JPS59117673A - Postprocessing system of character recognizing device - Google Patents

Abstract

PURPOSE:To shorten a postprocessing time by taking notice of the characteristics of a Japanese sentence repeating the same word or expression in one sentence to utilize the result read and recognized beforehand as a dictionary. CONSTITUTION:When a form A is recognized as a form B and the primary recognition result of the form B is C under this condition, it is checked whether the same character exists on a result D or not with respect to the character ''te'' (hand) of a candidate (1', 1). It is found that this character coincides with D10 of the result D. Next, when a candidate (2', 1) and D11 are collated with each other, they do not coincide with each other. Therefore, a candidate (2', 2) and D11 are collated with each other, and the second recognition result is judged to be ''sho'' (writing) because they coincide with each other. Similarly, the third character is recognized as ''ki'' (part of ''writing''), and a candidate (4', 1) is not equal to D13 and a candidate (4', 2) is not equal to D13, and thus, a series of processings are terminated up to the third character, and a new processing is started from the fourth character.

Description

Detailed Description of the Invention (A) Technical Field of the Invention The present invention relates to a post-processing method in a character recognition device, in particular, to output a plurality of candidates by a recognition means, and then select a true recognition result from among the candidates. When selecting post-processing,
The present invention relates to a post-processing method in a character recognition device that uses previously recognized results as a dictionary.

(B) Technical background and problems Traditionally, post-processing as described above has been carried out (1).
When the recognition target is limited, such as an address base or a product name, there is a method that uses the address base as a dictionary and performs post-processing on a word-by-word basis, or (11) a general method that uses a Japanese dictionary, for example, and Methods of processing are known.

However, the former method can only be used within a predefined range of words, and the latter method requires a large amount of memory to store the Japanese dictionary, and the post-processing time is extremely long. Become a dog.

(C) Object and Structure of the Invention The purpose of the present invention is to efficiently perform the above-mentioned post-processing. A character recognition device comprising: a means for extracting features of the character; and a recognition means for comparing the extracted features with the contents of a dictionary registered in advance and sequentially outputting recognition result candidates, In addition to providing a post-processing mechanism for correcting the recognition results, the post-processing mechanism also includes a result memory that stores a plurality of previously recognized character groups as a text dictionary, and a result memory that stores the newly extracted recognition results. a post-processing unit that checks whether the candidate matches any character in the content of the result memory based on the concept of longest match and obtains a true result from the recognition result candidate based on the matching character; It is characterized in that the true result is obtained from newly extracted recognition result candidates using the previously recognized results. This will be explained below with reference to the drawings.

(D) Embodiment of the invention Figures 1 (A) and (B) are examples in which a series of Japanese sentences are written on multiple forms, and Figure 2 shows the recognition and correction of the forms shown in Figure 1 (A). An example of the results shown in Figure 3 is shown in Figure 1 (
B) An example of the raw result of recognizing the illustrated form by the recognition means, FIG. 4 is an example of the result after post-processing based on the present invention, FIG. 5 is an example of the configuration of an embodiment of the present invention, and FIG. 5 shows an example process executed in the post-processing section shown in FIG.

Generally, in the case of Japanese sentences, the same word or expression may appear repeatedly within the sentence. In the present invention,
I am trying to take advantage of this point.

The following explanation assumes that there are forms as shown in Figures 1 (A) and (B), and that these are to be recognized. And, Figure 1 (A)
Assume that the form shown in is recognized as shown in FIG.

It is assumed that the result shown in FIG. 2 is a correct result obtained by correcting the raw result recognized by the recognition means according to the present invention by the display correction unit.

In this state, when recognizing the form shown in FIG. 1(B), let us consider a case where the result of raw recognition by the recognition means is that shown in FIG. 3. Note that for each character, in the illustrated case, a first candidate and a second candidate are obtained.

The following candidates are specified in the form K (■A), which is
FIG. 1(B) indicates that this is the 4th candidate for the No. 1 character on the illustrated form. In addition, the true recognition result on the form shown in FIG. 1 (A), that is, the recognition result shown in FIG. Indicate the result.

First, for K(■, 1), it is checked whether or not the same character exists based on the results shown in FIG.

In the illustrated case, K (■, 1) is the character "hand". In the case of K(■, 1), it matches Dlo.

Next, take K(■, 1)8. In this case, they are not equal. Therefore, K(■, 2) is compared with DI+. In this case, since they are equal, the 0th recognition result is considered to be "calligraphy". Similarly, the 0th character is ``ki''.

