JPS5949675A - Character recognition device - Google Patents

Abstract

PURPOSE:To generate a similar character discrimination dictionary containing erroneous recognition data and to improve the precision and speed of character recognition by storing the correspondence between character categories obtained by erroneous recognition and input character patterns which are objects of erroneous recognition. CONSTITUTION:An input character is converted into binary data by a tablet 1 and a recognition part 4 refers to the 1st and the 2nd dictionaries 2 and 3 to find recognition candidate characters according to similarity. Those recognition candidate characters are stored in a storage part 5 temporarily according to the similarity and the contents of the storage part 5 are displayed on a display part 6 according to the operation of an operation part 7. A storage part 8 is provided corresponding to the recognition part 4 and storage part 5 and the correspondence between erroneously recognized candidate characters and input character categories which are objects of erroneous recognition is stored therein. Then, character categories obtained by erroneous recognition and input character patterns which are objects of erroneous recognition are stored corresponding to each other to improve the precision and speed of the character recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Technical Field of the Invention The present invention relates to a character recognition device, and particularly to a character recognition device for handwritten characters.

(B) Technical Background In order to improve the recognition accuracy and speed of the handwritten character reading system B7, a method of limiting the writing style has been used for a long time, but as the user base expands, this method has been used for a long time. Like f! In response to this current situation, in addition to a dictionary that memorizes the characteristics of Pp quasi-character patterns, we have created a dictionary that can identify characters that are particularly likely to cause the error B4. For this reason, the character ml code station 11.t, which has a dictionary for similar characters 11 and 11, is used.

(0 Prior art and problems) A dictionary for identifying characters from the above-mentioned category A, for example, ``ne'', ``reJ, ゎj yo /Ii-61 similar characters that are easy to recognize from each other. It is provided for the purpose of identification.
Conventionally, for each similar character group, each standard character pattern or character pattern by specific 66 people and 4 other characters were used.
Based on the characteristics of the moon, I used the one created by JLL.

However, in general, in handwritten character patterns,
：The peculiarities of the Yt method are strongly expressed, and the power that has been applied to the character pattern by an unspecified person. ” had the disadvantage that it could not be adequately addressed.

α)) Purpose of the Invention The object of the present invention is to eliminate the drawbacks of the conventional example and provide a character recognition device that can handle character pattern recognition by an unspecified person.

(F) Structure of the Invention Characters according to the present invention; the four-way recognition device has IJ i, M! Character W that displays the H complementary character category and corrects misrecognition by visual inspection? Character category and nuclear error Nt obtained by misrecognition in J recognition device? The erroneous recognition data recording section records the input character pattern that is the target of &, and the erroneous recognition data recording section records the erroneous ? A similar character identification dictionary generated from AM data is provided.

(n. Fruits of the Invention) Below, the advantages of the present invention will be specifically clarified as ii by way of examples.

Figure 1 shows how Komaaki Honkoma is connected to the online handwritten letter dismissal device d.
1 shows an example of a 4-go weave that was used, 1 is a tablet that is manually processed and converted into binary data, and 2 is a standard character pattern special P k Sr: memorization for each character category. A first dictionary, 3 a second dictionary used as a similar character identification dictionary, 4 a human character pattern pattern 11 on the tablet 1, which is compared with at least one of the first dictionary 2 and the second dictionary 3 to identify similarities. The recognition unit calculates the 1st category of characters up to the 5th place in order of degree as recognition candidate characters, 5 is ng lij. Remember*! 1 storage unit, 6 is 21K 1 'ty
> 1 for each human-powered character pattern in the order of similarity according to the operation unit's operation unit temporarily stored in the storage unit 5.
A display unit that displays each character as t≧ and also displays the display illustrated in FIG. 2 below.

7, the operator visually determines whether the recognition candidate characters displayed on the display unit 6 are correct or not, and selects the correct one.

