JPH0620100A - Discrimination system of kanji to be input in arbitray stroke order - Google Patents

Abstract

PURPOSE: To recognize a character by reading a character inputted in various kinds of stroke orders, and reconstituting a peculiar stroke order. CONSTITUTION: This system is provided with a stroke base bank device for comparing the characteristic values of basic strokes in a stroke base 23, and obtaining the result of the identification of the strokes, data base device of a typeface structure resolving process for searching the number of steps, stroke central point, monitor system, number of stroke extraction, and 'stroke root' radical, and using it as the index of a structure resolving process, stroke order re-assembling device 25 for using the spatial relation of the central point of the strokes, and re-assembles the artificial stroke order of a reference character, device for calculating a directional relation between the adjacent strokes of a Chinese character inputted from a handwriting panel, and calculating a position relation between the strokes, and radical base and character base device including the number of the strokes of a reference typeface, artificial stroke order, stroke configuration, and inter-stroke relative position relation.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a kanji identification system for inputting in any stroke order.

[0002]

2. Description of the Related Art In the field of computer research, manufacturers are using on-line Chinese character recognition systems.
system, OLCCR), but this is a computer device in which the user directly inputs characters by handwriting, and it is necessary to memorize various complicated numbers and rules for assembling characters like key input. Absent.

At the starting line of the kanji identification system,
Our paper “On-line handwritten Chinese char
acter recognition by string matching ", Proc.1988 In
As shown in t.Conf. CPCOL pp.76-80), the computer cannot accurately identify the character unless the user performs handwriting according to the specified stroke order and stroke. However, in practice, most people use habitual writing styles that are slightly different from the standard stroke order and strokes, which makes the user considerably inconvenient.

In order to overcome the above limitation on the handwritten character recognition process, many researchers have devised various methods. Some tried to relax the stroke restrictions, and one researcher worked in stroke order. One intuitive solution is to regard different strokes and strokes in one kanji as different characters, but this method makes the database enormous.

In terms of stroke order, a group such as Hidai considered reassembling the character composition from the characteristics of the angle (tilt) and position of the stroke. (“Stroke re-ordering algorithm f
or on-line hand-written character recognition ", Pro
c. 8th int. Conf. PatternRecognition, pp. 934-936.
(Refer to 1986) Although this reconstruction method requires handwriting with accurate inclination, in reality each person's handwriting is not the same, and there are people with very large inclination angles, and the data of fairly detailed character composition is stored in the database. Need to be established.

Wakahara and Umeda tried to determine an input by making an inter-stroke distance matrix at intervals of each stroke and referring to the correspondence between the strokes and the strokes. (“Stroke nunmber and Stroke-order free on-li
ne character recognition by selective stroke linka
ge method ", 1983 Proc. 4th ICTP, pp. 157-162) However, this method also requires an enormous amount of time to calculate the distance between strokes and each distance element in the matrix. .

The group of Ye et al. Thought that the stroke order should be reconstructed by classifying strokes. ("Techniques
for on-line Chinese character recognition with re
duced writing constraints ", 1984 Proc. 7th Int.
Conf. Pattern Recognition, pp. 1043-1045) This is to measure and identify 500 Chinese characters in 3 to 10 strokes. Considering that this method consists of 5401 common characters consisting of 1 to 32 strokes, there are still some doubts in terms of practical use.

Y. J., published in 1986 and 1988, respectively.
Ishii Group paper (“Stroke-order free on-line ha
ndwritten Kanji character recognition method by m
eansof stroke representative points ", Trans. IEICE
J (Japan) J69-D, No.6, pp.940-948, 1986;
nji recognition method which detects writing error
s ", J. of Computer processing of Chinese of Chines
e & Oriental Languages, Vol. 3, No. 3 & 4, p
(See p. 351 to 365), the representative point of the stroke is divided into the starting point, the middle point, and the end point of the stroke using the spatial relationship between the stroke and the check direction. To take. It is thought that this concept hypothesizes the representative points of the strokes and the check direction, and that a stable spatial relationship of strokes can be obtained, but the experiment was conducted with Japanese handwritten kanji. Generally speaking, Japanese handwriting kanji is relatively well-prepared, but when you look at the Chinese characters that have 5401 common characters, the character form is more complicated, and the Chinese typeface is large and distorted. If the representative points of the strokes and the check directions are adopted, the composition decomposition for the typeface is not always sufficient. Furthermore, even if the database is small, considering the impact on memory space,
After all the problem remains.

