JPS61251982A - Optical character reader - Google Patents

Abstract

PURPOSE:To improve the recognition accuracy of a character to be read by sectioning a character pattern at every character into blocks, deciding the noise degree of each block and removing the block concerned. CONSTITUTION:The character pattern at every character is stored in a buffer memory 10, and composite similarity of the character pattern as binary-coded data outputted from the memory 10 is calculated in a calculating circuit 11 on the basis of a standard pattern from a dictionary memory 12, and a maximum similarity category is outputted as the recognized result. The character pattern in the memory 10 is scanned by a control circuit 13, and whether the each block of the character pattern is a noise or not is decided. When the block is decided to be a noise, it is removed. In this case a memory 14 stores threshold data necessary for noise decision processing. Then a deciding circuit 15 outputs a final recognized result from the recognized result outputted from the circuit 11 in accordance with a control signal N1 from the circuit 13.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to an optical character reading device that specifically reads characters such as alphanumeric characters other than Chinese characters. [Technical Background of the Invention and Problems Therewith] Conventional , optical character reading devices (OCR) perform recognition processing on a character-by-character basis when reading characters such as alphanumeric characters, kana characters, and symbols.For this reason, each character is recorded on a form, etc. When a character is read, if a part of the character is blurred or there is noise such as dust attached near the character, the OCR may misrecognize it.

[Objective of the Invention] The object of the present invention is to make it possible to reliably recognize even when a part of the character to be read is blurred or there is noise such as dust attached near the character, thereby greatly improving the reading accuracy. An object of the present invention is to provide an optical character reading device that can be improved in the following manner.

[Summary of the Invention] The present invention distinguishes the character pattern of each character into each block, determines the noise degree of each block, and if this noise degree is outside a predetermined tolerance range, removes the corresponding block as noise. Equipped with noise removal control means. In the noise degree determination process, noise threshold data set in advance for each type of character is used. When the noise removing means determines that a part of each block of the character pattern is noise, the similarity calculating means calculates a similarity value for the character pattern from which the noise has been removed. Further, the similarity calculation means calculates a similarity value for the character pattern before the noise removal process. The recognition result determination means is configured to compare each similarity value and output a final recognition result based on the comparison result.

With such a configuration, it is possible to remove noise attached near the character to be read from the character pattern, and it is also possible to improve recognition accuracy even when a part of the character is blurred.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a block diagram showing the configuration of an OCR according to an embodiment. In FIG. 1, a buffer memory 10 is a memory that stores character patterns for each character. The character pattern is binary data output from a photoelectric conversion circuit (not shown). The similarity calculation circuit 11 includes a buffer memory 1
For the character pattern output from 0, dictionary memory 1
A composite similarity calculation is performed based on the standard pattern from 2, and the maximum similarity category is output as a recognition result.

The noise removal control circuit 13 scans the character pattern in the buffer memory 10 and determines whether each block of the character pattern is noise. The noise removal control circuit 13 is
Processing is performed to remove blocks determined to be noise. The noise memory 14 stores noise threshold data necessary for noise determination processing by the noise removal control circuit 13. The determination circuit 15 outputs the final recognition result as an answer from the recognition results output from the similarity calculation circuit 11 in response to the control signal N1 from the noise removal control circuit 13.

In the OCR having such a configuration, the operation of this embodiment will be explained with reference to FIGS. 2 to 4. First, it is assumed that characters recorded on a form are scanned by a photoelectric conversion circuit (not shown) and a character pattern for one character is transferred to the buffer memory 10. When the character pattern is transferred to the buffer memory 10, a transfer end signal T is output from the OCR control circuit (not shown) to the noise removal control circuit 13. As a result, the noise removal control circuit 13 starts operating,
The buffer memory 10 is scanned and noise determination processing is performed for each block of the character pattern. Here, a character pattern consisting of multiple blocks refers to a character pattern consisting of two blocks, such as the alphabetic letter "i", for example.
Or, the character pattern consists of multiple blocks including noise blocks. The noise removal control circuit 13 is
Noise memory 14 for each block of character pattern
Noise determination processing is performed based on the noise darkness value data.

