JPS5985579A - Forming method of individual character form dictionary - Google Patents

Abstract

PURPOSE:To obtain a new dictionary of individual character form by having the weighted sum between the feature showing the character form obtained from a dictionary for many and unspecified persons and the feature of individual character supplied previously in response to the number of individual character so far obtained. CONSTITUTION:The feature X and its JIS code of individual character supplied from a character recognizer are stored in temporary storage devices 6 and 7 respectively. Here a dictionary F produced from many and unspecified persons is stored in a dictionary storage device 2 for many and unspecified persons. While the number N of supplied individual character is stored in an individual character number storage device 3 at a prescribed position. Then an individual character dictionary G containing the feature of supplied individual character is provided to an individual character storage device 4. Then the JIS code of the device 7 is read out by a controller 1, and therefore the contents corresponding to the devices 3 and 4 are read out. Then the contents N of the device 3 are updated with N+1, and the dictionary F of the device 2 and the updated contents G of the device 4 are updated and calculated 5 to transfer 8 an updated dictionary.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method of configuring a personal glyph dictionary included in a personal character recognition device intended for specific scribes. One example of a personal glyph dictionary that can be used is to collect a large number of characters from a target scribe and use the average to form a dictionary. but,
In this conventional method, a scribe must write down all categories to be recognized multiple times in order to obtain the desired recognition performance.1. Especially when the target is handwritten kanji, it is difficult to write down such a large number of characters. While the number of times it was written down was small, the correct reading rate was low, and it was considered a drawback that significantly diminished the value of the IM Humanities Character Recognition System.

<Summary of the Invention> In order to solve these drawbacks, the present invention has a dictionary created from a specific number of membrane pressure specific scribes, and as the target specific scribe inputs characters, personal glyph shapes are added to the dictionary. If necessary, even for characters that the scribe has not yet written, the shape of the characters is estimated from the characters that have already been written, and the dictionary is modified to suit the scribe. The purpose of this invention is to provide a method for configuring a personal glyph dictionary.

According to the present invention, when a character is input from a specified scribe to a personal character recognition device, a dictionary for general public is referred to based on the code indicated by the correct input character, and a characteristic representing the glyph shape of the character obtained from this is obtained; A new dictionary for the character is created by calculating a weighted sum of features of the same kind of the previously input personal input character according to the number of personal characters obtained so far.

In this way, as the number of times the same character is input increases, the recognition rate for that character improves. , Furthermore, select a number of glyphs similar to the glyph characteristics of the personally input characters from the scribe dictionary, and select from the glyph characteristics of other characters by the scribe of the selected glyphs that have not been previously input by the specific guest reporter. Create a new dictionary for characters.

<Embodiment> FIG. 1 shows an embodiment of the present invention, which is a personal glyph dictionary that modifies a dictionary of input character categories to suit the scribe as the target scribe inputs characters. Components are shown.

A control device 1 that controls the entire device; - A dictionary storage device 2 for unspecified users that holds a dictionary for specific and large numbers of membrane distortion; and a personal character count that stores how many personal characters have been input for each category across all categories to be read. A storage device 3 stores characteristics of a particular scribe's personal glyphs across all target categories, and a personal character feature storage device 4 stores the characteristics of a specific scribe's personal glyphs, respectively.Using the contents of storage devices 2 and 3゜4, A dictionary update calculation device 5 that calculates a new dictionary, a personal character feature temporary storage device 6 that temporarily stores the features of personal characters, a code indicating the personal character stored in the personal character feature temporary storage device 6,
For example, a personal character code temporary storage device 7 that temporarily stores JIS codes, and an updated dictionary transfer device 8 that transfers the contents of a new dictionary created by the dictionary update calculation device 5 from this dictionary construction device to the Sha character recognition device. are connected to a common path line 101 controlled by the control device 1. Device 2, 3,
Transmission and reception of data between 4°5.6 and 7.8, and control of starting and stopping are performed through the path line 101. The personal character features input from the character recognition device through the personal character feature input terminal 201 are transferred to the personal character feature temporary storage device 6, and the JIS code of the personal character input from the character recognition device through the personal character JIS code input terminal 202 is transferred to the personal character feature temporary storage device 6. is transferred to the personal character code temporary storage device 7. The contents of the new dictionary are transferred to the character recognition device through the updated dictionary output terminal 203 under the control of the updated dictionary transfer device 8.

