the Invention The present invention relates to a character recognition post-processing method and apparatus for performing character recognition processing on once recognized characters in order to prevent erroneous character recognition during character recognition. Regarding.

(2) Background of the technology 7 In general, character recognition devices observe characters written on paper, etc., extract the characteristics of the characters, and then compare the characteristics of the characters with the contents of a recognition dictionary to identify the characters. It is designed to identify.

However, if the characters written on a sheet of paper or the like are disordered9 or have curly characters, there is a risk that the characters may be misrecognized.

(3) Prior art and its problems Therefore, in the past, as a character recognition post-processing method,
There is a method that assumes that the recognition target is a word consisting of multiple characters, and re-recognizes the recognized characters word by word, thereby preventing misidentification of characters as much as possible. . That is, in the above post-processing method, a word dictionary is created in advance in which a large number of words including the input word are arranged in a searchable manner by the first character, and each character of the input word is recognized and the words are ranked in order of similarity. The candidate character strings selected in order are divided into word units to form a candidate word string, and the optimal word is selected from the word dictionary by comparing this candidate word string with the contents of the word dictionary. It is.

However, depending on the number of characters in a word, a situation may arise in which the number of words starting from a certain first character becomes extremely large. In such a case, □In the conventional character recognition post-processing method, the word dictionary is always searched by the first character of the word, so the search time for the word dictionary increases, and This causes a problem that the efficiency of character recognition post-processing becomes poor.

(4) Purpose of the Invention The present invention has been made based on the above-mentioned viewpoints, and its purpose is to shorten the search time of a word dictionary in character recognition post-processing. An object of the present invention is to provide a character recognition post-processing method and an apparatus thereof that improve efficiency.

(5) Structure of the Invention The basic structure of the character recognition post-processing method according to the present invention is to divide candidate character strings obtained by recognizing each character of an input word consisting of a plurality of characters into word units, and to generate candidates. As a word string,
When comparing this candidate word string with the contents of the word dictionary and selecting the optimal word from the word dictionary, the electronic index character position is set in advance according to the number of characters of the word to be registered in the word dictionary. In the word dictionary, registered word groups are arranged so that they can be searched by the characters at the optimal index character position, while the characters corresponding to the optimal index character position of the i-complement word string are used as index characters to search for the same number of characters as the candidate word from the word dictionary. The problem is that the search is made for the single word □. Furthermore, the basic configuration of the device for carrying out the above method includes an accumulation means for accumulating a candidate character string obtained by recognizing each character of an input word consisting of a plurality of characters as a candidate word string divided into vulgar words. , a word dictionary in which a group of registered words are arranged in a searchable manner according to the characters at the optimal index character position set in advance according to the number of characters of the word to be registered; a search means for searching for a candidate word from the word dictionary using the character corresponding to the optimum index character position as an index character; and selection means.

(6) Embodiments of the Invention Hereinafter, the character recognition post-processing method and apparatus thereof according to the present invention will be explained in detail based on the embodiments shown in the accompanying drawings. 1 is a block diagram showing a character recognition device to which a character recognition post-processing device according to the present invention is attached. □In the same figure, a character recognition device d, an observation unit 1 that optically reads each character of an input word written on a paper etc., photoelectrically converts the optical signal and outputs it, and a character figure output from this observation unit 1. □ An identification unit that extracts the characteristics of the input character based on the electrical signal representing the character. □ An identification unit that identifies characters in the recognition dictionary as input characters in descending order of similarity by comparing the characteristics with even standard features. 4
It is equipped with The character recognition post-processing device also includes a candidate word buffer 5 that stores each candidate character string identified by the identification unit 4' for each character of the input word as a candidate word string divided into word units; It is equipped with a word dictionary 6 that stores words including words in advance, and a word matching section 7 that matches the candidate word string with the contents of the word dictionary 6 and selects an optimal word from the word dictionary.

