JPH05120485A - Character reader - Google Patents

Abstract

PURPOSE:To improve reading precision without dropping of memory usage efficiency and processing speed. CONSTITUTION:A scanner 10 which scans characters to output an image data G1, a recognizing part 11 which extracts respective character patterns out of the data G1 and accesses a standard pattern R1 to extract N numbers of candidate character codes, an error table referring part 12 which further refers to an error table R2 for each candidate character code to extract M numbers of candidate character codes, a post-processing part 13 which finally identifies the most suitable character code out of (N+M) numbers of candidate character codes to introduce as a final derivated code G4 and an output part 14 which outputs the final derivated code G4 are provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character reading device, and more particularly to a character reading device which outputs a read character as a corresponding character code.

[0002]

2. Description of the Related Art Generally, a character reading device scans characters of a Japanese sentence written on a paper by light to read them, and first outputs image data. Then, the pattern of each character included in the image data is extracted, and this character code is output while converting each character pattern into a corresponding character code by pattern matching. The character code is generally used as input data for a computer, a word processor, and a notebook-sized electronic computer (electronic system notebook).

FIG. 4 is a block diagram of a conventional character reading device 2. The character reading device 2 inputs a scanner 10 that outputs a G1 image data by scanning a Japanese sentence previously written on paper with light, and extracts each character pattern included in the image data G1. Then, each character pattern is converted into a character code by pattern matching (pattern matching) with the standard pattern R1, and N candidate character codes and similarity data G2 are output for each character pattern as a pattern matching result. Recognition unit 1
1. Accessing the post-processing dictionary R3, the post-processing unit 13 for outputting the optimum final recognition code G5 finally recognized from the given N candidate character codes and the final recognition code G5 are connected to the next stage. The output unit 14 outputs the input data to a personal computer or a word processor.

The recognition unit 11 of the conventional character reader 2 matches the standard pattern R1 with each character pattern cut out from the image data G1 to determine N candidate character codes that can be the optimum character code, The N candidate character codes and the similarity data G2 are derived to the post-processing unit 13. The post-processing unit 13 sequentially inputs the N candidate character codes G2 given from the recognition unit 11 in the previous stage, develops the candidate character codes G2 by word spelling, and accesses the post-processing dictionary R3, while developing the developed word spelling. Finally, character recognition is performed depending on whether or not it is registered in this dictionary R3,
The final recognition code G5 is output.

[0005]

The standard pattern R1 accessed by the recognition unit 11 of the conventional character reading device 2 described above stores a standard pattern of a limited type in advance, so the pattern in the recognition unit 11 is stored. At the time of matching, the derived N candidate character codes were limited to the character codes corresponding to the character patterns stored in the standard pattern R1. Therefore, when a character whose standard pattern is not stored in the standard pattern R1 is read by the scanner 10 and input as the image data G1, even if the recognition unit 11 extracts N candidate character codes, I could not recognize it. If the standard pattern of all characters is stored in the standard pattern R1, the above-mentioned problem of unrecognization does not occur, but the realization thereof causes an increase in memory consumption in the standard pattern R1. The problem is that the effective use of the memory in the character reading device 2 is reduced and the recognition processing time including the access process of the standard pattern R1 is significantly increased, and the reading speed of the character reading device 2 itself is significantly reduced. .. Therefore, there is a problem that increasing the number of patterns stored in the standard pattern R1 in order to avoid becoming unrecognizable as described above cannot be a fundamental solution.

Therefore, an object of the present invention is to provide a character reading device with improved character reading performance without causing a decrease in utilization efficiency of memory capacity and a decrease in processing speed.

[0007]

A character reading device according to the present invention stores scanning means for scanning a character to output an input character pattern and a plurality of character patterns in advance in association with respective character codes. A first storage means and a first
Character code output means and character codes of a plurality of character patterns including character patterns not stored in the first storage means,
Second storage means for preliminarily grouping and storing each similar pattern, and determination means for determining that the corresponding matching result is not valid for each of the character codes output by the first character code output means, and the determination means. The second storage means is searched on the basis of the corresponding character code based on the judgment output of, and is output by the second character code output means for outputting the character code included in the corresponding group and the first and second character code output means. And a recognition output means for recognizing and outputting an optimum character code from a plurality of character codes.

The above-mentioned first character code output means performs pattern matching between the input character pattern output by the scanning means and each of the plurality of character patterns stored in the first storage means,
The character code corresponding to at least one character pattern whose matching result reaches a predetermined level is output.