In order to operate this dictionary construction device, first, the characteristics of the personal characters input from the character recognition device and their JIS codes are stored in temporary storage devices 6 and 7, respectively. This JIS code is a character recognized by a character recognition device, and if the recognition is incorrect, it is corrected by the individual inputting person.

The personal character feature temporary storage device 6 stores, for example, the stroke density feature X=(x
l, x2 . . . Xl . . .
iS code 3441 (hexadecimal) is stored. In the location corresponding to "Kan" in the unspecified majority dictionary storage device 2, there is a dictionary F- (f+
, f2...fl...fn) are stored, and the personal character number storage device 3 stores the number N of the personal character "Pakan"' that has already been input in a predetermined location, and the personal character characteristics In the storage device 4, the characteristics of the personal character kanji that have already been input are added.G=(g+r g2...gi...
・gn) is stored. However, at the initial stage when personal characters are not input, it is N−〇, and G=(0,
0...0).

Thereafter, the following operations are sequentially performed under the control of the control device 1. The JIS code stored in the personal character date temporary storage device 7 is read out, and the corresponding contents of the personal character number storage device 32 and the personal character feature storage device 4 are read out accordingly. Next, the contents of the personal character count storage device 3 (″
The number of "Kan") N is updated to N + 1. The contents of the personal character feature storage device 4 (the number of "Kan" features added) are:
G′=(g g6...gl...g included), g
'i=gi+Xi, and this value is newly updated to G-(g
+ + g2...gl...gn).

Next, using the dictionary F representing "Kan" and the updated G, a new dictionary Y-(yI, y2...yl ...Vn) +
y1-tN×(1/NXgi)+(1,0-tN)×f
1 is calculated by the dictionary update calculation device 5, and the calculation result is transferred as an update dictionary to the location corresponding to the character "Pakan" in the dictionary in the character recognition device through the update dictionary transfer device 8. Here, the dictionary update calculation is performed. tN is a numerical value between 00 and 1.0, and is determined depending on the number of input characters N,
This parameter tN is a parameter for the weighted average of personal characters and an unspecified large number of dictionaries, and is held in the dictionary update calculation device 5. This parameter tN is set so that the recognition rate is highest for each number of characters N. be done.

Since this device operates in this way, it is possible to construct an effective dictionary according to the number of individual characters obtained for each category (character).

FIG. 2 shows an apparatus for estimating a character shape of a category whose personal character shape is unknown from a character shape of a category whose personal character shape is known, and modifying a dictionary of the unknown character shape category.

In FIG. 2, parts corresponding to those in FIG. 1 are given the same reference numerals. Multiple general scribes extract the characteristics of each character from the data written in all target categories,
The multiple scribe character feature storage device 9 stores all of them individually, the characteristics of the already obtained personal character shapes stored in the personal character feature storage device 4, and the already obtained characteristics stored in the personal character number storage device 3. A pass line 101 includes a character shape estimation calculation device 10 that estimates the shape of a nonotegori whose personal character shape is unknown using the number of individual characters that have been written and the characteristics of the multiple scribe characters stored in the multiple scribe character feature storage device 9.
connected to.

The character shape estimation calculation device 10 sequentially performs the following operations. The personal character count storage device 3 is read, and the category CITl (m-1 to
Find M). For each category Cm, the contents G(Cm)−(g(Cm)
t 2g (Cm)2 ・・・・・・g(C1Tl)1
-... g(Crrl)n) is read out, and this content is divided by the number of characters N stored in the personal character number storage device 3.

That is, H (Cm,) - (h (Cm) + , h (
cml)2”””h(Cm)i”””h(Cm)n)
,.

