JPS6049482A - Approximate describer for on-line handwritten linear pattern - Google Patents

Abstract

PURPOSE:To decrease the processing time and to attain coding with a high data compression factor by obtaining automatically each approximate allowable error of a straight line, a circle and an arc respectively in consideration of a geometric form of an input linear pattern. CONSTITUTION:An input part 3 which feeds the time-series coordinate value data on a linear pattern is provided together with a flexion point extracting part 4 which extracts a flexion point out of the coordinate value data, an automatic approximate allowable error setting part 5 which calculates automatically an approximate allowable error in consideration of a geometric form of the linear pattern, and a linear element analysis/detection part 6 which analyzes roughly the linear pattern. Thus the fine and rough approximate coding operations are possible for the fine and rough linear elements since an approximate allowable error is automatically set to each element of a straight line, a circle and an arc respectively in consideration of the geometric form of the linear pattern. Furthermore, the part 6 detects hierarchically and preferentially the elements having high appearing frequencies. This device attains an approximate coding action in real time and with a high compression factor.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention approximates and encodes time-series coordinate value data of a line figure input by hand into a line figure input device in real time using a computer. The present invention relates to an online handwritten line figure approximation description device that displays the approximate description result of a handwritten line figure on a display.

Configuration of conventional example and its problems Conventional method - As a method for approximately encoding time-series coordinate value data of handwritten line figures, coordinate value data is sampled at equal time intervals, and each sampled pixel is interpolated with a straight line. This method is well known. However, since this method approximates a circle or an arc portion by a straight line, there are problems in that smooth approximation results cannot be obtained and a high data compression rate cannot be obtained.

Conventionally, as a method of analyzing dynamic coordinate value data and approximately describing the handwritten line figure using straight lines and circular arcs (including circles), the coordinate value data 1 of the handwritten line figure is divided into small areas as shown in Figure 1. After performing microscopic analysis such as determining whether each area is a straight line or a circular arc, and if it is an arc, determining the radius of the curved wheel, small areas with similar attributes are determined. Integrate.

Finally, a method was used to obtain approximation result 2 as shown in Figure 2. Approximate description result 2 is an approximate description of the coordinate value data 1 in FIG. 1, accurately and accurately. However, if we take into account that the coordinate value data 1 is handwritten input data and broadly analyze the entire coordinate value data 1, we will find that it should be determined that it represents one circle. .

Conventional techniques microscopically analyze handwritten line figures in this way, making it impossible to grasp the entire picture of the line figure, requiring a large amount of processing time, and requiring high data compression rate encoding. The problem is that it cannot be done.

OBJECTS OF THE INVENTION The present invention eliminates such conventional problems and efficiently extracts bending points by focusing on the drawing speed of line figures input by hand into a line figure input device. We globally analyze the line elements that are divided before and after the bending point, and use straight lines (including points) and arcs (including circles) to approximate the handwritten line figure with high data compression rate in real time. Perform encoding.

The present invention aims to provide an online handwritten line figure approximation description device that displays the approximation results on a display.

Composition of the Invention The online handwritten line figure approximation description device of the present invention includes: a handwritten line figure input unit that obtains time-series coordinate value data of a line figure that is input by hand; and a bend point that extracts a bend point from the coordinate value data. An extraction section, an approximation tolerance automatic setting section that automatically calculates an approximation tolerance considering the geometrical shape of the input line shape, and a straight line or A line element analysis and detection unit that hierarchically prioritizes and detects geometric elements that appear frequently, such as circles, and divides them into minimum units of approximate description, and stores encoded data of detected line elements in a list. Equipped with an encoded data storage unit and an approximation description output unit that displays the original image and approximation results,
The above-mentioned line figure is divided before and after the bending point, and each line element is grasped holistically, and each line element that falls within the approximation tolerance of each point straight line 9 circles 2 circular arcs automatically set from the geometric shape is divided into straight lines and circles. It preferentially detects items with a high frequency of appearance, performs approximation and encoding processing, and displays the approximation results on a display.

DESCRIPTION OF EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. Third
FIG. 3 shows a block diagram of the online handwritten line figure approximation description device of the present invention. In Fig. 3, numeral 3 denotes a handwritten line figure input unit. In this case, a tablet was used as an input device. 4 is a bending point extraction unit 91 handwritten line figure input unit 3
The bending points are extracted from the time-series coordinate value data of the handwritten line figure input in . Reference numeral 6 denotes an approximation allowable error automatic setting unit, which automatically sets approximation allowable errors for each line element of nine straight lines and one circular arc. Reference numeral 6 denotes a line element analysis and detection unit, which hierarchically detects each line element of one straight line, one circle, two circular arcs, and a point that falls within the approximation tolerance determined by the approximation tolerance automatic setting unit 6. 7
is an encoded data storage unit, and 09 is an approximate description output unit that stores encoded data of each line element, which is the minimum unit of approximate description detected by the line element analysis detection unit 6, in the list 8. Approximate description results are output on the display 10 from the encoded data of each line element obtained in the data storage section 7.