[0009]

The character reading apparatus according to the present invention further comprises the second storage means in addition to the first storage means so that the reading accuracy can be improved without significantly increasing the memory consumption and the processing time. It is characterized by The second storage means stores each character code of a plurality of character patterns including a character pattern not stored in the first storage means, and the first character code output means accesses the first storage means to output the character code, Since the two-character code output means accesses the second storage means and outputs the character code, the recognition output means is the first
Since the optimum character code is recognized from the plural character codes output from the second character code output means, the probability that the optimum character code exists in the plural character codes, that is, the probability that the optimum character code is recognized is improves. Further, since the second storage means stores only the character code while the first storage means stores the character pattern and the character code, the memory consumption thereof is smaller than that of the first storage means, Since the character code output means outputs the character code without performing the pattern matching, the pattern matching processing becomes unnecessary and the processing speed is improved despite the high recognition rate in the recognition output means.

[0010]

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

FIG. 1 is a block diagram of a character reading apparatus according to an embodiment of the present invention. 2 (a) to 2 (d) are views for explaining the procedure for creating the error table according to the first embodiment of the present invention.

FIG. 3 is a flow chart of the reading process of the character reading device according to one embodiment of the present invention. In FIG. 1, the character reading device 1 includes a scanner 10, a candidate character extraction unit 15 including a recognition unit 11 and an error table reference unit 12, and a post-processing unit 1.
3, a standard pattern R1, which is a kind of memory in which the output unit 14 and data are stored, an error table R2, and a post-processing dictionary R3.

The standard pattern R1 is configured to store a standard pattern for each of a plurality of characters and its character code in association with each other. Although details of the error table R2 will be described later, the character patterns corresponding to the character patterns stored in the standard pattern R1 and the character patterns not stored in the standard pattern R1 are configured to be grouped and stored for each similar pattern. Table. The post-processing dictionary R3 is a dictionary configured to store a plurality of words in character code units.

The scanner 10 optically reads the characters of the Japanese sentence written on the paper surface in advance and derives the image data G1. The candidate character extraction unit 15 inputs the image data G1,
Accordingly, (N + M) candidate character codes and the similarity data G3 are output. The candidate character extraction unit 15 includes a recognition unit 11 that accesses the standard pattern R1 and an error table reference unit 1 that accesses the error table R2.
Have two. The recognition unit 11 cuts out a character pattern of each character from the image data G1 and performs pattern matching between the cut-out character pattern and the standard character pattern stored in the standard pattern R1. Further, the character code corresponding to the standard character pattern is used as the candidate character code, and the data G2 of the similarity regarding the similarity of each pattern is output. The error table reference unit 12 accesses the error table R2 for each of the given N candidate character codes, identifies each candidate character code group, and reads M character codes from the identified group. The error table reference unit 12 derives (N + M) candidate character codes and similarity data G3, which is a combination of the N candidate character codes given and the M candidate character codes read by accessing the error table R2. To do.

The post-processing unit 13 is a candidate character extraction unit 1 in the preceding stage.
The (N + M) candidate character codes extracted in step 5 are sequentially input, and they are developed by word spelling to obtain the post-processing dictionary R3.
The final character code is identified and recognized from the validity of the spelling of the word by comparing and collating with the word stored in. In this way, the character code finally recognized by the post-processing unit 13 accessing the post-processing dictionary R3 is derived to the output unit 14 as the final recognition code G4.

The output unit 14 gives the final recognition code G
4 as input data to a personal computer or word processor, or to a printer or CRT
It operates to print or screen output on (cathode ray tube).

As described above, the character reading device 1 operates by reading the characters of a Japanese sentence with the scanner 10 and giving the read result as a character code G4 to a personal computer or a word processor.
Alternatively, since the printer or the CRT is configured to print out or output on the screen, the operator can operate the scanner 10
The read result of the Japanese sentence read in can be processed as data with a personal computer, etc., and the result can be confirmed by the print result or screen display.