The noise memory 14 stores in advance a table as shown in FIG. 2, for example. This table is a subset (
It consists of noise threshold data set for each character type (alphanumeric characters, kana characters, and symbols). The noise threshold data is
As shown in the figure, each block of the character pattern consists of data of a height threshold, HH1 width threshold, amount value, and area threshold (lower limit 31, upper limit 32). The noise removal control circuit 13 compares the height H1 width W and area S of each block of the character pattern with the noise darkness value data of the noise memory 14. For example, if the character pattern is a number, noise determination processing is performed based on noise threshold data corresponding to a subset of numbers. In the comparison process, the noise removal control circuit 13 determines that the block is a noise block if each of the height H1 width W and area S of the block is smaller than the noise threshold data HH, WW, 81. Remove from character pattern. In addition, the height H1 width W of the block is greater than each noise threshold data HH, W, satisfying the condition,
Assume that the area S is a comparison result such as [S1≦S<S2j. In this case, the noise removal control circuit 13 outputs the control signal N2 of "1" to the determination circuit 15 without executing the noise removal process in the first recognition process.

When the noise removal control circuit 13 finishes the noise processing, it outputs a control signal N2 to the similarity calculation circuit 11.

The similarity calculation circuit 11 performs similarity calculation on the character pattern output from the buffer 7 memory 10 based on the standard pattern data stored in the dictionary memory 12 based on the composite similarity method. Based on this calculation result, similarity calculation circuit 1
1 outputs the maximum similarity category as a recognition result. Here, the similarity calculation circuit 11 calculates the similarity value R using, for example, a calculation formula shown in equation (1) below.

R-<7-den>z/xYn2-1? Bv・<1>Here, f is a character pattern.

Judgment circuit 15 If t, i and the control signal N1 from the noise removal control circuit 13 are "O", output the maximum similarity category output from the similarity calculation circuit 11 as the final recognition result (answer). do. That is, the noise removal control circuit 1
3, this is a case where noise in the character pattern is removed or no noise is detected. Control signal N1 or
1”, the determination circuit 15 determines that the noise removal control circuit 1
3, outputs an instruction signal @I instructing to transfer the character pattern again. At this time, the determination circuit 15
The similarity value and its category (character code), which are the first recognition results output from the similarity calculation circuit 11, are temporarily stored. In response to the instruction signal @I from the determination circuit 15, the noise removal control circuit 13 removes the block that was put on hold in the noise determination process as described above from the character pattern as a noise block. That is, among the blocks of the character pattern, noise removal processing is performed on blocks whose only area does not satisfy the condition, for example, as described above.

As a result, the character pattern from which the noise block has been removed is transferred from the buffer 7 memory 10 to the similarity calculation circuit 11 again. The similarity calculation circuit 11 executes similarity calculation in the same manner as described above, and determines the maximum similarity category by the judgment circuit 15.
Output to. The determination circuit 15 is the first similarity calculation circuit 1
The category similarity value from 1 and the next similarity value are compared, and the category with the larger similarity value is output as the final recognition result.

The above operations will be specifically explained with reference to FIGS. 3 and 4. Assume now that a character pattern of numbers as shown in FIG. 3 is stored in the buffer 1 memory 10.

Here, each block 30 to 32 of the character pattern in FIG.
are blocks having the following conditions. That is, the block 30 has a height H>12 and a width W>1
0, area S>80, and block 31 has height H<12, width W<10, area S<
30, and the block 32 has a height of 12 and a width W<10, 30≦area S<80.

In the first process, the noise removal control circuit 13 removes block 31 as a noise block from the character pattern shown in FIG. 3 based on the table shown in FIG.
2 suspends noise removal. As a result, the character pattern consisting of blocks 30 and 32 is transferred from the buffer memory 10 to the similarity calculation circuit 11. The similarity calculation circuit 11 executes similarity calculation and, for example, if the similarity value is M30.
J, the recognition result corresponding to the number "2" is determined by the judgment circuit 15.
Output to. The determination circuit 15 stores this first recognition result and outputs an instruction signal (2) to the noise removal control circuit 13.