In the apparatus configured as described above, the operation of each processing section will be explained below.

(2) Handwritten line figure input unit 3 In this embodiment, a tablet whose sampling frequency of coordinate value data is constant is used as an input device for the handwritten line figure input unit 3 shown in FIG. Tablet - The coordinate value data of a line figure that is input by hand is obtained by the coordinate value data sampling function of the tablet. The handwritten line figure input section 3 receives this and outputs the coordinate value data for each line figure drawn with one stroke (hereinafter referred to as a stroke) to the following processing section.

■ Inflection Point Extraction Unit 4 In order to extract inflection points from time-series coordinate value data of a line figure that is handwritten into a tablet that samples coordinate value data at a constant frequency, the inflection point extraction unit 4 performs preprocessing. A bending point candidate point extraction process that extracts bending point candidate points by paying attention to the fact that the drawing speed of a line figure input by hand is drastically reduced near the bending point, and determining whether or not the bending point candidate point is a bending point. Performs bending point verification processing.

The bending point candidate point extraction process and bending point verification process will be described below.

(1) Bend point candidate point extraction process If all the minimum values of the drawing speed are extracted as bend point candidate points as shown by the arrows in Figure 4, the effects of noise caused by one-move play and quantization will be eliminated. It is not efficient as it includes everything. Therefore, as shown in FIG. 6, a pixel 13.14 where the drawing speed reaches the optimum value is set within the small area 11.12, which includes pixels where the drawing speed sharply decreases.
0 is extracted as a bending point candidate point. The small area is set by the following procedure.

(1) Time-series pixel array (P
+ , P2 , Pt-4, P
t-3, Ps-2・Pl-1, Pt, Pi+1,
--- , the drawing speed of each pixel in Pin l is
Since the tablet coordinate value data sampling frequency is constant, it is calculated from the distance between each adjacent pixel. Therefore, the drawing speed Si of the pixel Pl is expressed by Si=Dt, 1-1 (Di, i-+ is the inter-pixel distance between the pixels Pi and Pl-1).

(ii) Drawing speed of pixel Pi-4 calculated in (1) 5i-4 constant c (o<c<1)e multiplied by nfc value GXSl
Comparing −4 and 81, the following two cases are possible0 CjXSi−4≧St ・・・・・・・・・・・・・・・
・(a) GXSl-4< St ・・・・・・・・・・
・・・・・・(b)(iii) All l(s<t<n
: Repeat (1) and (li) for (n is the total number of pixels) - Check which state is in (a) or (b).

For example, if the state changes from (in) to (a) at i - jl, pixel j1 is the first pixel in the small area.

Conversely, if the state changes from (a) to (b) at t=j2, it can be seen that pixel (j2-1) is the last pixel of the small area, and the change from pixel j1 to pixel (k2-1) ) is set as one small area.

In FIG. 6, C=1. The case where C=0.6 is shown. As is clear from FIG. 6a, the small area 18, 19, 2 including the pixel 16.16g17 where the drawing speed takes the minimum value as it approaches C1.
0 is extracted.

Since the drawing speed of pixel 16 is fast, it is appropriate to determine that it is not an inflection point. FIG. 6 b shows the case of Kc=o, 6, where the small area is 18.19, and the small area containing 1 pixel 16 is not extracted, which is efficient.

・Extract the pixel with the minimum drawing speed in each small area extracted as described above as a bending point candidate point0 (2) Bent point verification process The bending point candidate extracted in the bending point candidate point extraction process. Verify whether a point is an inflection point.

First, in order to measure the opening angle of a pixel row centered on the bending point candidate point, the arm length for measuring the opening angle is determined by the following procedure.

(1) Give an initial value of 10 to the arm length l.

(11) As shown in FIG. 7B, the pixels 21.2 closest to the point l apart from the bending point candidate point 20 in both directions
Extract 2.

(iiD. Both pixels 21°2 extracted as shown in Figure 7)
A perpendicular line is drawn from the bending point candidate point 20 to the straight line connecting the points 2 to 2 to calculate the length d of one foot.

Ov) 1 if the leg length d of the perpendicular line is υ smaller than the threshold
．． 4-11+Δl. Repeat (++) to Gy) until 0 (V) d>D.