The character reading device 1 differs from the above-described conventional character reading device 2 in that the character reading device 1 has a large difference.
Is provided with the error table R2, and the error table reference unit 12 for accessing the error table R2 is provided. The character reading device 1 configured as described above uses the candidate character extraction unit 15 including the recognition unit 11 and the error table reference unit 12,
In the process of converting the image data G1 read by the scanner 10 into a character code, it is determined that the character code of the first to (N + M) candidates for each character pattern included in the data G1 seems to be the character code of the character pattern. The final character code is recognized by the post-processing unit 13. Therefore, the post-processing unit 13 of the conventional character reading device 2 determines the final recognized character from N candidate character codes, but the post-processing unit 13 of the character reading device 1 has (N + M) number of characters. Since the character is finally recognized from the candidate character code of, the recognition accuracy, that is, the character reading accuracy is dramatically improved as compared with the conventional one.

Next, the procedure for creating the error table R2 will be described. The error table R2 is created in advance by the procedure shown in FIGS. 2 (a) to 2 (d).

In FIG. 2A, the character types CH that can be input and output in the device 1 are predetermined, and for each of the character types CH, as shown in FIG. 2A, the category number Ci (i = 0.1) is set. .2.3 ...) are prepared in a one-to-one correspondence. The table in which the character type CH and the category number Ci have a one-to-one correspondence in this way is provided as a category correspondence table T1.

FIG. 2B shows a flow of the procedure for creating the error table R2, FIG. 2C shows a table T2 for creating an error table, which is accessed at the time of creation, and FIG. ) Shows the error table R2 created by the procedure of FIG. The procedure flow of FIG. 2B is prepared as a program in advance. Figure 2
Table T2 in (c) and table R2 in FIG. 2 (d)
The area is previously set in a memory space having a predetermined capacity.

The tables and processing flow programs shown in FIGS. 2A to 2D are prepared in a separate computer, and the error table R2 is created in advance in that computer, and the character reading device 1 is prepared. May be loaded into the internal memory of the character reading device 1, or these tables and programs may be prepared in the internal memory of the character reading device 1 so that the device 1 itself creates the error table R2 prior to the character reading operation. You can

Step ST1 of FIG. 2B, (ST in the figure
1), a storage area for the error table creation table T2 and the error table R2 is secured and the area is cleared (initialized).

Subsequent processing from step ST2 is performed by a processing unit having a function corresponding to the recognition unit 11 of the character reading device 1.

In the process of step ST2, it is determined whether the data input is completed, that is, whether the input of the image data of one character is completed. At this time, if the end of the input data is confirmed, the processing from step ST8 described later is executed, but if the end of the input data is not confirmed, the input 1 is input in the processing of the next step ST3.
Character recognition is performed on character image data. This character recognition is performed by accessing the standard pattern R1. At this time, the character code obtained by accessing the standard pattern R1 is sequentially output on the screen or printed out and presented to the operator, so that the operator sees the output character and reads the read character and the character recognition result. Check if the recognition result is correct by comparing with the recognition character. Whether or not the recognition result is correct is determined in the processing of the next step ST4,
If it is correct, the process proceeds to step ST2 again, and the process for the next character pattern is performed in the same manner.
If the recognition result is not correct, the process proceeds to step ST5 and thereafter.

In the process of step ST5, the process for creating the table T2 for creating the error table is started. The category correspondence table T1 prepared in advance based on the character type CH of the character pattern input in the process of step ST5 is accessed, the corresponding category number Ci is specified, and this is assigned to the variable x. In the processing of the next step ST6, the above-mentioned step ST
Similarly, the table T1 is accessed based on the character type CH of the recognition character obtained by the recognition result obtained in the processing of 3, the corresponding category number Ci is specified, and this is assigned to the variable y and set. Then, in the processing of the next step ST7, 1 is added to [x, y] of the table T2 for creating an error table to create (update) the table T2.
To do. Creation of the table T2 for creating the error table shown in steps ST5 to ST7 will be described with reference to FIG.

FIG. 2 (c) shows a table T2 created for creating the error table R2, which is a two-dimensional array table [x, y]. The structure of the table T2 is two-dimensionally composed of [x, y] and is composed of a character type x that can be input / output and a character type y that can be input / output. For example, if the recognition result is "sorrow" when the character "depression" is recognized, the table T2
Register "decline" in the "sorrow" element of. In other words, 1 is added to the corresponding element. In addition, if you recognize the character "Age" by another handwriting, the recognition result will be "Kyo".
When it is, “decline” is registered in the element of “K” in the table T2, that is, 1 is added. In this way, when the number of so-called recognition errors is sequentially registered in the table T2, when a character corresponding to the category number Ci is recognized, that character may be the category number Cidj, that is, an error table. The table R2 (see FIG. 2D) is completed.

