JPS634232B2 - - Google Patents

Description

[Detailed Description of the Invention] The present invention provides a kanji character recognition device, in particular, a candidate category extracting unit that narrows down candidate categories to a small number for input kanji characters, and then Check the division mode according to the structure,
The present invention relates to a kanji character recognition device that performs more detailed recognition of the divided partial patterns.

Recognition devices that recognize kanji characters have a huge number of categories, and many have relatively similar character structures, so it is necessary to narrow down the candidate categories to a small number and then perform more detailed matching. He is brought into recognition.

On the other hand, when recognizing kanji characters, since kanji characters can be classified according to their polarity, side, crown, etc.
Consideration is being given to extracting the candidate categories based on the bias, side, crown, etc., and then performing the detailed matching. However, in the case of a method that performs major classification according to the bias, side, or crown, for example, it is necessary to make sure that the number of characters with the same bias is sufficiently large, or that the input kanji characters are divided according to the bias, side, or crown. However, because the resulting division results are not necessarily correct, sufficient effects cannot be obtained.

In consideration of the above points, the present invention narrows down the candidate categories by another means in advance, and then examines the input kanji characters such as bia, ba, kan, etc. of the candidate category. The objective is to improve the recognition rate by correctly dividing the patterns in association with crowns, etc., and then performing detailed matching on the divided partial patterns. Therefore, the kanji character recognition device of the present invention performs character recognition that has a candidate category extraction unit that extracts features corresponding to input kanji characters and extracts candidate categories based on the contents of a dictionary section prepared in advance. In the device, a character structure dictionary that holds at least the division mode information divided based on the character structure including partial and/or side and/or crown corresponding to each category of characters, and the above-mentioned part for the input kanji character. A partial pattern extraction unit that extracts a pattern, a target partial pattern detection circuit unit that detects a partial pattern to be processed, a partial pattern feature extraction circuit unit that extracts features of the partial pattern to be processed from the input kanji characters, standard It is equipped with a standard partial pattern dictionary that stores partial pattern features and a partial pattern identification circuit section, and for candidate categories extracted by the candidate category extraction section, the content of the character structure dictionary and the partial pattern extraction section are At least a division mode check is performed based on the extraction result, and the partial pattern is identified based on the extraction result from the partial pattern feature extraction circuit and the contents of the standard partial pattern dictionary. It is a feature.
This will be explained below with reference to the drawings.

FIG. 1 is an explanatory diagram conceptually explaining a recognition processing mode according to the present invention, and FIG. 2 shows the configuration of an embodiment of the present invention.

Assume now that a Kanji character 1 is input and candidate categories 3-1, 3-2, and 3-3 are extracted by the candidate category extraction section 2, as shown in FIG. On the other hand, the character structure dictionary 4 contains division direction (division mode) information and partial patterns as a result of division when dividing characters according to character structures such as polarity, side, and crown, corresponding to each category character. Provided.

The character structure dictionary 4 is indexed based on the candidate categories 3-1 to 3-3, and partial patterns 5-1 to 5-3 convenient for identifying the candidate categories 3-1 to 3-3 are extracted. do. During this time, the input kanji character 1 is divided according to the character structure such as the above-mentioned bias, side, and crown in the partial pattern extraction section, and the partial pattern is extracted based on the contents of the character structure dictionary 4. The division positions and manner of division determined by the division are checked.

In this check, not only the division mode is checked, but also each partial pattern of the divided results is checked as necessary, and some noise contained in the divided results is corrected by the partial pattern extraction section. You can do it like this. Of course, if the input kanji character 1 cannot be divided by the partial pattern extraction section, or if the input kanji character 1 cannot be divided, the partial pattern and the candidate categories 3-1 to 3-3 extracted from the character structure dictionary 4 If the pattern is extremely different, the extraction result by the candidate category extraction section 2 will be output as a recognition result, or used for other purposes.

As mentioned above, the partial pattern for the input kanji character 1 falls into the candidate categories 3-1 to 3.
If there is a portion that matches the partial pattern -3, more detailed matching processing begins for the other partial patterns (in the illustrated case, partial patterns 5-1 to 5-3). That is, features are extracted for the portion of the input Kanji character 1 from which the partial pattern ``word'' has been removed, and the extracted features are compared with the contents of a standard partial pattern dictionary 6 prepared separately. In the case shown in Figure 1, it goes without saying that the partial pattern of the input kanji character 1 is ``Shin''.
The characters are compared and recognized as characters belonging to the category "recognized".