In this embodiment, the candidate word buffer 5 has a second
As shown in the figure, registers with the number of bits corresponding to the maximum number of characters of a human word are set for the number of candidate characters obtained from the identification unit 4 (
For example, the candidate character for the 1st (i=1.2.3)th character of the input word (number of characters n1, for example, 3) is stored in the i-th address of each register 5a to 5e from the upper register 5a. The candidate words are stored in rows with high similarity through the lower registers 5e, and the candidate words are arranged in each of the registers 5a to 5e.

Further, as shown in FIG. 2, the word groups registered in the word dictionary 6 are arranged so as to be searchable by the character at the optimum index character position. The optimum index character position is set in advance according to the number of characters of the word to be registered, and is an index character that can suppress the number of search words to the minimum value when searching for a word with a predetermined number of characters from the word dictionary 6. It shows the position of In this example, the number of characters in the input word is 1
The optimum index character position m=2 is set for 1=3.

Furthermore, as shown in FIG. 2, the word collation unit 7 includes a search means for searching the word dictionary 6 for a word having the same number of characters as the candidate word, using the character corresponding to the optimal index character position of the candidate word string as an index character. and a selection means for selecting the most suitable word from among the words searched from the word dictionary 6. In FIG. 2, 7a is a word character number register that stores the number of characters of a candidate word in the candidate word buffer 5, and 7b is an index character table that stores the optimal index character position m for the number of characters n of the word to be registered. 7d reads the contents of the word character number register 7a and reads the index characters from the candidate word buffer 5 by referring to the index character table 7b; The read address table 7e storing addresses is a word dictionary read control circuit which learns the addresses in the successive address table 7d from the contents of the word character count register 7a and the index character read circuit 1c, and sequentially reads words from the word dictionary 6. These constitute the above-mentioned search means. Further, reference numeral 7f is a word register that stores all the words read out from the word dictionary 6, and 7g is a similarity calculation circuit, in which each character of the word stored in the word register 7f is a candidate, and the word buffer 5 The method is to check whether the character is in the first rank of each candidate character string, and add this value for each character.If a certain character of the word stored in the word register 7f exists in the corresponding candidate character string, this value is added for each character. If not, the character will be given a higher ranking than the number of candidate characters. 7h is the similarity calculation circuit 7g #
T1 is a similarity register that stores the similarity calculated by 1, T1 is a minimum similarity register that updates and stores smaller values among the similarities sequentially stored in similarity register 7h, and 7j is the minimum similarity. an optimal word register that updates and stores words with a degree of similarity stored in the value register 7i;
7 is a comparator, and when the value of the similarity register 7h is smaller than the value of the minimum similarity register 71, the value of the brightness similarity register 7h is input.

The minimum similarity register is stored in the staff 1, and the contents of the word register 7f are stored in the optimum word register 7j. The word register 7f to the comparator 7 are configured with the selection means.

are doing.

Next, the operation of the character recognition post-processing device according to this embodiment will be explained. Now, suppose that the input word is "calculator" with character count n=3, and from the identification unit 4, for the input character "kei",
One character is given as a candidate for ``, revise,''``kei,''``shu,''``shu,'' and ``sue.''
Candidate characters for ``Ka'' and ``Book'' are given, and the input characters [Ki,! Suppose that candidate characters are given for ``ki'', ``iso'', ``horizontal'', ``ran'', and ``sei''. 1 In this case, each candidate character of each character is stored in the candidate word buffer 5 in order from the upper register 5a to the lower register 5e according to a predetermined order, and the candidate word "calculator"・0・ is stored in each register 5a to 5e. It will be stored. In this state, the word character number register 7a stores a ridge (3 in this case) corresponding to the number of characters of the candidate word, and when this value is sent to the index character reading circuit 7c, the index character table 7b is referred to. The index character reading circuit 7c corresponds to the optimum index character position m=2 from the candidate word buffer 5. The characters "san", "atsu", "etc.", "ga", and "book" of each candidate word are sequentially read out as index characters. Then, the contents of the index character and word character count register 7a are sent to the word dictionary reading control circuit 7e, and the contents of the index character and word character number register 7a are sent to the word dictionary reading control circuit 7e, and the contents are sent to the word dictionary reading control circuit 7e with reference to the reading address table 7d.
e sequentially reads words from the word dictionary 6 that have three characters and whose 29th character is "san", "atsu", "etc.", "ga", and "book". At this time, since the above-mentioned index characters are set to the optimal ones, the search time of the word dictionary 6 is limited to the search time for the word dictionary 6, such as the first character "edited □"... or the third character " It is shortened compared to the case where "machine"... is used as the index character.

