JPH04270482A - Printing character recognition device - Google Patents

Abstract

PURPOSE:To easily and accurately correct a character even when the character as a recognizing object is erroneously recognized by mechanical similarity judgement. CONSTITUTION:A deciding circuit 17 is provided to output the character code of the similar character as a clot pattern to the recognizing object character while referring to a memory 12 for rough classification dictionary and a memory 15 for precise classification dictionary, a display device 3 is provided to display the source dot pattern of the recognizing object character and the character similar to the recognizing object character, and a similar character table 19 is provided to arbitrarily record the character code and the attribute of the character having this character code by groups.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a printed character recognition device for recognizing, for example, printed characters.

0002

[Prior Art] There is a rapidly growing movement to efficiently utilize articles such as newspapers and books by digitizing them and creating a database. Character recognition devices that can encode characters are being developed. Some of these character recognition devices select candidate characters for characters to be recognized using a memory for a large classification dictionary that stores rough feature values and corresponding character codes, and select candidate characters that correspond to the character codes. Dot pattern (
The pattern of the character to be recognized is compared with the pattern of the candidate character using the memory for the subclassification dictionary that stores the character font data), and when the two patterns are similar, the recognition is performed. There are some that give the character code of the candidate character to the target character.

[0003] If the candidate character does not exist, or if the pattern of the candidate character and the pattern of the character to be recognized are not similar, the character to be recognized is rejected as unrecognizable. The following methods can be used to correct such rejected characters or erroneously recognized characters. 1. Display other candidate characters with high similarity and select from among them. 2. Modify by converting kana-kanji, inputting codes, or selecting by number of radical strokes.

0004

[Problem to be Solved by the Invention] However, although the candidate characters obtained by referring to the memory for the major classification dictionary are characters that are determined to be mechanically similar, they are not necessarily determined by mechanical similarity determination. Since the target character is not necessarily extracted accurately, the recognition target character may be incorrectly recognized when selecting from among the candidate characters. In such a case, it is necessary to input the character to be recognized through kana-kanji conversion or the like, but this has the disadvantage of being complicated.

[0005] Furthermore, when inputting characters through kana-kanji conversion, for example, the number "1" and the lowercase English letter "l"
or ``ri'' in hiragana and ``ri'' in katakana.
There is a risk of making a mistake. In view of the above, the present invention aims to enable a printed character recognition device to easily and accurately correct a character to be recognized even if the character is incorrectly recognized by mechanical similarity determination. purpose.

[0006]

[Means for Solving the Problems] The printed character recognition apparatus according to the present invention, as shown in FIG. a dictionary storage means (12, 15) for storing the characters in correspondence with each other; and a classification means (10) for referring to the dictionary storage means and outputting the character code of a printed character similar to the recognition target character as a dot pattern. ,11
, 17, 2), a display means (3) for displaying the original dot pattern of the character to be recognized and characters similar to the character to be recognized, and a display means (3) for displaying character codes and character attributes of the character codes for each group arbitrarily. storage means (19) for recorded similar characters; and storage means (12, 1) for the dictionary.
If the recognition result character obtained by referring to 5) does not match the recognition target character, the storage means (19) for similar characters is referred to and the characters of the group to which the recognition result character belongs and this character are stored. Character attributes (kanji, numbers, lowercase letters, hiragana, katakana, etc.) are displayed on the display means (3).

[0007]

[Operation] According to the present invention, characters that are deemed to be similar based on the operator's judgment or operational experience, etc., and the attributes of these characters are registered in advance in the storage means (19) for similar characters by group. I'll keep it. If the recognition result character obtained by referring to the storage means (12, 15) for the dictionary does not match the recognition target character, referring to the storage means (19) for the similar character, Since the characters of the group to which the recognition result character belongs are displayed on the display means (3), the character to be recognized can be easily identified from the group. In this case, since the attributes of each character are displayed together with the characters of the group, there is no chance of confusing the number "1" with the lowercase letter "l", for example.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the character recognition device according to the present invention will be described below with reference to the drawings. FIG. 1 shows the character recognition apparatus of this example. In FIG. 1, 1 is an image scanner, and image data of a printed document read by this image scanner 1 is supplied to a workstation 2. In FIG. This workstation 2 first extracts a character string by projecting the image data in one direction, then identifies the circumscribing frame of each character by projecting within this character string, and identifies the dots inside this circumscribing frame. An input pattern IP of individual characters consisting of a pattern is supplied to a character recognition board 5 via an input/output interface circuit 4.

