JPH051512B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a character recognition device (hereinafter referred to as
The present invention relates to character recognition (OCR), and in particular to a character recognition device that can reduce misrecognition.

In recent years, the progress of OCR has been remarkable, and OCR, which can print alphanumeric characters and alphanumeric characters and read handwritten characters, is widely used in form processing operations, etc. Development of word recognition technology is also active, and various methods are being tried.

Particularly in the recognition of handwritten characters including kanji, a method is desired that reduces recognition errors.

[Conventional technology]

Figure 3 shows a block diagram of OCR for Japanese characters, which targets handwritten characters including kanji.

In the figure, form 1 is a form in which the customer's address, name, product name, etc. are written in each field.

The reading unit 2 has a function of scanning the reflected light of the light irradiated onto the form 1 through a lens system 21 with an image sensor 22, reading one frame of characters, and transmitting the characters as image data to the binarization circuit 3. have

The main control section 4 has a function of controlling each section and executing a character reading and recognition processing program.

The image memory 5 stores binarized image data, that is, image data of read characters.

The character cutting circuit 6 has a function of cutting out one character from one frame of characters stored in the image memory 5 and sending it to the recognition circuit 10 based on the format information sent from the format information memory 9.

The feature extraction circuit 7 has a function of extracting character features sent from the recognition circuit 10, such as the number of strokes of the character, curve coefficients, etc., and sending them to the recognition circuit 10.

The dictionary memory 8 stores characteristics of characters that serve as recognition standards, and upon request from the recognition circuit 10,
It has a function of sequentially sending data to the recognition circuit 10.

The format information memory 9 stores information indicating the position where characters are written on the form 1, and has a function of sending the read position of the characters to the image memory 5, the single character cutting circuit 6, and the recognition circuit 10. .

The recognition circuit 10 receives from the feature extraction circuit 7 the features of the character sent from the single character extraction circuit 6, compares them with the features of the characters sequentially sent from the dictionary memory 8, determines the degree of matching, and determines the degree of matching. It has a function that sends out character codes as candidate strings in order.

Since it has such a configuration and function, the character recognition method will be explained. First, the characters on the form 1 are read and the binarized image data is stored in the image memory 5.

Next, the image data is sent to the single character cutting circuit 6,
Based on the character position information sent from the format information memory 9, one character is cut out and sent to the recognition circuit 10.

The recognition circuit 10 sends the input character data to the feature extraction circuit 7, which extracts and receives the features of the character data. Therefore, the characteristics of the characters are sequentially read out from the dictionary memory 8 and compared with the characteristics of the character data.
Characters with a high degree of matching are selected as candidate characters for recognition.

If there are multiple candidate characters, character codes are sequentially output as a candidate string.

Character recognition is sequentially performed on the image data stored in the image memory 5 in this manner.

[Problem that the invention seeks to solve]

According to the above conventional method, handwritten characters including kanji are recognized by the number of strokes of kanji, curve coefficient of hiragana, etc., but there is a problem that similar characters are often misrecognized due to handwriting habits etc. .

[Means for solving problems]

The present invention is a character recognition device that irradiates light onto characters written on a medium and recognizes character image data obtained from the reflected light of the irradiated light, based on character characteristics of the image data. a first recognition means for recognizing, and a second recognition having a person pattern storage means for recognizing based on the character pattern of the image data, storing the recognized character pattern, and outputting the character pattern at the time of subsequent recognition. means, and when the degree of recognition coincidence is lower than the desired value as a result of recognition by the first recognition means, the corresponding image data is transferred to the second recognition means.
This is a character recognition device that recognizes the character pattern using the recognition means and updates and stores the recognized character pattern in the filler pattern storage means, thereby solving the above-mentioned problems.

