JPH1027218A - Method for evaluating on-line handwritten character pattern data base - Google Patents

Abstract

PROBLEM TO BE SOLVED: To quantitatively evaluate an on-line handwritten character pattern data base by defining information for quality evaluation based on the quantitative information of the handwritten pattern data of a data base. SOLUTION: A quality information definition part 203 receives the quantitative primary information of the data of a handwriting pattern from the data base 201 and defines the information for the quality evaluation of the data base 201 based on it. The information of quality defined in the quality information definition part 203 is stored in the data base 201 or stored in a different file as indicated by a broken line and is referred to at any time on the side of an on-line handwritten character recognition development system 200. Thus, a clear objective reference for evaluating the result of the research and development of a handwritten character recognition technique utilizing the on-line handwritten character pattern data base is provided and substantial contribution is made to the research and development of the handwritten character recognition technique.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to the field of online handwritten character recognition, and more particularly, to the quality evaluation of an online handwritten character pattern database.

[0002]

2. Description of the Related Art A general flow of research and development of online handwritten character recognition can be shown in FIG. First,
For example, handwriting data (time series of xy coordinate values) of handwritten characters is input using an online handwritten character input unit such as a tablet and an electronic pen, and stored in the data file 101 (step 100). The handwriting data of the handwritten character handwriting stored in the data file 101 is organized and an online handwritten character pattern database 103 is constructed (step 102). Using this online handwritten character pattern database 103, research and development of online handwritten character recognition is performed (step 10).
4), the result 105 is evaluated (step 10)
6). Steps 104 and 106 are repeated.

[0003] Regarding the construction of an online handwritten character pattern database, it is necessary to collect a considerable amount of handwriting patterns to which realistic font deformation is added in order to enhance the recognition of online handwritten characters. With this in mind, there has been proposed a method for creating an online handwritten character pattern database with no restrictions on text format and fonts (Masaki Nakagawa et al., "Collection and Use of Online Handwritten Character Patterns without Text Format Fonts", IEICE. Technical Report, PRU95-110, pp. 43-48 (199
5)).

[0004]

Various methods for online handwritten character recognition have been proposed, and their advantages and disadvantages are discussed. However, if it is unclear what kind of online handwritten character pattern database was used to develop the online handwritten character recognition method, it is impossible to objectively evaluate the method.

In the above-described conventional method of constructing an online handwritten character pattern database, the quantitative primary information (for example, the number of writers, the number of character types, the number of handwritten patterns) of the data of handwritten character handwriting patterns is provided. Etc.) are given. However, since it does not provide quality information (eg, politeness of handwritten characters, variations in the number of patterns for each character type, etc.), it is necessary to evaluate the results of research and development of online handwritten character recognition technology using a database. The objective criteria were unclear.

It is an object of the present invention to provide a method for evaluating the quality of an online handwritten character pattern database in order to solve the above-mentioned problems.

[0007]

The gist of the method for evaluating an online handwritten character pattern database according to the present invention is to process quantitative primary information of handwriting pattern data in the database to obtain secondary information, The information for quality evaluation of the online handwritten character pattern database is defined based on the information (claim 1).

The information for quality evaluation to be defined includes, specifically, information on the frequency of appearance of handwriting patterns of each character type (Claim 2) and the quality of handwriting patterns of each character type in terms of the number of strokes. (Claim 3), information indicating the quality of handwriting patterns of all character types based on the number of strokes (Claim 4), and information indicating the quality of handwriting patterns of each character type of each writer based on the number of strokes (Claim 4). Item 5), information indicating the quality of writing patterns of all character types of each writer as viewed from the number of strokes (Claim 6), and information indicating the degree of difference in the quality of handwriting patterns of each writer as viewed from the number of strokes (Claim 6) Item 7),
Information indicating the relative evaluation of the quality of the strokes of the handwriting patterns of each writer in terms of the number of strokes of the handwriting patterns of all the scribes (claim 8), and / or
Alternatively, at least one of information (claim 9) indicating a relative evaluation of the quality of the handwriting pattern of each writer group in terms of the number of strokes of the handwriting pattern with respect to the quality of the handwriting pattern of all the writer in terms of the number of strokes is included. It will be apparent from the following description that other information may be included.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be specifically described below. FIG. 2 is a block diagram for explaining an embodiment of the present invention. Reference numeral 200 denotes an online handwritten character recognition development system, and reference numeral 201 denotes an online handwritten character pattern database (hereinafter, abbreviated as a database) used in the online handwritten character recognition development system 200. The construction method may be the same as the conventional method.