Find h(Cm)1-(1/N)×g(Cm)1. This feature H(Cm) and the feature Z(Cm) −(z(Cm)+ , Z(Cm)z
...Z(Cm)i...Z(Cm)n)
Compare with. In other words, the unselected features H(Cm) and Z(Cm
) and the distance d(H(Cm).

Calculate Z(Cm))-4e, 1h(C), -Z(C), l.

ml 11 1 Furthermore, as a measure of average comparison over all Cm (m=1 to M), D (n, z) = mg1a(n
(cm), Z(Cm:)). According to this standard of measurement, among the multiple scribes stored in the device 9, the top 0 scribes whose character shape is similar to this personal character are
Select a person. Characteristics of selected scribe 0 qZ-(Q
Z,.

qZ2...qzl...qZn) (q-1~
Q) is used to estimate the estimated feature P = (P+, P2-...Pi...-Pn) for the unknown category of individual glyphs, that is, the characters that have not been input yet.
, Pi=1/QXΣqZ1q=1. Next, this estimated feature is considered in the same way as the personal character feature obtained directly from the scribe, and is input into the personal character feature storage device 4, and the dictionary is modified by the method explained in FIG. In this case, the estimated feature P may be used as a new dictionary.

Because of this structure, even when the glyphs for some categories of scribes are obtained, for the glyphs of categories that are not written in the trenches, the glyphs of the scribe who write the same + shape can be substituted, It is possible to obtain a higher recognition rate than when using a dictionary for general public as is.

<Effects> As explained above, according to the present invention, it is possible to write down a large number of individual characters in advance, and as the scribe inputs characters, the scribe's shape is efficiently input, and the individual A dictionary that reflects gender will be constructed. That is,
The character recognition device equipped with the device according to the present invention has a dictionary suitable for the general public in the initial state, but when a scribe inputs characters, along with the JIS code of the recognition result (if the recognition is incorrect) When input character features are input into the device according to the method of the present invention, personal character features are weighted to maximize the recognition rate according to the number of input characters. A dictionary is created by taking a weighted average of the general dictionary and the general dictionary. Furthermore, it takes a lot of time to input all the characters to be recognized multiple times, but even if only some categories have been input, categories that have not yet been written can still have similar scribe shapes. This method has the advantage that the recognition rate can be automatically improved by estimating the features and modifying the dictionary.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of an apparatus using the method of this invention, and FIG. 2 is a block diagram showing an embodiment of an apparatus to which a partial modification of the method of this invention is applied. 1 Control device, 2... Dictionary storage device for general public, 3
・・Personal character number storage device, 4.Personal character feature storage device,
5...Dictionary update calculation device, 6...Personal character feature temporary storage device, 7..Personal character JIS code temporary storage device, 8
...update dictionary transfer device, 9.multiple scribe character feature storage device, 10.character shape estimation wave q2 device, 101..common path line, 201 individual character feature input terminal, 202
... Personal character JIS coat input terminal, 203... Update dictionary output terminal. Patent applicant: Nippon Telegraph and Telephone Corporation Agent: Takashi Kusano, Figure 1

Claims (2)

[Claims] (1) In a personal character recognition device equipped with a personal character recognition device targeted at specific scribes, -membrane distortion is created from the characters of a large number of specific scribes, and stores the characteristics representing the shape of each character. as the characters of the specified scribe are input into the personal character recognition device, the code indicating the correct input character is used to refer to the dictionary for the general public; By performing a weighted summation of the features of a dictionary suitable for an unspecified large number of input characters and the features of the same type as the dictionary extracted from the input personal characters, depending on the number of personal characters input in the past, the new character is A method for configuring a personal glyph dictionary, which is characterized by creating a dictionary. (2) Describe the characteristics that represent the shape of each character written by an unspecified number of scribes.1. A multiple scribe dictionary is provided, which selects from the multiple scribe dictionary a character shape similar to the characteristics of the input character, and based on the characteristics of the character shape of the scribe of the selected character shape, the input character of a specific scribe is selected. A personal glyph dictionary construction method according to claim (1), characterized in that a new dictionary is created for characters that are not included in the personal glyph dictionary.