The error table R2 will be briefly described with reference to FIG. 2D. From the table T2 for creating the error table, the category number Ci = 21 (decline) is changed to the category number Ci = 22 (sorrow) and the category number. Ci = 2
It is recognized that it may be 3 (K). In the error table R2, i is the number of categories recognized for creating the table R2, and j is the optimum number of characters registered in the table R2, which is determined based on the memory capacity and experience of the device 1 itself.

As described above, the creation of the error table R2 based on the table T2 for creating the error table is performed in the processing of steps ST8 and ST9, and the value of the character of the corresponding y column is set for each x column of the table T2. , J j for each x column
The error table R2 is created by outputting the characters in the column.

As described above, since the standard pattern R1 is referred to when the error table R2 is created, characters that are not registered in the standard pattern R1, for example, the standard pattern R1.
If a JIS second-level character or the like that is not reflected in is recognized, the JIS second-level character code will appear in the error table R2.
Even if the second level character is entered, the error table R2
Since that character code exists, it is sent to the post-processing unit 13 as a candidate character code.

Further, a peculiar font that is not reflected in the standard pattern R1 (a font that is not printed in the Gothic typeface or Mincho typeface of a newspaper and is decorated by design) is also recognized as input data when the table R2 is created. By doing so, since the character code of this peculiar font is reflected in the error table R2, the candidate character extracting unit 15
Will be extracted as a candidate character code.

As described above, it is possible to increase the number of recognition target characters and the number of recognition target fonts by providing the error table R2 and the error table reference unit 12 for accessing the same, without any operation of the standard pattern R1. it can.

Further, the standard pattern R1 accessed by the recognition unit 11 requires a memory capacity of 384 bytes in a 32-bit machine for storing data of one character, but the error table R2 shows a character code. Since it is configured to store only data, it can store 2 bytes for one character data. Therefore,
Rather than expanding the pattern data with the standard pattern R1 to improve the reading accuracy of the character reading device 1, it is better to provide the error table R2 to give more candidate character codes to the post-processing unit 13. The memory consumption of the device 1 has an advantage that it may be small for its reading accuracy.

Next, the reading operation of the character reading device shown in FIG. 1 will be described according to the processing flow of FIG.

It is assumed that the standard pattern R1, the error table R2, and the post-processing dictionary R3 are stored inside the character reading device 1 in advance.

In step S1 of FIG. 3 (abbreviated as S1 in the figure), the scanner 10 scans Japanese characters written on the paper surface to derive image data G1. Image data G1
Is given to the candidate character extraction unit 15.

In the processing of the next steps S2 to S4, the recognition section 11 cuts out one character from the supplied image data G1 and stores each standard pattern and the pattern stored in the standard pattern R1 for each cut-out character pattern. Matching (pattern matching) is performed, and N character codes corresponding to a standard pattern that matches a predetermined matching condition are extracted. The recognition unit 11 performs pattern matching between the given character pattern and each of the plurality of character patterns stored in the standard pattern R1, and if the similarity of the pattern corresponds to a predetermined threshold value, the corresponding character pattern. Since the character code corresponding to is extracted as a candidate character code, at least one or more N character codes are extracted as candidates.

When the recognition unit 11 extracts the N candidate character codes and the similarity data G2 as described above, the data G2 is provided to the error table reference unit 12. In the process of step S5, the error table reference unit 12 determines whether or not each of the given N candidate character codes is a character code to be extracted. This determination is made based on whether or not the given similarity data has reached a predetermined threshold value. This similarity data is used for the scanner 1 and the character patterns stored in the standard pattern.
It represents distance data with the input character pattern read by 0, and it is known that the larger the distance data is, the less similar the patterns are. Therefore, if the similarity data is small, the patterns are similar. Can be said. Therefore, if the similarity data does not reach the predetermined threshold value, the error table reference unit 12 determines that the extracted candidate character code is not valid, that is, is suspicious, and the processing in step S6 and subsequent steps described later. Will move to
If it is not determined that the candidate character code extracted based on the given similarity data is suspicious, the processing from step S8 described below is executed.