[0009]

As described above, in a system for identifying a typeface of Chinese characters on a line, it is necessary to develop a system capable of identifying the typeface in any stroke order by making it convenient for the user to handwrite. Happened. Therefore, regardless of the stroke order at the time of input, we will assemble with a fixed "artificial stroke order" and compare it with the features of the subsequent strokes to develop a system that accurately calculates and expresses the degree of crossing of the letter shape. It has become.

An object of the present invention is a system for identifying characters by handwriting in any stroke order and identifying a font online. Even if the user inputs characters according to his or her custom, the system can secure a high identification rate.

Another object of the present invention is to provide a system for decomposing a wide variety of different handwritten typefaces in an accurate and stable manner and for recognizing characters online, which is a kind of fixed artificial stroke order. ) To reassemble.

In order to further save the computer memory space, it is necessary to gradually identify the kanji form on the line.
One of the purposes is to display the decomposition order of the character structure in (hieranchical) and to identify the Chinese characters with many strokes without requiring a huge storage space on the system.

[0013]

As a system for identifying Chinese characters on a line, the present invention provides a stable font structure for the purpose of developing processing of Chinese characters with many strokes handwritten in an arbitrary stroke order. It is an invention of the method and system for disassembling, and the optimal stroke representative point and monitor method are set so as to match the special form of Chinese characters. Since the strokes are arranged in the order of strokes, the system can accurately discriminate between various customary handwriting strokes and handwritings and obtain accurate discrimination results. In selecting the representative point of the stroke, the second feature of the present invention is the midpoint of the Y-axis projection and the lowest point, and the monitor direction is horizontal and vertical, two directions outside, and the character material is assembled even during the monitor. This monitoring method is represented by PQWGD by filtering the strokes that are not suitable for extraction in consideration of the factors that stabilize the completeness and decomposition. In addition, the system of the present invention considers the “root” (radical) of the Chinese character as an element of the character composition when decomposing the font and extracting the stroke, so when the system decomposes the structure of the font,
The radical "character root" was taken out as early as possible, and then the decomposition step of each stroke was determined by referring to the radical structure decomposition program, and the stepwise display method was achieved. It is possible to save the trouble of disassembling each stroke for one kanji and avoid the huge waste of computer memory.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram of a kanji character identification system according to the present invention, in which a user writes characters on a handwriting panel (writing tablet) 11 and then the obtained digital data 14 is subjected to character identification processing. The identification data 15 is sent to the block 12, the input handwritten character data 14 is analyzed, the most probable character is calculated, and then the identification result 15 is sent to the image display block 13.

Taking the system unit as an example, the handwriting panel 11 is a digitizer manufactured by Graphtec.
(KD4030), the resolution is 10 lines / mm, the sampling speed is 110 dots / second, and there is a fine grid-like scale on the upper part. When the user inputs characters, write them in the graph scale in order. The character recognition processing block is a central processing unit, which uses a PC / AT 80386 CPU, and the image display block 13 displays it on a fluorescent screen.

The character identification processing block 12 will be described in detail below.

FIG. 2 is a font recognition processing block, which processes according to the flow of the font data 14 of handwritten characters according to the present invention.

After inputting with handwritten data 14, first preprocessing
Processed in step 21 to eliminate the noise of the input strokes and produce the effect of smoothing. The preprocessing method according to the present invention adopts a constant threshold method. An appropriate constant has already been provided, the difference between the point X or Y coordinates adjacent to each other in the handwritten character data 14 is read, and if the difference is larger than that value, it is reserved. Unnecessary dark points that occur due to the slow writing speed are eliminated, and the effect of being relatively smooth is obtained, and the calculation time in the stroke identification process can be shortened and complexity can be reduced.

The data obtained through the pre-processing 21 is subjected to a line block, that is, a segmentation processing 22, and along a sequence of straight lines, a stroke near a non-straight line is promoted to obtain a characteristic point of the stroke.

In the present invention, the values in eight directions are used as the quantization direction of the line connecting the input points.

If the constituent vector falls in the same direction area, the middle points of the three points can be deleted and processed as an octagonal vector area with a front and rear difference of 22.5 degrees. If removed, the feature point with the most representative value is obtained.

Subsequently, the operation of the stroke identification 23 is performed at the characteristic point after the line division processing. That is, the characteristic points of each stroke are connected to a line in order, and then the quantization direction and the length direction are set as the characteristic points of the stroke, and the characteristic value of the stroke and the characteristic of the basic stroke of the stroke base 32 existing in the memory. Compare with the value,
The result of character recognition can be obtained. In the stroke base,
Characteristic data for 21 basic kanji strokes is stored in memory.

After the strokes are identified, the preliminary class 24 process is performed. In the example of the present invention, the major classification of input fonts is based on the number of strokes. That is, the input character is compared with various reference characters having the same stroke number.

