JPS6059487A - Recognizer of handwritten character - Google Patents

Abstract

PURPOSE:To call out an erroneous recognition pattern having the highest degree of resemblance and to delete it by using the character code of an input pattern which produces the erroneous recognition and a pattern to be deleted as an input to perform matching again likely at a recognition mode. CONSTITUTION:The resemblance is calculated between an input character pattern and each dictionary pattern in a dictionary 10 by a resemblance calculating circuit 5. Then these patterns are sorted in the order of higher degrees of resemblance of each pair of serial numbers of patterns and a corresponding character code. The result of this sorting is sent to a deciding circuit 9, and the extracted character code is sent to a display device. When an undesired pattern is selected and then deleted, it bears the highest resemblance to the input character pattern. In this case, the input character pattern retrieved out of a memory 6 by a misread input character number is sent again to the circuit 5 to delet the undesired character. Then the serial number of a pattern having the same character code as that to be deleted by a deletion control circuit 8 is selected out of the output of the circuit 5 to delete.

Description

Detailed Description of the Invention (1) Technical Field of the Invention The present invention has a function of registering personal character patterns, and for a dictionary in which one character has multiple dictionary patterns, the degree of similarity with an input handwritten character pattern is determined. The present invention relates to a handwritten character recognition device that includes a recognition unit that calculates and determines.

(2) Takeki's technology and problems Generally speaking, it is well known that when it comes to character 1g knowledge, a dictionary based on the user's individual character shapes (patterns) shows a higher IUA rate than a general-purpose dictionary. ing. However, it is difficult to construct a personal dictionary of 2000 character types before using the recognition device.

Therefore, it is conceivable to use a general-purpose dictionary at first, and then register individual glyphs one by one during actual operation.

There are two possible methods for registering personal glyphs: One is to completely replace the previously registered glyph with the glyph you are about to register, and the other is to leave the glyph that was registered as is.
This is a new method of additional registration. The former is a single character type (
Although there is always only one pattern in the dictionary for each category (category), the latter method is better in order to improve the recognition rate, since handwritten character shapes vary widely even among individuals. In this case, a plurality of dictionary patterns exist for one category in the dictionary.

A problem that arises when implementing a recognition device using such a registration method is the existence of unnecessary patterns that are registered due to user's erroneous operations. It is desirable that such patterns can be deleted, since they may be unnecessary or cause misunderstandings.

Usually, users become aware of the existence of such unnecessary and harmful patterns when erroneous recognition occurs due to such patterns. At this point, all we know about this pattern is its character code.

However, for one character code (category),
Since there are multiple dictionaries, it is not possible to specify the character pattern to be deleted using the character code alone.

Conventionally, when specifying a turn to be deleted from among patterns having a plurality of identical character codes, 1) all dictionary turns having the same character code are displayed.

2) Have them specify the appropriate turn from among them.

Such a method was adopted.

However, with this method.

1) It is necessary to display the patterns in the dictionary in a way that users can understand. If the glyphs at the time of registration cannot be reproduced, it is almost impossible to force normal users to specify unnecessary patterns. .

2) Even when the registered character shape is reproduced, an expensive graphic output device is required to visually express it.

3) Deleting one pattern requires several steps of operations.

There were many shortcomings.

(3) Purpose of the Invention The purpose of the present invention is to provide a function for deleting erroneously recognized patterns, thereby eliminating dictionary patterns that have an adverse effect on the recognition of personal character patterns and recognizing handwritten characters with high precision. The object of the present invention is to provide a recognition device.

(4) Structure of the Invention In order to achieve the above-mentioned object, the handwritten character recognition device of the present invention has an input section into which a character pattern is input, and a dictionary pattern in which a character =-do indicating a category of 23 characters is added. a dictionary, a matching unit that matches the input character pattern with the dictionary pattern, and a display that displays a character corresponding to a character code assigned to a dictionary pattern that the matching unit determines is most similar. an instruction input section for instructing deletion of the dictionary pattern that caused the misrecognition when the characters displayed on the display section are misrecognized; and the instruction input section. The present invention is characterized by comprising means for searching the dictionary and retrieving the dictionary pattern that caused the erroneous recognition based on an instruction from the department, and deleting the dictionary pattern obtained by the medal search means.

(5) Embodiments of the Invention In general, users wish to register their personal glyphs when an input glyph is misrecognized.

If you rewrite the character you are trying to register at that point, the shape of the character will often be different from the previous one.
In order to register the glyph itself that caused misrecognition, it is necessary to save the input glyph turns for a certain amount of time.