Numeral 203 receives quantitative primary information of handwriting pattern data from the database 201, processes it to obtain secondary information, and defines information for quality evaluation of the database 201 based on the secondary information. It is a quality information definition part. The quality information defined by the quality information definition unit 203 is stored in the database 201 or in another file 204 as shown by a broken line, and can be referred to at any time on the online handwritten character recognition development system 200 side.

An embodiment of quality information definition by the quality information definition unit 203 will be described below. Here, a concept necessary for smoothly performing the following description will be described.

Point, p When the electronic pen touches the digitizer and the touched location is detected at a certain time, the coordinate value (x,
y) is called a point. This point is described as p = (x, y).

Stroke, s When a writer writes a character on a digitizer using an electronic pen, contacting the electronic pen with the digitizer is referred to as pen-down, and speaking of the electronic pen from the digitizer is referred to as pen-up. And The stroke is a time series of points representing the movement trajectory of the pen tip from pen down to pen up. N points, p_1, p_2, p_ in stroke s
When there are 3,..., P_n (n ≧ 1), the stroke s is written as s = (p_1, p_2, p_3,..., P_n). Points are the special case of stroke (n = 1).

Handwritten character handwriting pattern, C The handwritten character handwriting pattern is a time series of strokes. When a certain character C is written with a w stroke (w ≧ 1) and the stroke order is s_1,..., S_w, it is written as C = (s_1,..., S_w). As long as confusion does not occur, the handwritten character handwriting pattern may be called a “character”. A dot, or strone, is a special case of letters.

[Embodiment 1] In this embodiment, the quality information definition unit 203 defines quality information relating to the appearance frequency of character types.

The number of target character types in the handwritten character pattern database is Ncat, and the set of target character types is SETcat = ｛C
AT_1, CAT_2, ..., CAT_Ncat}. Then, the number of target character types in the handwritten character pattern database,
A set QcatN = (Ncat, NptnMean, NptnSD) of the average of the number of handwriting patterns of the character type and the standard deviation of the number of handwriting patterns of the character type is defined as the quality related to the appearance frequency of the target character type.
tN is output as quality information.

Here, the meanings of Ncat and NptnMean are obvious. NptnSD means that the larger the value of NptnSD, the greater the bias in the appearance frequency of the target character type. NptnSD = 0
Means that the appearance frequency of the target character type is uniform.

[Embodiment 2] In this embodiment, the quality information definition unit 203 determines the quality of the handwriting pattern of the character type as the quality evaluation information of the database 201 from the viewpoint of whether or not the written characters are written carefully. Define information.

The number of handwriting patterns of the i-th character type CAT_i (1 ≦ i ≦ Ncat) in the handwritten character pattern database is Nptn_i, and the sum of the number of handwriting patterns of all character types is Np.
tn = Nptn_1 + Nptn_2 +... + Nptn_Ncat. Also, a set of handwriting patterns of the character type CAT_i is set as SETptn_i
= {Cptn_i_1, Cptn_i_2, ..., Cptn_i_Nptn_i}. Handwriting pattern Cp of the j-th character type in SETptn_i
Nstk_i_j is the stroke number of tn_i_j (1 ≦ j ≦ Nptn_i)
The average of the number of strokes of the handwriting pattern of this character type is set to NstkMean_i. The standard deviation of the number of strokes of the handwriting pattern of this character type is NstkSD_i.

A set of the number of handwriting patterns of this character type, the average of the number of strokes of the handwriting pattern, and the standard deviation of the number of strokes of the handwriting pattern, Qptn_i = (Nptn_i, NstkMean_i, NstkSD_i), is used as the quality of the handwriting pattern of the character type CAT_i. Define.
Here, the larger the value of NstkSD_i, the greater the change in how to write the handwriting pattern of the i-th character type. NstkSD_i = 0 means that the handwriting pattern of this character type was written with the same number of strokes.

The sum of the standard deviations of the number of strokes weighted by the number of handwriting patterns for each character type NstkSD = (Nptn_1 × NstkSD_1 + Nptn_2 + Nptn_
2 × NstkSD_2 + ... + Nptn_Ncat × NstkSD_Ncat)
/ Nptn, and the combination of the number of target character types and the standard deviation of the number of strokes of the weighted handwriting pattern Qptn = (Ncat, NstkSD) is defined as the quality of the handwriting pattern of all character types. This Qpt
n and Qptn_i are output as quality information.

[Embodiment 3] In the present embodiment, the quality information definition unit 203 defines the quality of the database 201 based on the number of strokes of the quality of a handwriting pattern written by a certain handwriting person.