FIG. 2 shows the configuration of an embodiment of the present invention. Reference numerals 1, 2, 4, and 6 in the figure correspond to those in FIG. 1, 7 is a partial pattern extraction section, 8 is a normalization circuit section, 9 is a feature extraction circuit section, 10 is a dictionary, 11 is an identification circuit section, 12
1 is a peripheral distribution calculation circuit unit which examines the character peripheral distribution of the input kanji character 1 and prepares for division position detection, 13 is a character division position detection circuit unit, 14 is a division determination circuit unit, 15 16 is a target partial pattern detection circuit section, 16 is a normalization circuit section,
Reference numeral 17 represents a partial pattern feature extraction circuit, 18 represents a partial pattern identification circuit, and 19-1 to 19-3 each represent a minimum mismatch detection circuit.

As is conventionally known, the candidate category extraction unit 2 identifies the input Kanji character 1 based on the contents of the dictionary 10, and extracts candidate categories 3-1 to 3-3 as shown in FIG. Based on this result, information corresponding to the candidate categories 3-1 to 3-3 stored in the character structure dictionary 4 is extracted.

Meanwhile, for the input Kanji character 1, the division position and division direction are extracted by the partial pattern extraction section 7 including the peripheral distribution calculation circuit section 12 and the character division position detection circuit section 13, as is conventionally known. be done. Based on the information extracted from the character structure dictionary 4, the division determination circuit 14 checks whether information such as the division direction matches. The target partial pattern detection circuit section 15 detects the portion of the partial pattern that is to be the object of subsequent comparison, and instructs to extract the corresponding partial pattern from the input Kanji character 1. In the case shown in FIG. 1, the video signal of the partial pattern "Ninja" is extracted and supplied to the normalization circuit section 16. Then, the partial pattern feature extraction circuit section 17 extracts the features regarding the "ninja". Then, the standard partial pattern read from the standard partial pattern dictionary 6 is compared with the standard partial pattern in the identification circuit section 18. In the case shown in FIG. 1, it can be considered that the standard partial patterns "Shin", "Sale", and "〓" are compared. In the case shown in FIG. 1, the standard partial pattern "Shinobu" is most often matched, and the answer is obtained through the minimum mismatch detection circuit section 19-3.

Note that if the partial pattern extraction section 7 does not detect the division itself, the minimum mismatch detection circuit section 19-1 obtains an answer. Further, in cases where the division manners are significantly different, an answer can be obtained in the minimum mismatch detection circuit section 19-2.
Of course, the minimum discrepancy detection circuit units 19-1 to 19
-3 results may be combined to make a final recognition determination.

As described above, according to the present invention, recognition processing using character structures such as slant, side, and crown is used for more detailed recognition processing after major classification. Therefore, it is possible to check the character structures such as the above-mentioned bias, side, and crown for the number of candidate categories narrowed down by the above-mentioned major classification. For this reason, even if the process of the partial pattern extracting unit 7 that divides the input kanji character 1 contains some noise, the correct partial pattern will be extracted in subsequent partial pattern extraction, and the correct partial pattern will be extracted and compared with the standard partial pattern. It becomes possible to perform verification. Furthermore, since the division determination is made once, it is less likely that the comparison will be performed in an undesired manner when comparing with the standard partial pattern.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram conceptually explaining a recognition processing mode according to the present invention, and FIG. 2 shows the configuration of an embodiment of the present invention. In the figure, 1 is an input kanji character, 2 is a candidate category extraction unit, 6 is a standard partial pattern dictionary, 7 is a partial pattern extraction unit, 14 is a division determination circuit unit, 15
17 represents a partial pattern detection circuit, 17 represents a partial pattern feature extraction circuit, and 18 represents a partial pattern identification circuit.

Claims (1)

[Claims] 1 In a character recognition device that has a candidate category extraction section that extracts features corresponding to input kanji characters and extracts candidate categories based on the contents of a dictionary section prepared in advance, A character structure dictionary that holds at least division mode information divided based on a character structure including polarity and/or side and/or crown, a partial pattern extraction unit that extracts the partial pattern for the input kanji character, and a processing target. a target partial pattern detection circuit unit that detects a partial pattern;
It is equipped with a partial pattern feature extraction circuit section for extracting the features of the processing target partial pattern in the input Kanji character, a standard partial pattern dictionary for storing the standard partial pattern features, and a partial pattern identification circuit section, Regarding the candidate categories extracted by the candidate category extraction section, at least a division mode check is performed based on the contents of the character structure dictionary and the extraction results by the partial pattern extraction section, and the partial pattern feature extraction circuit section A kanji character recognition device characterized in that the partial pattern is identified based on the extraction result and the contents of the standard partial pattern dictionary.