[Effect]

According to the present invention, instead of the conventional method of recognizing only based on the characteristics of image data, the degree of coincidence is determined by the first recognition means similar to the conventional method and the recognition by the first recognition means. When the character pattern is low, the image data is recognized by comparing it with the filler pattern stored in the filler pattern memory, or the image data is normalized to a predetermined size and the normalized character pattern and the standard pattern memory are Misrecognition of similar characters can be reduced by performing recognition by a second recognition means, such as by comparing and recognizing standard patterns stored in the same character.

Furthermore, since the filler pattern memory is updated each time recognition is performed, recognition processing efficiency can be improved.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 1 and 2. FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a flowchart of FIG. 1. The same reference numerals indicate the same objects throughout the figures.

In FIG. 1, the main control section 4a has the same function as the main control section 4 described in the conventional example, that is, a first recognition means that performs recognition processing based on the characteristics of image data of read characters. It also has a function of controlling the second recognition means that performs recognition processing based on the character pattern of the image data.

The determination circuit 11 compares and recognizes the features extracted from the image data with the character features in the dictionary memory 8, and determines whether or not a predetermined degree of matching has been reached. If the degree of matching is low, the recognition circuit 11 10a. In this case, the process moves to the second recognition means that performs matching based on the pattern of the image data.

The normalization circuit 12 has a function of normalizing the character pattern sent from the image memory 5 to n×m (for example, 48×48 dots).

The normalized character pattern memory 13 is a storage means for storing character patterns normalized by the normalization circuit 14.

The filler pattern memory 14 is a storage means that updates and stores a character pattern with a high degree of matching that has already been recognized, that is, a filler pattern each time it is recognized.

The standard pattern memory 15 is a storage means for storing standard patterns of characters.

The modified pattern generation circuit 16 detects a portion of the normalized character pattern with a low degree of matching, moves the pattern in that portion vertically or horizontally, and generates a modified pattern for increasing the degree of matching. Has a function.

For example, as shown in FIG. 3, the degree of matching of the normalized character pattern to the standard character pattern "3" is calculated to detect a portion with a low degree of matching, that is, portion a shown by hatching in the figure. First, move it downward to a specified size (for example, 1/8 of the height of the character pattern)
Move and transform the standard character pattern “3”
If the degree of agreement is high, the deformed pattern is sent to the deformed pattern memory 17,
If the degree of matching is low, part a is moved to the left and transformed, and if the degree of matching with standard character pattern "3" is high, the transformed pattern is stored in modified pattern memory 1.
Send to 7. If the degree of match is still low, move part a to the right and then upward to transform it, and if the degree of match with the standard character pattern "3" is still high, then
The deformed pattern is sent to the deformed pattern memory 17. If the degree of matching with the standard character pattern "3" is still low, the character is determined to be a reject character.

The modified pattern memory 17 is a storage means for storing the modified patterns generated by the modified pattern generation circuit 18.

The matching circuit 18 has a function of matching the input character pattern and the filler pattern, the normalized character pattern and the standard pattern, and the modified character pattern and the standard pattern, respectively.

The matching degree determination circuit 19 has a function of determining the degree of matching based on the matching result in the matching circuit 18, and notifying the recognition circuit 10a.

Since the apparatus has such a configuration and function, the recognition processing method will be explained with reference to the flowchart shown in FIG.

First, the characters on the form 1 are read, and as in the conventional example, the extraction circuit 7 extracts the characteristics of the image data, and the recognition circuit 10a compares them with the characteristics of the characters sent from the dictionary memory 8 to determine the degree of matching. is required.

The determination circuit 11 determines whether the degree of coincidence has reached a predetermined value (for example, 70%) and notifies the recognition circuit 10a.

Therefore, the recognition circuit 10a sends out characters whose matching degree has reached a predetermined value as a recognition result, that is, as a candidate character.

If the degree of matching is low, the recognition circuit 10a shifts to a recognition method based on the character pattern of the image data, and first, the recognition circuit 10a matches the cut-out character pattern sent from the single-character cutout circuit 6 to the matching circuit 18.
It also instructs the filler pattern memory 14 to send the filler pattern to the matching circuit 18 for matching.