After the system has performed all the above steps, the present invention proposes a method for matching the various stroke order and handwriting deformation situation that may occur on the input font, thereby reassembling the stroke order and the positional relationship between strokes. The calculation is performed, the series between the reference character and the input character shape is compared, and the reference character group with the highest possibility is selected as the candidate character group. This process is described in detail below.

In the character recognition system of the present invention, the strokes are recognized and classified, and then the strokes are reassembled 25. When reproducing, first, the disassembly order of the stroke composition follows the disassembly order of the reference characters having the same number of strokes in the database 33, and the structure of the input character is disassembled, and the stroke of the input character is used as the reference character. After reassembling in the order of the artificial brushes, compare the features of the strokes.

In the structure decomposition method designed by the present invention, in order to save the memory space of the memory block, a stepped structure is particularly adopted to utilize that many common radicals are contained in the polymorphic character peculiar to Chinese characters. The structure decomposing process of radicals is a necessary matter, and in the present invention, when decomposing a structure of each character as much as possible, when decomposing a structure of one character, the radical is firstly analyzed if possible. Is a method of decomposing and decomposing the structure of one radical by the index.

Considering that the input Chinese characters are stably decomposed into radicals in the process of structural decomposition, and that the handwriting of a general writer when a complicated Chinese character is input changes, the structural decomposition process is performed. In designing, more diverse monitoring methods and stroke representative points are designed to match the changes in the structure peculiar to Chinese characters. This is different from what was published in the Y. Ishii group paper. It is not possible to stably decompose a font with large changes simply by monitoring from the two directions of the stroke origin, middle point, end point, horizontal, and vertical. Therefore, according to the present invention, even if the user writes in the stroke order according to personal habits or the handwriting is considerably inaccurate, the user can stably disassemble and reassemble the strokes to obtain an accurate identification result.

Next, the representative points of the strokes and the monitor method adopted in this description will be introduced. First, each symbol will be described.

Representative point of stroke: S: starting point of stroke; T: second middle feature point; M: midpoint of stroke; Y: projection midpoint of Y axis in stroke; L: lowest point of stroke; E: End point of stroke; Monitor method of stroke: H: Horizontal scan from left to right R: Horizontal scan from right to left V: Vertical scan from top to bottom B: Vertical scan from bottom to top P: From right It is difficult to discriminate by the horizontal scanning to the left, the lowest intermediate point character. Q: The lowest intermediate point character, which cannot be discerned by the horizontal scanning from left to right. W: The upper and lower vertical scanning, it discriminates among the three lower starting points. Difficult end point minimum character G: The lowest point that is difficult to determine by scanning from left to right is on the lower right side D: Don't care, that is, the representative point cannot be determined directly Strokes with low input order for one or more strokes Positional stroke allowance between strokes described above For the point and monitor method, a new stroke representative point such as TY L and a check method such as RBPQWGD are used, and an "X", that is, a don't care condition, for example, a certain stroke is an individual The "X" is used when the majority of people recognize that the input order of strokes is correct, although it is not possible to select one stable stroke representative point due to a considerable deformation due to a habit. At this time, since the order of inputting the strokes is not changed, it is possible to avoid disassembling one stroke in one character in an unstable state.

In the process of decomposing the character structure, the display items in the database include the number of steps, radical code / 0, stroke representative point, monitor method and count of stroke number, in which radical code / 0 is radical. It is used as an index of the process of structural decomposition in the search, and 0 indicates that decomposition of radicals is not required. The structural decomposition process data of radicals includes the number of steps, stroke representative points, monitoring method and extraction of strokes. Has been.

[0031]

[Table 1]

If the structure is decomposed using "Morrow" as an example, Table-A
As shown in Fig. 3, there are three radicals, namely "woman" and "mouth".
"Month" is displayed using the structure decomposition process database display method. Table-B shows the structural decomposition process of the Chinese character "Sho". As shown in Table-B, in the first step (MH3), the radicals corresponding to the “woman” of “Morrow” are extracted. M is a representative point of a stroke, H is a monitor method, and 3 is the number of strokes taken out. The second step uses the radical code 3031 as an index,
In the decomposition process in which radicals are indexed, the "female" part is completed in two steps according to the data of the structure decomposition process. They are displayed as (EB2) and (ER1) respectively, and subsequently, as shown in Table-B, the character "female" can be identified in two steps.
After reading 31 radicals, the number of steps is 2 + 2 and 4
Becomes At this time, the radicals in the "mouth" are extracted again with (EV3). Similar to the principle described above, in the fifth step, the decomposition order of the 3006 radicals is captured, and the radical for the last “month” captures the decomposition order of the “month” from the radical code of 4004 and completes the reassembly in the stroke order.