The noise removal control circuit 13 controls the suspended block 3.
2 is removed as noise, and the character pattern of only block 30 is transferred to the similarity calculation circuit 11. The similarity calculation circuit 11 outputs a recognition result corresponding to the number "7" to the determination circuit 15 with respect to this character pattern, for example, with a similarity value M10J. As a result, the determination circuit 15 compares the similarity value of the first recognition result with the next similarity value, and
2" is output as the final answer.

Similarly, it is assumed that character patterns (numbers) are stored in the buffer memory 10 as shown in FIG. here,
It is assumed that each block 40, 41 of the character pattern in FIG. 4 has the following conditions.

That is, the block 40 has a height H>12, a width W>10, and an area S>
aO, and the block 41 has a height of 12 and a width W<10.30≦area S<80.

In the first process, the noise removal control circuit 13 suspends the block 41 and transfers the character pattern consisting of blocks 40 and 41 to the similarity calculation circuit 11.

For this character pattern, the similarity calculation circuit 11 outputs a recognition result corresponding to the number "2" with a similarity value r120J, for example. The determination circuit 15 instructs the transfer of the character pattern again in the same manner as described above.

The noise removal control circuit 13 removes the block 41 as noise and transfers only the character pattern of the block 40 to the similarity calculation circuit 11. For this character pattern,
The similarity calculation circuit 11 outputs a recognition result corresponding to the number "7" with a similarity value r140J, for example. This results in
The determination circuit 15 outputs the recognition result corresponding to the number "7" as the final answer.

In this way, for each block of character pattern,
A noise determination process is executed based on the noise threshold data. Depending on the result of this noise determination process, each similarity value between the character pattern from which noise blocks have been removed and the character pattern before the noise removal process is calculated. Based on this calculation result,
The recognition result with the larger similarity value is set as the final recognition result. Therefore, for a character pattern in which a part of the character is blurred (Figure 3), accurate recognition results can be obtained even if block 32, which is originally part of the character pattern, is determined to be noise. It is. In addition, for character patterns (Figure 4) when noise is attached near the characters,
A character pattern with noise blocks removed can be obtained. Therefore, it is difficult to recognize the original character pattern.
Since 1i processing can be executed, accurate recognition processing is possible.

In other words, even if the noise threshold is reduced in order to improve noise detection, it is possible to improve the recognition accuracy for blurred characters that are likely to cause false recognition due to noise removal.

[Effects of the Invention] As detailed above, according to the present invention, when characters to be read are alphanumeric characters, kana characters, symbols, etc., parts of the characters are blurred or noise such as dust is attached near the characters. Therefore, even if there is some defect in the recorded state of the characters, the reading accuracy of the characters to be read can be greatly improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an optical character reading device according to an embodiment of the present invention, FIG. FIG. 4 is a diagram showing an example of a character pattern for explaining the specific operation of the same embodiment. DESCRIPTION OF SYMBOLS 10... Buffer memory, 11... Similarity calculation circuit, 12... Dictionary memory, 13... Noise removal control circuit, 14... Noise memory, 15... Judgment circuit.

Claims (1)

[Claims] A buffer memory that stores character patterns for each character,
a noise memory that stores in advance noise threshold data for each type of character, which is necessary for determining whether a character is a noise, and a noise memory that distinguishes the character patterns in the buffer memory into each block, and divides the character patterns into each block based on the noise threshold data of the noise memory. noise removal control means that determines the degree of noise of a block and removes the corresponding block as noise if the degree of noise is outside a predetermined tolerance; and a similarity between the character pattern output from the buffer memory and the standard pattern. a similarity calculation means for calculating a degree value, and a similarity value from the similarity calculation means for the character pattern from which noise has been removed when a part of each block of the character pattern is determined to be noise by the noise removal means; and recognition result determination means for comparing the similarity value with the character pattern before noise removal processing and determining the final recognition result based on the comparison result. .