Therefore, when considering the deletion of unnecessary patterns, the input pattern that caused the erroneous recognition that led to the deletion is
If you think that it is preserved, you will be able to do it. In that case, the unnecessary dictionary word (turn) becomes the first candidate because it has a higher degree of similarity than the category (character type) originally intended by the user, so if we match it with this input word turn again, we get It should show an extremely high degree of similarity.

In other words, the input cover turn that caused the misrecognition and the character code of the pattern to be deleted are input, and matching is performed again in the same way as during recognition. What is necessary is to delete the one with the highest degree of similarity.

In this method, the circuit required for deletion is either a recognition unit or an input pattern storage circuit.

Since it is possible to use almost everything that already exists, it can be easily realized.

The figure is a configuration explanatory diagram of an embodiment of the present invention.

In the figure, handwritten character data input from a tablet 1 is separated by a circuit 2 into input character data and command data.

The input character data is subjected to preprocessing by the feature extraction circuit 3, and an input character pattern is output. At this stage, each input character pattern is given a number (
jd) is given.

The input character pattern is sent to the similarity calculation circuit 5, which calculates the degree of similarity with each dictionary pattern in the dictionary 10, and sets the serial number of this dictionary pattern in the dictionary and the corresponding character code as a set to calculate the similarity. The results are sent to the judgment circuit 9, which extracts one or more character codes from among them and sends the results to the display device 11 as recognized character codes. 0 In addition, the input cover turn is stored in the input character no. turn storage device 6.
On the other hand, the input command data is solved by the command interpretation circuit 4, and if the command is registration or deletion, it is stored in the respective control circuits. Sent to 7.8.

In the case of registration, under the control of the registration control circuit 7, the input character pattern stored in the storage device 6 is searched based on the input character number, the character to be registered is added to it, and the dictionary 10 Store in.

In the case of deletion, the following operation is performed under the control of the deletion control circuit 8 using the input character number that was misread and the character code of the notation turn to be deleted.

The input character pattern retrieved from the storage device 6 based on the input character number is sent again to the similarity calculation circuit 5, and the pattern -
Matching is performed. The sets of character codes corresponding to the serial numbers of each pattern in the dictionary, sorted into one pattern with a high degree of similarity, are sent to the determination circuit 9.

On the other hand, the character code to be deleted has already been sent to the determination circuit 9 from the deletion control circuit 8, and in that case, the determination circuit 9 will be able to confirm the serial number and character code sent to the similarity calculation circuit 5. From the sequence of pairs, select the one that has the same character code as the character code to be deleted and has the highest degree of similarity, and use that serial number as the serial number of the dictionary pattern to be deleted.
Issue a deletion request to 0. The dictionary 10 deletes the dictionary pattern having the requested serial number.

(6) As described in detail, dictionary patterns that are considered unnecessary or harmful for recognition can be efficiently deleted using the same procedure as registration. Therefore, even if a character pattern that causes erroneous recognition is registered due to an erroneous operation during registration, it can be deleted by inputting a similar character pattern and performing the same number-turn matching as recognition. Registration is easy, increases the accuracy of the dictionary, and eliminates unnecessary
Efficient use of storage devices can be achieved by eliminating harmful memories0

[Brief explanation of the drawing]

The figure is an explanatory diagram of the configuration of an embodiment of the present invention, in which 1 is a tablet, 2 is an input segmentation circuit, 3 is a feature extraction circuit, 4 is a command interpretation circuit, 5 is a similarity degree calculation circuit, 6 indicates an input character turn storage device, 7 indicates a registration control circuit, 8 indicates a deletion control circuit, 9 indicates a determination circuit, 10 indicates a dictionary, and 11 indicates a display device 0 Patent applicant Fujitsu Limited Successor applicant Patent attorney 1 ) Yoshishige Saka

Claims (1)

[Claims] a Kukata part into which character bata ≦7 is input; a dictionary in which a dictionary pattern to which a character code indicating a character category is assigned is stored; and a collation part to match the Kukata character pattern with the dictionary pattern. , a display section that displays a character corresponding to a character code given to a dictionary pattern that is determined to be most similar by the matching section, the handwritten character recognition device comprising: an instruction input section for instructing deletion of the dictionary pattern that caused the misrecognition when a character is misrecognized; and an instruction input section that searches the dictionary for the dictionary pattern that caused the misrecognition based on the instruction from the instruction input section. 1. A handwritten character recognition device, comprising: means for extracting dictionary patterns, and deleting dictionary patterns obtained by the search means.