The number of character types written by a writer who provided handwriting patterns to the handwritten character pattern database
catwrtr (1 ≦ Ncatwrtr ≦ Ncat). The i-th character type CATwrtr_i (1 ≦ i ≦ Nca written by this writer)
The number of handwriting patterns in t) is Nptnwrtr_i, and the sum of the number of handwriting patterns of all character types written by this writer in the database is Nptnwrtr = Mptnwrtr_1 + Nptnwrtr_2 + ... + Nptnwrt
r_Ncat. Also, a set of handwriting patterns of CATwrtr_i filled out by the scriber is expressed as SETptnwrtr_i = iCptnwrtr_i_1, Cptnwrtr_i_
2, ..., Cptnwrtr_i_Nptnwrtr_i｝.

The number of strokes of the j-th handwriting pattern Cptnwrtr_i_j (1 ≦ j ≦ Nptnwtr_i) in SETptnwrtr_i is Nstkwrtr_i_j, the average number of strokes of the handwriting pattern of this character type is NstkMeanwrtr_i, and the stroke of the handwriting pattern of this character type is Standard deviation of numbers is NstkSDw
rtr_i. Then, the quality of the handwriting pattern of the j-th character type entered by the scribe is determined by the number of handwriting patterns of the character type,
A set of the average and standard deviation of the number of strokes of the handwriting pattern Qptnwrtr_i = (Nptnwrtr_i, Nstk Meanwrtr_i, Nst
kSDwrtr_i). Here, as NstkSDwrtr_i is larger, it means that the writer's writing style of the i-th character type is more unstable. NstkSDwrtr_i = 0 means that the writer's i-th character type is constant.

The sum NstkSDwrtr = (Nptnwrtr_1 × NstkSDwrtr_1 + Np) obtained by weighting the standard deviation of the number of strokes of each writing pattern written by the writer with the number of patterns.
tnwrtr_2 × NstkSDwrtr_2 + ... + Nptnwrtr_Ncatwrt
r × NstkSDwrtr_Ncatwrtr) / Nptnwrtr.

Then, the set of the number of target character types written by the writer and the weighted sum of the standard deviation of the number of strokes of each handwriting pattern, Qptnwrtr = (Ncatwrtr, NstkSDwrtr), is written by the writer. Defined as the quality of the handwriting pattern of the species. These Qptnwrtr and Qptnwrtr_i are output as quality information.

[Embodiment 4] In this embodiment, the quality information definition unit 203 defines the quality of the database 201 based on the quality of the handwriting pattern of the writer.

Let Nwrtr be the number of writers who have provided handwriting patterns to the handwritten character pattern database. Further, as described in the third embodiment, the number of target character types written by the K-th writer (1 ≦ k ≦ Nwrtr) is determined by Ncatwrt.
r_k, and the weighted sum of the standard deviation of the number of strokes of each handwriting pattern written by the writer is NstkSDwrtr_k. Then, the quality of the handwriting pattern of the k-th writer is set as Qptnwrtr_k = (Ncatwrtr_k, NstkSDwrtr_k).

The average of the number of target character types written by the scribes is Ncat Meanwrtr, and the standard deviation thereof is NcatSDwrtr. The average of the sum of the standard deviations of the stroke numbers of the handwriting patterns written by the writers is NstkSDeanwrtr, and the standard deviation of the sum is NstkSDSDwrtr. Then, the number of character types in the database, the average number of target character types entered by the scribes, the standard deviation of the number of target character types entered by the scribes,
A set of the standard deviation of the sum of the standard deviations of the stroke numbers of the handwriting patterns written by the scribes and the standard deviation of the sum of the standard deviations of the stroke numbers of the handwriting patterns written by the writer Qwrtr = (Ncat, NcatMeanwrtr, NstkSDMewrtrtr,
NstkSDSDwrtr) is defined as the quality of the database by the quality of the scribe's handwriting pattern. These Qwrtr and Qptnwrtr_k are output as quality information.

Here, the larger the value of NcatSDwrtr, the greater the difference in the number of target character types entered by the writer. NcatSDwrtr = 0 means that the scribers have entered the same number of matching character types. NstkSD
The larger the SDwrtr, the greater the difference in the quality of the handwriting pattern written by the scribe.

[Embodiment 5] In this embodiment, the quality information definition unit 203 defines the position of the quality of the handwriting pattern of the writer in the database 201.

Let Nwrtr be the number of writers who have provided handwriting patterns to the database. As described in the third embodiment, the number of target character types written by the k-th writer (1 ≦ k ≦ Nwrtr) is Ncatwrtr_k, and the standard number of strokes of each handwriting pattern written by this writer is weighted. Let the sum of the deviations be NstkSDwrtr_k. Then, the quality Qptnwrtr_k = (Ncatwrtr_k, NstkSDwrtr_k) of the handwriting pattern of the k-th writer is defined.

As in the case of the fourth embodiment, the quality of the database using the quality of the handwriting pattern of the writer Qwrtr = (Ncat, NcatMeanwrtr, NcatSDwrtr, Nstk)
SDMeanwrtr, NstkSDSDwrtr).