In the normal character recognition mode, when there is a candidate character corresponding to the input pattern IP, this character recognition board 5 calculates the residual difference between the candidate character and the pattern of each candidate character and its input pattern, and calculates the degree of similarity. supplies, for example, pairs of character codes and residuals of the top 10 candidate characters to the workstation 2 via the input/output interface circuit 4. When the residual of the highest candidate character is smaller than a predetermined level, this workstation 2 transfers the character code of the highest candidate character to its input pattern IP.
When there is no candidate character or the residual of the highest candidate character exceeds a predetermined level, it is determined that the character cannot be identified and a reject code is assigned to the input pattern.

The workstation 2 also has a character font ROM, and displays the identification result of the input document in the form of characters on the display screen 3A of the display device 3 (see FIG. 2). Finally, a high-intensity square figure, for example, is displayed in the part of the character to which the reject code has been assigned. Furthermore, as will be described later, the workstation 2 can correct erroneously recognized characters using the display screen 3A shown in FIG.

To explain in detail the structure of the character recognition board 5 and its operation in the character recognition mode, in the character recognition board 5 shown in FIG. It is supplied to the normalization processing circuit 7 via the port memory 6A. This normalization processing circuit 7 expands and contracts the input pattern IP to 2
A normalized pattern NP of 4×24 dots is obtained, and this normalized pattern NP is stored in a first-in first-out memory (hereinafter referred to as “FIFO”).
It is abbreviated as "memory". ) 8 and an input pattern set circuit 9 to a subclassification circuit 10. Further, the input pattern IP is supplied to the major classification processing circuit 11 via the second dual port memory 6B, and the major classification processing circuit 11 extracts a rough feature amount from the input pattern IP.

The feature quantities in this example are obtained by encoding the pattern structure of the character portion in the vicinity of each side of a rectangle circumscribing the input pattern IP as sides A to D. The pattern structure near each side is the value obtained by projecting the character pattern from each side onto the observation line several dots away from each side. The number of dots in a section can also be used. In this regard, reference numeral 12 denotes a memory for a large classification dictionary, and in this memory 12 for a large classification dictionary, character codes of characters having the feature amount are stored in correspondence with the feature amount of the character.

The major classification processing circuit 11 refers to its major classification dictionary memory 12 and selects characters (hereinafter referred to as "candidate characters") having the same feature code as the feature code corresponding to the feature extracted from the input pattern IP. ), and store the character codes of these candidate characters in FIFO.
It is supplied to a candidate pattern set circuit 14 via a memory 13. In this example, the candidate characters are the feature code of side A ~
A character whose feature code on the D side is equal to the feature code on each of the four sides of the character to be recognized. Reference numeral 15 indicates a memory for a subclassification dictionary, and this memory 15 for a subclassification dictionary stores a character code and a dot pattern (character font) corresponding to the character code for each character code. The dot pattern is also a normalized pattern of 24×24 dots. For example, if the character code is a JIS code, multiple normalized dot patterns are stored for each character code according to the font to be recognized (Mincho font, newspaper Mincho font, Gothic font, etc.) I'll keep it. If the character code is defined in a format that can also specify the font,
The number of dot patterns corresponding to each character code may be one.