In the process of step S6, the error table reference unit 12 accesses the error table R2 in response to the determination that the candidate character code extracted in the process of step S5 is not valid. At this time,
The error table R2 is searched for each of the suspicious candidate character codes, and a plurality of candidate character codes corresponding to the corresponding character code are further extracted. As shown in FIG. 2 (d), if it is determined that the candidate character code such as "decline" is suspicious, the error table reference unit 12 accesses the error table R2, and further "sorrow" about the character code "decline". Read the character code including "Kyo". In this way, N candidate character codes are further extracted by referring to the error table R2 for each of the candidate character codes determined to be suspicious. The (N + M) candidate character codes and the similarity data G3 are processed by the post-processing unit 1 in the processing of the next step S7.
3 is derived.

In the next step S8, the post-processing unit 13 sequentially develops the given candidate character code into word spelling, and performs language processing by referring to the post-processing dictionary R3 while considering the similarity data. .. If the word spelled and expanded is registered in the post-processing dictionary R3, each character of the expanded word spell is set as the final recognition code G4 and the output unit 1
4, and if not registered in the post-processing dictionary R3, the final recognition code G4 that gives the most optimum character code is derived to the output unit 14 in consideration of the similarity data.

The output unit 14 operates so as to output the final recognition code G4 given in the processing of the next step S9 as the input data of the computer or word processor in the next stage, or to print out or screen output to the printer or CRT. ..

Subsequently, in the process of step S10, it is determined whether or not the process is completed. This is the scanner 1
It is determined according to the fact that the scanner input at 0 is not performed for a predetermined time period. When it is judged that this process is finished, a series of character reading is finished, but if the character input continues in the scanner 10, the process immediately returns to the process of step S1 and the next input character is similarly (N + M) pieces. The process is repeated so that character recognition is performed by extracting the candidate character code of.

As described above, the candidate character extraction unit 15 of the character reading device 1 sets each of the N candidate character codes extracted by referring to the standard pattern R1 as the optimum character code based on a certain criterion. It is determined whether or not it should be extracted, and when it is determined that it should not be extracted, the error table R2 is referred to in order to extract a further character code corresponding to the candidate character code, and M character codes as additional candidate character codes. Is extracted, M character codes are added to N character codes, and the character codes are derived to the post-processing unit 13 to increase the recognition rate of final recognition in the post-processing unit 13.

[0044]

As described above, according to the present invention, the character code which is not stored in the first storage means is stored in the second storage means, and the first character code output means accesses the first storage means to access the character code. Is output, and the second character code output means accesses the second storage means and outputs the character code.
Since the recognition output means recognizes the final code from the plural character codes output by the first and second character code output means, the probability that the optimum character code is read is improved.

Further, since the second storage means is configured to store only the character code, its memory capacity may be smaller than that of the first storage means, and in view of the improvement of the reading rate mentioned above. The effect is that memory consumption is not large.

Further, since the second character code output means does not perform pattern matching but only accesses the second storage means and outputs the character code, the pattern matching here becomes unnecessary and the processing of the external device itself is performed. There is an effect that the speed is improved even though the reading accuracy is high.

[Brief description of drawings]

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

2A to 2D are views for explaining a procedure for creating an error table according to an embodiment of the present invention.

FIG. 3 is a flowchart of a reading process of the character reading device according to the embodiment of the present invention.

FIG. 4 is a configuration diagram of a conventional character reading device.

[Explanation of symbols]

1 Character reading device 10 Scanner 11 Recognition unit 12 Error table reference unit 13 Post-processing unit 14 Output unit R1 Standard pattern R2 Error table R3 Post-processing dictionary G1 Image data G2 N candidate character codes and similarity data G3 (N + M) Individual candidate character code and similarity data G4 final recognition code In the drawings, the same reference numerals indicate the same or corresponding portions.

Claims (1)

[Claims] 1. A scanning means for scanning a character to output an input character pattern, a first storage means for storing a plurality of character patterns in advance in association with respective character codes, and the scanning means for outputting. A pattern matching is performed between the input character pattern and each of the plurality of character patterns stored in the first storage means, and a character code corresponding to at least one character pattern whose matching result reaches a predetermined level is output. A one-character code output unit, a second storage unit that stores in advance each character code of a plurality of character patterns including a character pattern that is not stored in the first storage unit by grouping the similar patterns in advance, and the first character code For each of the character codes output by the output means, a determination that determines that the corresponding matching result is not valid A second character code output unit that searches the second storage unit based on the character code corresponding to the judgment output of the judgment unit and outputs the character code included in the corresponding group; A character reading device, comprising: a recognition output unit that recognizes and outputs an optimum character code from a plurality of character codes output by the two-character code output unit.