JPS5848180A - Character decision processing system - Google Patents

Abstract

PURPOSE:To read a character string, which contains different kind of characters whose forms are similar, with high precision by previously inputting information on whether two character strings are adjacent to each other or not through a form, and then counting the frequencies of those two character strings during reading operation. CONSTITUTION:Information on whether two character strings are adjacent to each other or not is inputted previously through a form, and during reading operation, the frequencies of those two character strings are counted to select a possible character for an input character. For example, the form is scanned through a scanning part 20 under the command of a control part 10, and the picture signal of the input character obtained by photoelectric conversion is transferred to a recognition part 30 to perform recognition processing, thereby sending the code of the candidate character for the input character to the control part 10. Then when the input form has underfined front and back relation, the readout result is written in a front and back relation table storage part 60 through an interpretation part 50 for the form having underfined front and back relation. When the input document is a form to be read, the candidate character is sent to a decision part 70, whose decision result is sent out through an output line 72; and a character string frequency counting part 80 counts the frequencies of those two character strings.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character determination processing method, particularly a character determination processing method in a character reading device that reads a form in which a plurality of character types such as kana, alphabetic characters, and numbers are written. It is a stone related to.

If you mix different types of characters such as kana, alphabets, numbers, etc., a set of characters that are very similar to each other, for example, kana 1 k 1
There are many alphabetic characters, 111, numbers 171, and kana characters such as 1 and 1. When reading a character string containing a mixture of different characters, conventional character reading devices emphasize the differences between these similar characters. For example, in the case of the number 07'', the right line segment is convex to the left. The reading accuracy was ensured by writing the letters in a curved manner, and in the case of kana no ri 6, making them curve convexly to the right.

However, with this method, the psychological burden on the scribe increases, and at the same time, it is necessary to examine minute differences in character shapes during recognition processing, so the recognition logic circuit is large and the equipment is expensive. There was a drawback that it became a thing. In response to this problem, a method has been proposed in which a dictionary of words to be read is prepared in advance, the reading results are compared with the dictionary, and the reading results are corrected to improve reading accuracy. However, with these methods, when the number of target words is large, the storage capacity of the dictionary becomes large, which makes the entire device expensive, increases the processing time required for matching, and reduces the reading speed. This method has drawbacks, such as the fact that the procedure for creating a dictionary is cumbersome because it is necessary to create a dictionary according to the type of words to be read.

The present invention is a character string KTh containing a mixture of letters, letters, and numbers.
This method focuses on the fact that when different characters are adjacent, their combinations are often limited to some extent. (referred to as relationships) using a form; and 1. In contrast to counting the S* of two character strings during reading, we provide a character determination all method that stores context information within the character reading device and uses that information to select candidate characters for input characters. The purpose of this is to easily store information that is useful for making judgments in the device without going through the troublesome procedures required in the conventional method, and to use that information to identify characters that are similar in shape. The objective is to read with high accuracy a character string containing characters of different characters.

FIG. 1 shows an embodiment of the present invention, in which 10 is a control section, 2
0 is a scanning unit, 30 line recognition unit, 40 is an operation unit with a display, 50 is a context definition form interpretation unit, 60 is a context table storage unit, 70 is a determination unit, 80 is a character string frequency counting unit, 71 is a Connection with the control unit @ 72 is an output line.

In operation, in step 5, the scanning section 20 scans the form according to a command from the control section 100, and transfers the image signal of the input character obtained by photoelectric conversion to the recognition section 30. Followed by 'i! 1#It section 30
performs recognition processing (according to a command from the control unit 100), and sends ;-, a candidate character for the input character, to the control unit 10.

Here, if the character code in the form type field of the input form indicates that the input form is a context definition form, the control unit 10 converts the read result of the form into a context definition form. It is sent to the interpretation section 50.

At this time, if necessary, the control section 10 sends the reading results to the operation section 4.
The operator can check the reading result 9 on the display and correct the error/reject character ' by using the operating section and the keyboard 40. The corrected character code is sent to the context definition form interpretation section 50.

The context definition form interpretation unit 50 interprets the contents of the context definition form and writes the information about the context regarding the two character strings written on the form into the context table in the context table storage unit 60.

If the character code written in the form type field of the input form indicates the form to be read, the control unit IO converts the character code that is a candidate for the input character into a unit character string (
For example, each character string of one field is sent to the determination unit 70. The determination unit 70 scans the candidate character code of the unit character string, and based on the reading results of the characters located before and after the input character to which a plurality of candidate characters correspond, the determination unit 70 reads the context table storage unit 60. The candidate characters that can be adjacent to the characters before and after the input character are determined by referring to the contents of the context table stored in the input character.