If the word read out from the word dictionary 6 is, for example, "calculator," the word is stored in the word register 7f, and the similarity of the word to the candidate word is calculated by the similarity division circuit 7g. Ru. In this case, the first character of the above word "ke" is located in the second position of the candidate string, and the second and third characters "san" and "ki" of the above word are located in the first position of the candidate character string. Since it is located at , the similarity calculation circuit 7g
calculates 2.+1+1 and stores the calculated value 4 as the similarity in the similarity register 7h. In this state, the content of the previously stored minimum similarity register 71 is larger than the content of the similarity register 7h since the word "calculator" does not exist, so the comparator 7 As a result, the contents of the similarity register 7h are stored in the minimum similarity register 71, and the contents of the word register 7f are stored in the optimum word register 7j. Thereafter, the words read from the word dictionary 6 are sequentially stored in the word register 7f, and the similarity of each word is calculated and stored in the similarity register 7h.
Since the similarity of each word is greater than the similarity of the word "calculator", the lowest similarity value register 71
The contents of the optimum word register 7j are not updated and retain their original contents. In this way, when all the words have been searched, the minimum similarity value is finally stored in the minimum similarity register 7i', and the word corresponding to the minimum similarity value is stored in the optimal word register 7j. "Calculator" is selected, and the optimal word "Calculator", which is the final recognition result of the input word, is read out from the optimal word register 7j.

The specific configuration of the character recognition post-processing device is not limited to that shown in the above embodiments, but may be created using a single microprocessor, etc. You may change the design as appropriate. -□ ゛(7) Effects of the invention □ “As explained above, according to the character recognition post-processing method and device according to the present invention, the character recognition post-processing method and the device can be used for character recognition post-processing. The search time for the Oiso word dictionary can be reduced compared to the conventional method, and the processing efficiency can be improved accordingly. □

[Brief explanation of drawings]

Fig. 1 is a block diagram showing an example of a system attached to the character recognition post-processing device i character recognition device according to the present invention;
The figure is a block diagram showing one embodiment of the character recognition post-processing device according to the present invention.5...Candidate word buffer (storage means)6...Word dictionary''

Claims (1)

[Claims] l) Recognizing each character of an input word consisting of a plurality of characters,
The candidate character strings are divided into word units to form candidate word strings, and this candidate word string is compared with words and the contents of the dictionary to select the most suitable single from the word dictionary. The optimal i index character position is set in advance according to the number of characters of the candidate word, and the registered word groups are arranged in the word dictionary so that they can be searched by the characters at the optimal index character position, while the optimal index of the candidate word string is A character recognition post-processing method characterized in that a word having the same number of characters as a candidate word is searched from the word dictionary using a character corresponding to a sentence position as an index character. □ 2) A storage means for accumulating candidate character strings as candidate word strings partitioned into ``Kogogo'' units by recognizing each input single character consisting of multiple characters, and a storage means for storing candidate character strings as candidate word strings divided into ``Kogogo'' units and a word dictionary in which a group of registered words are arranged in a searchable manner according to the characters at the optimal index character position set in advance; a search means for searching for a word having the same number of characters as the candidate word in the word dictionary using the character as an index character; and a selection means for selecting an optimal word from among the words searched from the word dictionary. A character recognition post-processing device.