The candidate pattern set circuit 14 extracts the normalized pattern RP corresponding to the character code of the candidate character supplied from the major classification processing circuit 11 by referring to the memory 15 for the subclassification dictionary. , this normalized pattern RP is sequentially supplied to the subclassification circuit 10. This subclassification circuit 10 compares the normalized pattern NP of the character to be recognized with the normalized pattern RP of the candidate character,
A first residual ΔIP obtained by erasing pattern NP with pattern RP and a second residual ΔRP obtained by erasing pattern RP with pattern NP are stored in the FIFO memory 1.
The character code of the candidate character output from the major classification processing circuit 11 is also supplied to the determination circuit 17 via the FIFO memory 16C. This determination circuit 17 determines the first residual ΔIP and the second residual ΔIP.
For example, the final residual ΔRP is obtained by adding the residual ΔRP.

[0015] The closer this residual Δ is to 0, the more similar the character to be recognized and the candidate character are. High similarity. Therefore, the determination circuit 17 pairs the character codes C of the top 10 candidate characters with high similarity and the residual difference Δ of this character code, and sequentially stores them in the FIFO memory 18.
If there is no candidate character, a predetermined reject code is output to the input/output interface circuit 4 via the input/output interface circuit 4. These pairs of character code C and residual Δ or the reject code are supplied to the workstation 2.

19 is random access memory (RAM)
This similar character table 19 is connected to the workstation 2. To explain the memory structure of the similar character table 19, its contents are divided into a registered character code table that always exists and a temporary similar code table that is created from time to time. Figure 4 shows the memory structure of the registered character code table created by the operator based on operational experience, etc. As shown in Figure 4, the data corresponding to each address is stored in the character code, next address, and table number. The table number indicates the number of the group to which the similar characters belong, and the similar characters are characters that the operator has determined to be similar and have divided into groups. In this case, the character code part includes the character codes of all characters and character attributes (e.g. numbers, lowercase English letters, uppercase English letters, commas, hiragana, katakana, full periods, English symbols, kanji,
Character data C1, C2, C3, . . . indicating symbols, etc.) are stored in order corresponding to addresses N1, N2, N3, .

[0017] Then, in the next address part of the data of address Ni, the address of the next character belonging to the group with the same number as the group to which the character of character data Ci belongs is stored, and the table number in that data is stored. The number of the group to which the character of the character data Ci belongs is stored in the part. Specifically, in Figure 4, characters 0, 8, B
, o, g at addresses N1, N3, N6, N7, respectively.
Corresponding to N9, if the character set (0, o) belongs to the group with table number 1 and the character set (8, B, g) belongs to the group with table number 6, then address N9
Address N6 corresponding to character B is written as the next address of the data corresponding to character 8 of 3, and address N6 corresponds to character B.
The address N9 corresponding to the character g is written as the next address of the data corresponding to the character B of 6. Since no other similar characters are registered in the group with table number 6, 0 is written as the next address for the data at address N9. Similarly,
The address N7 corresponding to the character o is written in the address next to the data at the address N1 corresponding to the character 0.

As shown in FIG. 4, in this example, similar characters belonging to the same group are connected by pointers, so when a certain character is specified, it is easy to search for the next character belonging to the same group. can. For example, when character B belonging to the group of table number 6 is specified, the next character g belonging to the group of table number 6 can be easily specified from the next address N9 corresponding to the character B. Further, when a character for which the next address does not exist, such as the character g of address N9, is specified, the process returns to the first character of the group to which the character belongs in the registered character code table.

In this example, the registered character code table can be added to or deleted from the registered character code table at any time, but for this purpose, the registered character code table shown in FIG. It is necessary to form a table of similar codes to show. As shown in FIG. 5, the data corresponding to each address in the similar code table is composed of a table number and a character code, and the table number part contains table numbers 1, 2, and 3.
. In this case, each table number is written consecutively and repeated as many times as there are characters belonging to that table.

When the contents of the similar code table of FIG. 5 are displayed on the display device 3 of FIG. 1, the result will be as shown in FIG. In FIG. 6, the table on the left shows the table number of each group, and the table on the right shows similar characters belonging to the group with that table number and attributes of each similar character. The similar characters in this figure 6 are
These characters are not characters that are determined to be similar by the character recognition board 5 in FIG. 1, but are characters that are determined to be similar by the operator and are grouped. Using this display, it is possible to add similar characters of a desired table number, delete similar characters of a desired table number, or set a similar character group of a new table number.