When the unit character string valve processing is completed, the determination unit 70 outputs the determination result to the output line 72.

String! The frequency counting unit 80 counts the frequency of two character strings based on the determination result of the single blood character string sent from the determination unit 70. At this time, for 92 character strings whose frequency exceeds a certain value, the content corresponding to the two character strings in the context table stored in the context table storage unit 60 is added to the adjacent possible information. Rewrite it as

The configuration of the present invention is to input a context definition form and count the appearance frequency of two character strings when reading a document to be read, thereby accumulating the context relationship of characters in a front and rear table in a character reading device. It uses the accumulated context information to determine the appropriate candidate character from among multiple candidate characters), and the mixed characters of multiple character types that have many characters that are similar in glyph shape are given higher priority. Can be read with precision.

Figure 2 shows an example of a context definition form, with 51
O is a form type field, 52o is a previous character group specification field, 530Fi is an adjacent information specification field, 54
0 is the rear character group designation field.

・The form capital punishment field is a field provided to indicate whether the form is a context or definition form, and if a predetermined character or character string is entered in this field, the form This means that it is a relationship definition form.

The preceding character group designation field 520K is filled with the character group located before, and the subsequent character group designation field 540 is filled with the character group located after. The adjacent information specification field 530K is
Preceding character group specification field 52G written in the relevant gender
A character symbol indicating whether or not the characters inside and the characters in the rear character group designation field 840 can be adjacent to each other (in Fig. 2, -1 stirrup is a symbol indicating that they can be adjacent to each other, and '-1 is a symbol that indicates that they cannot be adjacent to each other. ) is entered.

For ease of description, fielding to specify character groups uses a predetermined order of round characters, such as 0, 1.2...9 in numbers, or aiweoori in kana, and uses, for example, the sentence " When specifying a character group from the letter C. to the letter Cmt, it is assumed that the description 1 (C□-Csy- is allowed).

The example in Figure 2 means that in the second line, a kana is not adjacent to the number, and in the third line, the kana -riichi 9M chi is allowed after the number. Indicates that Therefore, the second and third lines indicate that only kana 1, 1, 1, and 1 can be adjacent after the number. The context definition form folding unit 50 functions to write descriptions onto the context definition form.

FIG. 3 shows an example of a context table.

As shown in FIG. 3, the context table contains information about whether or not two characters can be adjacent to each other, such as kana, alphabets, numbers, and symbols, such as 1, 0, etc. It is shown in three values of g.

In this ζ, α is a number that satisfies 0 (gi After a symbol, a kana character indicates that there are no adjacent alphabetic characters, and after a symbol, a kana character indicates that it is unknown whether or not they are adjacent.

In the context table, as an initial value, l is assigned to a combination of the same character type (for example, kana and kana, or alphabetic character and alphabetic character, etc.), and - is assigned to a combination of different character types k. do. This initial value setting reflects the fact that in a character string, the probability that 52 adjacent characters are of the same character type is higher than the probability that they are of different character types.

FIG. 4 shows the configuration of the determination section.

710 is a candidate character register, 720 is a reject character detection circuit, 730 is a context table address generation circuit,
740 is an addition circuit, 750 is a maximum value detection circuit), 7
1 is a connection with the control section, and 72 is an output line.

The purpose of the operation of the judgment section is to identify the person to whom multiple candidate characters correspond,
The purpose of this method is to correct rejects (for input characters, by selecting from candidate characters the character that is most likely to be adjacent to the characters before and after the mosquito input character). Codes of candidate characters of the unit character string sent from the input line 71 are sequentially stored in the candidate character register. Next, the reject character detection circuit 720 scans the register and detects a certain reject character C4 that is rejected because a plurality of (-) candidate characters C%, . . . , C; Detect the position of.

1. Next, if the reading results of the characters located before and after the character position are not rejected, the candidate character C (-Isc=
+11- Read from the candidate register, 2s set of 2 character strings C(-s C3s C'4 C4+x, C(-s
C"4 s C'404+skei..., C(-m c
FCI C(+* is sequentially sent to the context table address generation circuit 730 jc.

The context table address generation circuit 73G generates two character strings C C i IC), C) C i+1 . . . . Sequentially generates the addresses of the context tape in which context information regarding #C file C1+1 is stored, and sends the address information to the connection 7311c with the context table storage unit. Context information R(C4-1 bow)*R(C3C(
+m) e...