The above-mentioned "娟" character uses horizontal lines and vertical lines, and the strokes between the three radicals are completely decomposed.
As shown in FIG. 3A, the stroke between radicals is decomposed in several appropriate situations, and then the radical of each character is indexed by each radical code, which is the process of structural decomposition.

The font shown in FIG. 3B is the same as that shown in FIG.
It is difficult to develop radicals on the horizontal and vertical lines. In this case, strokes in a plurality of radicals must be decomposed one by one, and the structure cannot be decomposed by borrowing a radical code.

FIG. 4 shows a situation in which the representative points of the typefaces decomposed in this description and the monitor method are applied to the typeface decomposition by way of example.

FIG. 4A shows an application example of the second feature point.
Island (7 radicals in (a) of Fig. 4 (A))

[0037]

[Equation 1]

And divide into three strokes "mountain". The disassembly method is (TV7), T is the stroke representative point, V is the monitor method, 7
Is the number of strokes to be extracted, and the boar in (b) of FIG. 4A takes out the first stroke and then takes out non-radicals at (TR8). 4B is an application example of the L point, and the exemption of the (a) is (L.
V.3) decomposes the radical "mouth" into three strokes 3, 4 and 5 and strokes 6 and 7. In this character example (MV3) or
Do not use (EV3). This is because the midpoint of stroke 5 is the same as the midpoint of stroke 6, the end point of stroke 7 is at a high position, and the letter (LV4) is radical day strokes 1, 2, 3, 4
And disassembled into strokes 5 and 6, without using (MV4) and (EV4), 4M and 5M are the same height, and 5E is also quite high.
The sign of M is the midpoint of the fourth stroke, and 5E is the end point of the fifth stroke.

Referring to FIG. 4C, a good example of the Y-axis projection midpoint is that the Y-axis projection midpoint is obtained by the average number of the highest and lowest Y coordinate values of the stroke, that is, (Ymax + Ymin ) / 2, and in terms of a stroke composed of one stroke and one stroke, two feature points are left after the preprocessing and the processing of the line layer. In this way, the y coordinate of the Y point and the y coordinate of the M point become the same. In the case of using a plurality of line blocks, the y-coordinate value of the Y point and the y-coordinate value of the M point always have different values. It can be easily compared with the y coordinate value of the point. In (a), "many" has two identical radicals in parallel and stable decomposition is difficult, but in the present invention, strokes 1 and 2 are used in (SV2) and strokes 1 and 2 are used in step 2 (YV1). After development, in step 3, (YV1) is used to decompose the stroke 3 and the residual stroke of the constituent radical "Yu". (B)
"Aka" of (YV1) develops stroke 1 and other strokes,
Do not use (SV1) or (EV1). This is because 1S has the same height as 2S and 1E can also have the same height as 3E and 4E.

FIG. 4D, which requires a special monitoring method.
As shown in

[0041]

[Equation 2]

In terms of the radicals of "and", the monitoring system of P is used (MPn).

[0043]

[Equation 3]

The first step in which (a) and (b) in FIG. 4D are decomposed is (MP6). Figure 4
(E) In the two letters with the radical "gate" in,
You can use (SW8) when you want to develop the strokes of itself and other radicals completely. Referring to (F) of FIG.
radical

[0045]

[Equation 4]

(SGn) has n strokes (a), (b) and (c) on the left side. n is 3, 6, 8
Becomes

Finally, as shown in FIG. 5, it is very difficult to develop a plurality of radicals. (Shown with thick black lines in the figure.) When most people write certain characters, it is very rare to write from top to bottom, outside to inside, left to right according to standard rules. The stroke order of radicals and other strokes may be mistaken. At this time, the structure can be developed and identified by the "condition-free" decomposition method of the present invention.

In the process of reassembling in the stroke order according to the present invention clearly according to the above contents, it is applied to the processing of a large class of intermediate characters with the "fixed artificial stroke order" which is stably fixed, and is applied to the handwriting method of the handwriting and the stroke order of the user. On the other hand, it shows greater tolerance.

After arranging in the "artificial stroke order", the system immediately calculates the positional relationship between strokes. This calculation is to find the direction relation between two adjacent strokes (tail: T) and the starting point (Head: H), and there are four types, TH, TT, and HT.