Next, a method for correcting erroneously recognized characters will be explained with reference to FIGS. 2 and 3. FIG. 2 shows a display screen 3A of the display device 3 of this example, and this display screen 3A includes:
A recognition result display area 20 is provided to display recognition results of characters for one page of an input document. Similar to the input document, the recognition results are displayed horizontally. A cursor 21 indicates a character to be corrected. A surrounding area display section 22 and an original pattern display section 32 are provided at the lower left of the recognition result display area 20, and by moving the cursor 21 over a desired character in the recognition result, the original pattern display section 32 and the surrounding area are displayed. The original dot pattern of each character to be corrected and the peripheral pattern of a predetermined range around the character are displayed on the area display section 22. In the example in Figure 2, the correct character in the input document is the number "8", but the recognition result is the lowercase English letter "8".
g”.

Reference numeral 23 denotes a character attribute button (the "button" in this example is not a physical button but a display on the display screen) that indicates the character attribute of the recognition result of the character to be corrected. The character resulting from the recognition of the character to be corrected is displayed on the character of interest button 24 to the right of the character attribute button 23. When the recognition result character of the correction target character is “g” as in this example, the character attribute button 23 displays the character “English” (or the character “lowercase English”). In addition, in this example, a cross-shaped cursor 3 is input using a coordinate input unit such as a mouse (not shown).
By moving 5 onto each button and operating the coordinate input switch, the function of that button can be executed.

In this example, by selecting the character attribute button 23 with the cursor 35, the displayed character of the noted character button 24 can be replaced with the next character belonging to the same similar character table in FIG. The character attribute button 23 can display the attribute of the next character. With this function, as will be described later, it is possible to easily modify characters to be modified.

Reference numeral 25 indicates a function button area, in which candidate buttons 26A, code buttons 26B, etc. are displayed. For example, when the candidate button 26A is selected with the cursor 35, ten candidate characters determined to be similar to the character to be corrected are displayed in the candidate character display area 33 together with their residuals. Since the operator can recognize the original character from the display on the peripheral area display section 22 and the original pattern display section 32, if the original character exists among the 10 candidate characters, the operator can select that character with the cursor 35. The character code of the selected character is assigned as the character code of the unrecognized character. If the desired character does not exist among the candidate characters, the character code can be directly input by selecting the code button 26B, and the kanji can be input by using the kana-kanji button 26C and the radical stroke count button 26E. . In addition, by selecting the table button 26D, a list of character codes (for example, JIS codes) and corresponding characters is displayed, and the external character registration button 26F and learning deletion button 26 are displayed.
G is used when registering characters with special fonts. Further, 27 indicates a re-recognition button, and 28 indicates a backspace button. The re-recognition button 27 is used when re-recognizing a character to be corrected.

Reference numeral 29 indicates a modification history display section, and this modification history display section 29 displays 10 characters from the left in descending order of frequency of appearance, which are obtained by modifying characters or adding character codes after rejection. When a desired character is displayed on the display section 29, by selecting the desired character with the cursor 35, an accurate character code can be easily assigned to the unrecognized character. 30 indicates the next button, and 31 indicates the previous button. When the next button 30 is selected, the cursor 21 moves to a character that is the same as the character that is being focused on and exists later, and when the previous button 31 is selected, the cursor 21
moves over the character that is the same as the character you are looking at but is earlier.

Reference numeral 34 indicates a similarity button, and when this similarity button 34 is selected, the following display is performed in the recognition result display area 20. 1. Displaying a list of similar character tables 2. Additional registration to similar character table 3. New registration to similar character table 4. Deletion of registration of similar character table For example, if you select the top list display, the list of similar character tables shown in FIG. 6 will be displayed in the recognition result display area 20, and the next additional registration will cause the character button The character displayed in 24 can be added to the table of desired similar characters, and the next new registration allows that character to be registered in a new table, and the last registration deletion allows the character to be registered in the desired table. Desired characters or the desired table itself can be deleted.