If this is determined as the recognition result (f), and it is an error, the recognition candidate character with the next highest degree of similarity is displayed on the display section 6, and the Rei Shikiura character that was incorrectly recognized and the incorrectly recognized character. The input character category that was the target of 4.
Note 2 1. <i J-★part to be recorded in Mli, 8 is the previous 1rj iig , i, , erroneous recognition in which the recognition candidate character that was recognized and the input character category that was the target of the erroneous recognition are recorded in correspondence. Recognition of input character patterns is performed by the second storage section, which is used as a data recording section and has a configuration of 0 or more. Next, the generation of the second dictionary 3 used as a similar character identification dictionary will be explained.

For example, the first step is to recognize the input character pattern ``ne''.
The character category “wa” is obtained as a candidate character for recognition.
If this is an erroneous recognition, the character category "wa" and the input character category "ne" are associated and recorded in the second storage unit 8 as (wa-ne). Next, the input character pattern "re" is recognized f) 'r 1's Fi4 #I#
If the character category "wa" is obtained as a complementary character and this is an erroneous iWR, the erroneous recognition data (Wa-ne) wo (Wa-ne) recorded in the storage unit 8 of f:1G2.
Like ne, re) - to renew.

In this way, after obtaining erroneous recognition data for several character categories, the process moves to the generation mode of the second dictionary 3. FIG. 2 shows the contents displayed on the display unit 6 when generating the second dictionary 3. , 61 is an erroneous recognition data display section in which the erroneous recognition data recorded in the second storage section 8 is displayed for each character category. 062 is a character number display that displays the character number marked with - in the feature menu to be described later. Part 63 is a feature menu, which displays features in handwritten strokes that should be noted in order to identify characters that are likely to be misprinted, such as "loop at the end of a stroke" and "intersection of two strokes."&- display section, 64 and 65t constitute a registration table display section, and 64 is p14 ng identification data display section 6
1st place displayed in 1:+ij Hf&R% complementary character (
1) in the example of the burial mound), and which can be compared with the input character category that was the direct target of misrecognition (in the illustrated example, ``ne'' and ``re''). - is selected from the feature menu display section 63 by the operator, and the feature menu 1- is displayed according to the character number displayed on the character number display section 62.

Reference numeral 65 is a stroke display section in which the stroke number of the input character pattern in which the feature displayed on the feature display section 64 appears is displayed according to an operator's instruction.

The feature menu displayed on the feature menu display section 63 can be commonly applied to the character categories of each JM, and therefore is created in advance and used in a fixed manner. On the other hand, the display contents of the registration table, that is, the characteristic display section 64 and the stroke display section 65, can be selected and instructed by the operator based on the display contents of the misrecognition data display section 61 and the feature menu display section 63. At this time, the second
A dictionary 3 is generated.

In this way, the second dictionary 3 for each character category is generated using the error data accumulated every time an error +u4 iil occurs in the recognition of input character patterns up to now. Once this is complete, the program proceeds to normal character recognition mode.

Each time such a cycle is repeated, the contents of the second dictionary 3 become adapted to the peculiar Chinese writing habits, thus improving the recognition speed by 117 degrees.

0) As described in detail of the invention, according to the present invention, a dictionary for each similar character HA is generated and enriched each time it is used, and it is possible to deal with character patterns created by unspecified writers.

[Brief explanation of the drawing]

FIG. 1 shows a system block diagram of an embodiment of the present invention,
Reference numeral 3 designates a second dictionary used as a similar character identification dictionary, and reference numeral 8 designates a second storage unit used as a data recording unit. ′
Further, FIG. 2 is an explanatory diagram regarding the generation of jisai 3 of t32.

Claims (1)

[Claims] Is the input character pattern recognized? In a character recognition device that displays and corrects misrecognition by visual inspection, the character category obtained in misrecognition and the input character pattern that was the subject of misrecognition. What is claimed is: 1. A character recognition device comprising: an erroneous recognition data recording section for recording erroneous recognition data in correspondence; and a similar character identification dictionary generated from the erroneous recognition data recorded in the erroneous recognition data recording section.