Through the above procedure, the arm length can be calculated taking into account the geometrical shape in the vicinity of one bending point candidate point.The opening angle of the pixel row centered at the bending point candidate point can be calculated using the zero-order rounded arm length. calculate. As shown in FIG. 8, the intersection angle of a straight line connecting the bending point candidate point 20 and both pixels 21 and 22 whose distance exceeds the arm length E for the first time with the bending point candidate point 20 as the center is defined as the opening angle.

Based on the calculated opening angle, a first verification is performed to determine whether the bending point candidate point is a bending point. At this time, the following two parameters set in advance are used to reduce the processing time for extracting bending points.

(1) Recognition Minimum Arc Diameter When attempting to reverse the drawing direction during handwriting input, a phenomenon occurs in which the drawing trace becomes rounded near the bending point due to the effects of stylus pen slipping, etc. In order to ignore this influence, when a bending point candidate point is extracted on a circular arc drawn with a diameter less than this parameter, it is not recognized as one circular arc and is unconditionally set as a bending point.

(11) Recognized bending point maximum opening angle The bending point candidate points having an opening angle greater than this parameter are:
It is considered to be a point on a straight line, not an inflection point.

A second verification is performed on the bending point candidate points for which it cannot be determined whether or not they are bending points through the above processing, based on information on a 12-dimensional plane. That is, as shown in Fig. 9, the straight line 23.24
Drop a perpendicular line from the pixel located between each end,
If all the leg lengths from each pixel are equal to or less than the threshold value, the bending point candidate point 20 is extracted as the bending point.

■ Approximation tolerance automatic setting section 6 When two straight lines with different lengths are drawn by hand as shown in FIG. 10, the error ε2 of the longer line segment is obviously larger. Therefore, it is desirable to determine the allowable error in approximation depending on the line segment length.

Therefore, as shown in Figure 11, approximation tolerances for straight lines, circles, and arcs are automatically set in proportion to the line segment length.

Consideration is given to strictly approximating dense line elements and roughly approximating coarse and negative line elements. However, the upper limit value εh is set in consideration of the width of one tablet, and the lower limit value εl is set in consideration of the error in reading the coordinate values of the tablet.

■Line element analysis and detection unit 6 Analyzes human strokes in a global manner and divides each stroke hierarchically according to the following steps in order to preferentially detect line elements that appear more frequently, such as straight lines and circles. , performs detection processing for each line element.

(1) Detection process of 2 points in 1 straight line (stroke and pre-segment unit) Whether the pre-segment divided before and after the bending point (stroke if there is a bending point) is 1 straight line or a point. to judge. The conditions for detecting a pre-segment as a straight line or a point are listed below.

(1) Straight line conditions: As shown in FIG. 12, perpendicular lines are drawn from all pixels in the pre-segment 26 to the straight line 26 connecting both end points of the pre-segment 26.

If the lengths of all legs are within the straight line approximation tolerance, it is determined that the line is a straight line.

(11) Condition for a circle: In order for the pre-segment 27 as shown in FIG. 13 to be a circle, the following three-row property must be satisfied. In FIG. 13, the distance E between the coordinates of the start and end points 28 and 29 of the pre-segment 27 is less than or equal to the threshold value.

(b) Check the flatness of the pre-segment 27.

That is, the aspect ratio (dy/dX) of the rectangle circumscribing the pre-segment 27 in FIG. 13 is within the permissible aspect ratio.

(0) The intersection of the diagonal lines of the rectangle circumscribing the pre-segment 27 is set as the temporary center point 30 of the -pre-segment 27. The distance between the center point 30 and all pixels is calculated, and all pixels are within the allowable eccentric length range FLmtn''RmaX.

(111) Point condition: If the distances between the starting point of the pre-segment 27 and all pixels are all less than the threshold value, it is determined that the point is a point.

If the pre-segment is neither a straight line nor a single point, it is determined that it is a curved line.

(2) Dividing processing of curved pre-segment The curved pre-segment is divided into a straight line part and a curved part (right turning line and left turning line) according to the following procedure.

(1) Determine the turning direction of each pixel of the curved pre-segment.

If the pixel 31 is on the left of the straight line connecting the pixels 32 and 33 with respect to the drawing direction, as shown in Figure 14a, it is called a right-turning line, and is represented by label 2, and conversely as shown in Figure 14. If it is to the right, it is called a left turn line and is indicated by label 1. Note that if the pixel is on a straight line, the label of the previous pixel is inherited.

(11) As shown in FIG. 15, the curved pre-segment is divided into continuous pixel groups in the same turning direction.