To explain an example of how to use the similar character table of this example, when candidate button 26A is selected with cursor 35 in FIG. 2, characters "g", "6" and "9'' are displayed together with each residual. In this example, it can be recognized that the original character is "8" from the display in the peripheral area display section 22 and the original pattern display section 32, so it can be determined that the candidate character display area 33 does not contain the target character. I understand. Then, the operator selects the character attribute button 23 with the cursor 35. Correspondingly, as shown in FIG. 3, the next similar character belonging to the same table as the character "g" (as seen in FIG. 6, it is the character "8") is displayed on the notable character button 24. The similar character button 23 displays the attribute of the similar character (in this example, "number").
) is displayed.

When the character displayed on the character of interest button 24 is the desired character, the operator selects the character of interest button 24 with the cursor 35 to change the character code of the character of the character of interest button 24. The cursor 2 in the recognition result display area 20 is attached to the character to be corrected.
The selected character is displayed in 1. In this case, the attribute of the character is displayed on the character attribute button 23, so for example, the number "1" may be confused with the lowercase letter "l", or the number "0" and the uppercase letter "O" may be confused. There is no chance of getting them mixed up. On the other hand, if the character displayed on the character of interest button 24 is not the desired character, the operator selects the character attribute button 23 with the cursor 35, and the character of interest button 24 displays the next character from the same similar character table. is displayed, and the character attribute button 23 displays the attribute of the next character.

As described above, according to this example, the target character and the recognition result character are registered in advance in the group with the same table number in FIG. Even if the desired character is not displayed, there is an advantage in that the character code of the desired character can be easily assigned to the character to be corrected by simply selecting the character attribute button 23. This makes it possible to easily correct erroneous recognition results. In this case, since the attribute of each character is displayed on the character attribute button 23, there is an advantage that the target character will not be misidentified.

Note that the present invention is not limited to the above-mentioned embodiments, and it goes without saying that various configurations may be adopted without departing from the gist of the present invention.

[0031]

[Effects of the Invention] According to the present invention, since an operator can group and register characters that cannot be determined to be similar by a machine in the storage means for similar characters, it is possible to correct characters that have been erroneously recognized by the machine. There are benefits that can be easily made. Also, since the attributes of each character are displayed on the display means,
There is an advantage in that characters can be corrected accurately.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram showing an embodiment of a character recognition device according to the present invention.

FIG. 2 is a front view showing an example of a display screen of the display device according to the embodiment.

FIG. 3 is a diagram of main parts showing an example of changes in display content on the display screen of FIG. 2;

FIG. 4 is a diagram showing a memory configuration of a registered character code table of a similar character table in one embodiment.

FIG. 5 is a diagram illustrating a memory configuration of a similar code table of a similar character table according to an embodiment.

6 is a diagram showing an example of the contents of the similar code table of the similar character table of one embodiment displayed on the display screen of FIG. 2; FIG.

[Explanation of symbols]

1 Image scanner 2 Workstation 3 Display device 5 Character recognition board 12 Memory for major classification dictionary 15 Memory for subclassification dictionary 10 Subclassification circuit 17 Judgment circuit 19 Similar character table

Claims (1)

[Claims] 1. A data input means for obtaining a dot pattern of a character to be recognized, a dictionary storage means for storing character codes and feature quantities of printed characters in correspondence, and a dictionary storage means for storing a character code and a feature amount of a printed character in correspondence. a classification means for outputting a character code of a printed character similar to the recognition target character as a dot pattern; a display means for displaying the original dot pattern of the recognition target character and a character similar to the recognition target character; Separately, it has a storage means for similar characters in which a character code and the character attributes of this character code are recorded, and the recognition result character obtained by referring to the storage means for the dictionary matches the recognition target character. A printed character recognition device characterized in that, if there is no similar character, the character of the group to which the recognition result character belongs and the attribute of this character are displayed on the display means by referring to the storage means for similar characters.