R (C bow C4 + 1) (Each value is either O# a # 1, and the number that satisfies the above-mentioned 0 × a × ÷) is the addition circuit 740
K is input.

1゛ Addition circuit 740 each candidate character C: (&ml, ・
... regret), R(C4-m ci) + R(debt C(+
x) and sends the calculation result to the maximum value detection circuit 750. The maximum value detection circuit 750 has R(C4-1ci)+R(
Select CI, o, for which OiC4+t) is the maximum, and write C4° at the corresponding character position in the candidate character register. Therefore, the previously written OS...cl is rewritten as epic.

When the above processing is completed for all characters in the candidate character register that have multiple candidates, the code of the next candidate character finally obtained is outputted to the output line 72. At this time, the character C4 whose one-complementary characters have not been squeezed out until the end is output as a diectokud. Figure 6 shows the configuration of the character string frequency counting circuit. -810 is a shift register, 820 is a work.

memory, rear address generation circuit, 830 is working memory,
840 is an addition circuit, 850 is a comparison circuit, 72 is a connection with the determination unit 70, and 81 is a context table storage unit 60.
This is the connection.

The purpose of operation of this character string frequency counting circuit is to count the occurrences S* of two literary strings, and for two character strings that exceed a certain value, determine that two empty characters can be adjacent to each other, and determine the context. The content of the corresponding location of the table can be adjacent (i.e. 1) K-
It's about changing.

The character code sent from the - connection 72 is sequentially stored in the shift register 810, and the shift register 810 always holds the ;- code for two characters. Working memory address generation circuit 820K
The sending 0 working memory 830 is a memory that stores the appearance frequency of each two character string, and the working memory address generation circuit 820 recognizes that all of the 92 characters of codes that are sent are reject codes. If not, a corresponding address is generated for the two character strings in the working memory 830, and address information is sent to the working memory 830. The read contents of the working memory 880 are sent to the adder circuit 840K, which adds 1 to the contents, writes the addition result to the original address of the working memory, and sends it to the comparator circuit 85K.
Also send to 0.

The comparison circuit 850 determines whether the addition result exceeds a predetermined color value, and if it exceeds the desired color value, the comparison circuit 850
Writes 1 to the address of the context table that corresponds to the character string.

As explained above, according to the present invention, the context of characters in a document to be read is inputted into a context definition document, and the frequency of appearance of two character strings is counted when reading the document to be read, so that the character strings can be recognized. Data can be stored in an embedded context table and this information can be used to determine the correct candidate character from a pool of candidate characters. In addition, the above-mentioned character judgment processing method can achieve high accuracy without increasing the scale of recognition logic in reading characters with mixed character types such as kana, alphabets, and numbers, which have many morphologically similar characters. It is possible to realize the reading function.

[Brief explanation of the drawing]

Figure 1 is a boot diagram of one embodiment of the present invention, Figure 2 is a configuration diagram showing an example of a context definition form; Figure 13 is a configuration diagram showing an example of a context table, and Figure 4 is a configuration diagram showing an example of a context definition form. FIG. 5 is a block diagram of one embodiment of the determination section. FIG. 5 is a block diagram of one embodiment of the character string frequency counting section. In the figure, 10 is a control unit, 20 is a scanning unit, 3G is a recognition unit, 4
0 is the operation unit, sO is the context definition form interpretation unit, 6G is the context table storage unit, 70 company determination unit, ’71 is the connection between the control unit and other parts, 72 is the output line, and 80 is the character The column frequency counter is shown. Patent applicant Nippon Telegraph and Telephone Public Corporation Patent attorney Mori 1) Kansai 2 eyes 3 figures 4 A t 5 figures

Claims (1)

[Claims] (1) In a character reading method that optically reads forms written in a mixture of characters of multiple character types, a recognition means capable of outputting multiple candidate characters of =- for one input character. , reads the context definition form that describes the arrangement of two character strings in the document to be read, and creates a context table that indicates whether or not pairs of two consecutive characters are allowed to be adjacent to each other. and means for selecting a character that can be adjacent to a previous or subsequent character from among candidate characters for an input character based on information in the context table. (2. In the character determination processing method recited in claim 4, item 1, the means for creating a context table reads the characters on the input form and at the same time, from the reading result, the two characters appearing on the form - 1 A character determination processing method comprising means for counting the appearance frequency of the two character strings and rewriting the contents of a context table (according to the appearance #JR) of the two character strings.