Next, the features of the reference typefaces existing in the radical base bank 34 and the character base bank 35 and the features between the input characters are serially compared (27) as the features of the stroke. The features of strokes include the number of strokes, the artificial stroke order, the form of strokes, and the relative positional relationship between strokes. The input font and the reference character of the same stroke are compared one-on-one, and the reference character with the smallest cumulative calculation value is set as the first candidate for the input character.

Finally, when the first candidate, which is compared (27) in series, belongs to a similar character group, the process proceeds to the next detailed identification 28, in which there are ten or more kinds of basic feature classification strokes and the corresponding characters Establish a binary “Decision Tree” of the group, and find the character that best fits the detailed discrimination decision,
The output result 15 of the identification font is obtained.

In order to prove the effect of the present invention, 5401 Kanji characters were adopted and tested with 80090 characters handwritten by Chinese. As a result, the stability of the typeface disassembled by the system reached around 99%, and the characters that were accurate in sampling identification were as high as 96.9%, which is a sufficient kanji (Chinese character) identification system according to the present invention. However, it has been proved that it has high performance and is excellent for input processing of a Chinese computer.

[Brief description of drawings]

FIG. 1 is a block diagram for identifying kanji characters online.

FIG. 2 is a flowchart of a kanji character identification process input in an arbitrary stroke order in the embodiment according to the present invention.

FIG. 3A is an example of a Chinese character that is divided into several radicals by using simple horizontal and vertical lines, and FIG. 3B is an example of a character in which a partial radical is stroke-by-stroke decomposed.

FIGS. 4A to 4F respectively show examples and fonts to be applied when the character structure is decomposed by the representative point of the stroke and the monitor method according to the present invention.

FIG. 5 illustrates, by way of example, a glyph that allows the structure of a glyph to be resolved under any conditions in accordance with the present invention.

[Explanation of symbols]

 11 Handwriting Plate 12 Character Identification Electronic Parts 13 Image Display Parts 14 Handwritten Character Data Input Digitally 15 Results of Identified Characters 21 Preprocessing 22 Line Blocking Processing 23 Confirmation of Strokes 24 Major Classification 25 Reassembly in Stroke Order 26 Correlative position calculation between strokes 27 Serial comparison of stroke features 28 Identification details 32 Stroke base 33 Stroke structure decomposition program database 34 Radical "root" base 35 Character bank 36 Data bank of similar strokes

Claims (3)

[Claims] 1. A kanji character identification system for inputting kanji characters handwritten in an arbitrary stroke order on a handwriting panel (writing tablet) and outputting an accurate identification result to a display device of the system, the kanji characters being input from a handwriting panel. The pre-processing and line-block processing device, which is the part that obtains the feature points of each stroke by simplifying the data, the stroke identification device, and the feature value of the stroke is obtained using the feature points of each stroke, and the feature of the basic stroke on the stroke base. Stroke decomposition with a stroke base bank device to compare the values and obtain the result of stroke identification, a device that stores the characteristic values of various basic strokes, and a structure that displays the process decomposition data of each typeface with a layered structure. The process data is searched for the number of steps, stroke representative points, monitoring method, stroke number extraction, and “root” radicals to search for the structural decomposition process. By using the spatial relationship between the font type structure decomposition process database that includes the radical code as an index and the representative point of the stroke, the structure decomposition process database memory of the typeface is used to decompose the reference character with the same stroke number into the decomposition order. As a reference, a stroke-order reassembly device that reassembles the reference characters using the artificial stroke order, and a device that calculates the positional relationship between strokes that calculates the directional relationship between adjacent strokes of Chinese characters input from the handwriting panel, and It is a memory that stores the stroke characteristics of typefaces and their radicals. It includes radical-based and character-based devices that include the number of strokes, artificial stroke order, stroke form, and relative positional relationship between strokes, and typefaces and reference fonts that are input one-to-one. A kanji character recognition system that consists of a device that compares the characteristics of strokes and determines the discrimination result in series. Tem. 2. The stroke representative point described in the structural decomposition process data includes at least one of the characteristic points of the stroke second, the stroke at the projection midpoint of the y-axis, and the minimum representative point of the stroke. The kanji identification system according to claim 1, which is included. 3. The following items in the monitoring method (check) of the structural decomposition process data; P: horizontal scanning from the right to the left, the lowest stroke at the middle point which is difficult to decompose, Q: from the left The horizontal stroke to the right makes it difficult to determine the lowest intermediate point stroke. W: The lowest stroke point in the three strokes whose starting point is relatively low in the vertical scan (scan) from top to bottom. , G: a character image in which the lowest point that is difficult to distinguish in the horizontal direction from left to right is in the lowermost part on the right side, D: at least one item irrelevant to the direction is included, and the kanji character identification according to claim 1 system.