Next, K(■, 1)Nha3 and K(■, 2β mountain.

Therefore, the series of processing ends up to the 0th processing, and a new processing starts from the 0th processing.

Find something that is equal to Tsume D, K (■, 1) from the results shown in Figure 2. In this case, since there is no match for K(■, 1), we next search for the same one as K(■, 2). The result is K(■, 2) = D. o
is found, and K(■, 1) = D,. .

is obtained, and the 0th character is assumed to be "Kan". Furthermore, K([stock],') = D++o, but K(
■,' ) 'ED+20 K(0,2)
Since '= DI20, the 0th character is not modified.

In this way, the result shown in FIG. 4 is obtained.

FIG. 5 shows the configuration of an embodiment of the present invention.

Reference numeral 1 in the figure is an observation unit, which reads the forms shown in FIGS. 1(A) and 1(B). Reference numeral 2 denotes a feature extraction unit (means), which extracts the features of each character, as is conventionally known. 3 is a recognition dictionary in which features of standard characters and their category names are stored.

Reference numeral 4 denotes a recognition unit (means) that compares the character features supplied from the feature extraction unit 2 with the standard character features read out from the recognition dictionary 3, and converts each character into the characters shown in FIG. In the example, the first candidate and the second candidate are output. 5 is a candidate memory, as shown in FIG.
The candidates corresponding to the th character are stored respectively. Reference numeral 6 denotes a display correction control section which reads a plurality of candidates for each character from the candidate memory 5 and displays them on the display/correction section 7, for example, regarding the first character on the form. Further, 07, which stores the true result obtained based on the above candidates (as shown in FIG. It also displays the results of corrections, and allows the operator to select the correct characters if necessary. 8 is a post-processing unit that processes the contents of the candidate memory 5 into a result memory 9;
09 is a result memory (text dictionary) that performs post-processing in the form described in connection with Figs. 2 to 4 based on the contents of It is stored as shown in the figure and used for post-processing.

The processing in the post-processing section 8 shown in FIG. 5 has been described with reference to FIGS. 2 to 4, but in more detail,
Processing is performed as in the embodiment shown in FIG. 6th
The numbers shown in the figure correspond to the numbers such as ■, ■, etc. shown in Figure 3, and J corresponds to 112, ..., 12 shown in Figure 2.
0, which corresponds to a number such as 0, corresponds to a route in which matching is performed based on the second candidate.

(E) As described in detail, according to the present invention, the capacity of the dictionary required for post-processing is limited to the unit of a series of sentences, compared to having a general Japanese dictionary. Since the 0"if requires only a very small capacity, the processing efficiency is also excellent.Furthermore, it is possible to perform corrections not in units of words but in units of sentences, etc., greatly improving the accuracy of post-processing.

[Brief explanation of drawings]

Figures 1 (A) and (B) are examples where a series of Japanese sentences are written across multiple forms, Figure 2 is an example of the result of recognizing and correcting the form shown in Figure 1 (A), and Figure 3 is an example of the result of recognizing and correcting the form shown in Figure 1 (A). The figure is Figure 1 (
B) An example of the raw result of recognizing the illustrated form by the recognition means, FIG. 4 is an example of the result after post-processing based on the present invention, FIG. 5 is an example of the configuration of an embodiment of the present invention, and FIG. 5 shows an example process executed in the post-processing section shown in FIG. In the figure, l is an observation unit, 2 is a feature extraction unit, 3 is a recognition dictionary, and 4
5 is a recognition unit, 5 is a candidate memory, 6 is a display correction control unit, 7 is a display/correction unit, 8 is a post-processing unit, and 9 is a result memory (text dictionary).

Claims (1)

[Claims] Character input means for inputting characters to be recognized, feature extraction means for extracting features of the characters, and recognition for sequentially outputting recognition result candidates by comparing the extracted features with the contents of a dictionary registered in advance. A character recognition device comprising: a post-processing mechanism for correcting recognition results; It is checked whether the newly extracted recognition result candidate matches any character in the contents of the result memory based on the longest match concept, and based on the matching character, the back of the recognition result candidate is extracted from the recognition result candidate. Post-processing in a character recognition device, characterized in that it has a post-processing unit that obtains results, and uses previously recognized results to obtain true results from newly extracted recognition result candidates. Method 0