(iii) (The following process is performed on pixel groups that are less than the minimum number of pixels required to form one pixel group among the pixel groups divided by tD.

(a) As shown in FIG. 16a, when the number of constituent pixels of one adjacent fixed direction turning pixel group 2.3 is equal to or greater than the minimum number of constituent pixels, the turning direction of the pixel group 34 of interest is Change the pixel group rotation direction to 36°360.

(b) As shown in FIG. ).

Gv) Focusing on the turning direction of the fixed direction turning pixel group corrected in (iii), the one-curve pre-segment is divided into a straight line part, a right turning line part, and a left turning line part (hereinafter referred to as segments).

(3) Verify whether the straight line segment determined in the straight line segment verification process (2) is within the straight line approximation tolerance. A perpendicular line is drawn from each pixel in the straight line segment to the straight line connecting both end points, and if the length of the legs from all pixels is equal to or less than a threshold value, it is determined that the straight line is a straight line. Furthermore, if even one pixel exceeds the threshold, it is regarded as either the left or right turning line portion, depending on whether the pixel is biased to the left or right side of the straight line connecting both end points.

(4) Curve segment division processing @) The curved segment consisting of the left and right turning lines extracted in (3) is divided into partial turning lines (hereinafter referred to as primitives) that can be approximated by one circular arc using the following procedure.

(+) Forcibly approximate the left and right turning lines with one circular arc. As shown in Figure 17, the distance between the approximate arc and each pixel is
If all are within the approximation tolerance, this segment is treated as a primitive, which is the minimum unit for approximation encoding.

If even one pixel exceeds the approximation tolerance in the process of Ui) (i), the turning line is divided into two partial turning lines, and the process (1) is performed for each partial turning line.

■ Encoded data storage unit 7 Stores the encoded data of primitives divided into the minimum units for approximate encoding in list 8 in ■ Approximate description output unit 9 The handwritten input line figure, that is, the original image, and the The approximation result described from the coded data of each primitive extracted from the handwritten line figure is displayed on the display 10. Effects of the Invention As described above, the present invention provides a geometrical shape of a line figure input by hand. The approximation tolerances for linear elements, circular elements, and arc elements are automatically set in consideration of can be converted. Furthermore, by globally analyzing line figures input by hand, linear elements and circular elements that appear frequently are prioritized and detected hierarchically, allowing for approximate encoding with high compression rates in real time. , it is an extremely efficient device for development into recognition processing.

[Brief explanation of the drawing]

Fig. 1 is a coordinate value data diagram of a handwritten line figure, Fig. 2 is an approximate figure diagram showing the approximate description result of a handwritten line figure, and Fig. 3 is a configuration of an online handwritten line figure approximation description device in an embodiment of the present invention. 4 to 6 are velocity distribution diagrams for explaining the principle of extraction of bending point candidate points.
Figures to Figure 9 are diagrams for explaining the principle of bending point verification. Figure 10 is a comparison diagram showing the error proportional to the line segment length of a handwritten line figure, and Figure 11 is a comparison diagram of the line segment length and approximation tolerance. Correspondence diagrams showing relationships - Figure 12 is a diagram of the principle of straight line verification, Figure 13 is a diagram of the principle of circle verification - Figure 14 is a diagram of the definition of turning direction, Figure 16 is a division diagram of curved segments, Figure 16 is a diagram of the principle of circular verification. The corrected labeling diagram, FIG. 17, is a diagram of the principle of arc verification. 3...Handwritten line figure input section, 4.River...bending point extraction section, 6...Approximation tolerance automatic setting section, 6.
. . . Line element analysis detection section, 7 . . . Encoded data storage section, 9 . . . Approximate description output section. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
Figure @2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 6 α b Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure Lb Figure 15 O

Claims (1)

[Scope of Claims] A handwritten line figure input unit that obtains time-series coordinate value data of a line figure that is input by hand; a bending point extraction unit that extracts bending points from the coordinate value data; and a line figure that is input. Considering the geometric shape of the straight line 9 circles. An automatic approximation waiting error setting unit that sets each approximation tolerance for arcs to be small for dense line elements and large for coarse line elements, and to broadly analyze input line figures. In addition, there is a line element analysis and detection unit that hierarchically prioritizes and detects frequently appearing graphical elements such as straight lines and circles and divides them into minimum units of approximation description, and encoded data of the detected line elements. an encoded data storage unit that stores the coded data in one nost, and an approximation description output unit that displays on a display an approximation result that is described based on the encoded data of the original picture and line elements,
An online handwritten line figure approximation description device that performs line encoding processing on each line by dividing the line figure before and after a bending point, and displays approximate description results on a display.