The rank of the quality of the handwriting pattern written by the k-th scribe is given by Qrankwrtr_k = (Ncatdif_k, NstkSDdif_k) where Ncatdif_k = Ncatwrtr_k-NstkSDMewrwrt
r NstkSDdif_k = NstkSDwrtr_k-NstkSDMewrt
Defined as r. This Nrankwrtr_k, Qptnwrtr_k, Qwr
tr is output as quality information.

Here, Ncatdif_k> 0.0 means that the k-th writer has entered more than the average target character type. NstkSDdif_k> 0.0 means that the quality (attentiveness evaluated by the number of strokes) of the handwriting pattern written by the k-th writer is lower than the average.

[Embodiment 6] In this embodiment, the quality information definition unit 203 defines the position of the quality of the handwriting pattern of the writer group in the database 201.

Nwrtr is the number of writers who have provided the handwriting pattern to the database, and a group of writers WG = {WRTR_1, WRTR_2,..., WRTR_m}
(1 ≦ m ≦ Nwrtr). As in the third embodiment, the quality of the handwriting pattern of the k-th writer (1 ≦ k ≦ m) Qptnwrtr_k = (Ncatwrtr_k, NstkSDwrtr_k) is defined.

As in the fourth embodiment, the quality of the database using the quality of the handwriting pattern of the writer Qwrtr = (Ncat, NcatMeanwrtr, NcatSDwrtr, Nstk)
SDMeanwrtr, NstkSDSDwrtr).

As in the case of the fifth embodiment, K
The position of the quality of the handwriting pattern written by the scribe is defined as Qwrtr_k = (Ncatdif_k, NstkSDdif_k). QrankgrpWG = (NcatdifWG, NstkSDdifWG) where NcatdigWG = (Ncatdif_1 + ... + Ncatdif_
m) / m-Ncat Meanwrtr NstkSDdifWG = (NstkSDdif_1 + ... + NstkSD
dif_m) / m-NstkSDMeanwrtr. This QrnakgrpWG, Qwrtr, Qrnakwrt
r_k is output as quality information.

Here, NcatdifWG> 0.0 means that the writer group WG has entered more target character types than the average. NstkSDdifWG> 0.0 means that the quality of the handwriting pattern written by the writer group WG (the politeness evaluated by the number of strokes is worse than the average).

[0041]

As is apparent from the above description, the present invention defines an online handwritten character pattern by defining information for evaluating the quality of an online handwritten character pattern database from a character type aspect and a writer's aspect. It provides a clear objective criterion for evaluating the results of research and development of handwritten character recognition technology using a database, and greatly contributes to research and development of handwritten character recognition technology.

[Brief description of the drawings]

FIG. 1 is a diagram showing a general flow of research and development of online handwritten character recognition.

FIG. 2 is a block diagram for explaining an embodiment of the present invention.

[Explanation of symbols]

 200 Online Handwritten Character Recognition Research and Development System 201 Online Handwritten Character Pattern Database 203 Quality Information Definition Unit 204 File

Claims (9)

[Claims] 1. An online handwritten character pattern database evaluation method, wherein information for quality evaluation of an online handwritten character pattern database is defined from quantitative information of handwriting pattern data of the online handwritten character pattern database. . 2. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes information on the frequency of appearance of handwriting patterns of each character type. Evaluation method of online handwritten character pattern database. 3. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes information representing the quality of the handwriting pattern of each character type in terms of the number of strokes. A method for evaluating an online handwritten character pattern database characterized by: 4. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes information representing the quality of handwriting patterns of all character types as viewed from the number of strokes. An online character pattern database evaluation method characterized by the following. 5. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes information representing the quality of the handwriting pattern of each character type of each writer as viewed from the number of strokes. A method for evaluating an online handwritten character pattern database, wherein the method includes: 6. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes information representing the quality of handwriting patterns of all character types of each writer as viewed from the number of strokes. A method for evaluating an online handwritten character pattern database, wherein the method includes: 7. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes information indicating a degree of a difference in quality of a handwriting pattern of each writer as viewed from the number of strokes. A method for evaluating an online handwritten character pattern database, comprising: 8. A method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes handwriting of all scribes in terms of the number of strokes of the handwriting pattern of each writer. An evaluation method for an online handwritten character pattern database, which includes information indicating a relative evaluation of quality with respect to the number of strokes of a pattern. 9. The method for evaluating an online handwritten character pattern database according to claim 1, wherein the information for quality evaluation defined includes the quality of all scribes in terms of the number of strokes of the handwriting pattern of each writer group. An evaluation method for an online handwritten character pattern database, which includes information indicating a relative evaluation of the quality of a handwriting pattern in terms of the number of strokes.