The matching result is sent to the matching degree determination circuit 19, which determines whether the matching degree has reached a predetermined value and notifies the recognition circuit 10a.

When the degree of matching reaches a predetermined value based on this judgment result, the recognition circuit 10a takes the same action as above, and also sends the cut-out character pattern to the filler pattern memory 14 to identify the matched filler. It is updated and stored in place of the pattern.

When the degree of matching is low, the character pattern of the extracted character is sent to the normalization circuit 12 for normalization, and the normalized character pattern is stored in the normalized character pattern memory 13.

Next, the normalized character pattern and the standard pattern are sent to the matching circuit 18 for matching.

As a result of the matching, the degree of matching is determined in the same way.

When the degree of matching reaches a predetermined value as a result of the judgment, the same procedure as above is taken, and the normalized pattern recognized as a candidate character is sent to the filler pattern memory 14, and the filler pattern of the corresponding character is Updated.

When the degree of matching is low, the normalized character pattern is first sent to the deformed pattern generation circuit 16, and the part with a low degree of matching, for example, part a of the character pattern explained in FIG. 3, is moved vertically or horizontally, and the character Deformation pattern memory 1
Store in 7.

Therefore, the modified pattern and the standard pattern are sent to the matching circuit 18 for comparison.

As a result of the matching, the degree of matching is determined in the same way.

When the degree of matching reaches a predetermined value, the same procedure as above is taken, and the deformed pattern recognized as a candidate character is sent to the filler pattern memory 14, and the filler pattern of the corresponding character is updated.

If the degree of matching is low, a reject character (for example, "?") is output as an answer.

Recognition processing is performed based on the characteristics of the characters in the image data read in this way, and when the degree of matching is low, the recognition processing is shifted to recognition processing based on character patterns, and moreover, pattern matching with higher accuracy is performed in stages. Misrecognition can be further prevented, and since character patterns with a high degree of recognition matching are updated and stored in the filler pattern memory 14, a high degree of matching can be obtained at an early stage even with unlearned handwritten characters and similar characters. can improve the efficiency of recognition processing.

〔Effect of the invention〕

As described above, according to the present invention, it is possible to reduce erroneous recognition of unlearned handwritten characters and similar characters, and the efficiency of recognition processing can be improved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment according to the present invention, FIG. 2 is a flowchart of FIG. 1, FIG. 3 is an explanatory diagram of the embodiment, and FIG. 4 is a block diagram showing a conventional method. In the figure, 4 and 4a are main control units, 5 is an image memory, 6 is a single character extraction circuit, 7 is a feature extraction circuit,
8 is a dictionary memory, 9 is a format information memory,
10, 10a is a recognition circuit, 11 is a judgment circuit, 12
is a normalization circuit, 13 is a normalization character pattern, and 14 is a normalization circuit.
15 is a filler pattern memory, 15 is a standard pattern memory, 16 is a modified pattern generation circuit, 17 is a modified pattern memory, 18 is a matching circuit, and 19 is a matching degree determination circuit.

Claims (1)

[Scope of Claims] 1. A character recognition device that irradiates light onto characters written on a medium and recognizes image data of the characters obtained from reflected light of the irradiated light, comprising: a first recognition means that recognizes based on character characteristics; a first recognition means that recognizes based on character patterns of the image data;
a second recognition means having a filler pattern storage means for storing the recognized character pattern and outputting the character pattern during subsequent recognition; and as a result of the recognition by the first recognition means, the degree of recognition coincidence is lower than a desired value, the corresponding image data is recognized by the second recognition means, and the recognized character pattern is updated and stored in the filler pattern storage means. Device. 2. Claim 1, wherein the second recognition means comprises: normalization means for normalizing the size of the image data; and standard pattern storage means in which a standard pattern of characters is stored. Character recognition device as described in Section. 3. The character according to claim 2, wherein the second recognition means includes a deformation pattern generation means for moving a portion of the normalized image data with a low degree of matching in a predetermined direction. recognition device.