JP2801194B2 - Linear component extraction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a system for inputting handwritten characters and figures as a sample point series of time-series coordinate data. And a method for extracting a straight line component for extracting a sequence of points constituting. [Prior Art] Conventionally, position coordinates are taken in at a fixed rate from a coordinate input device such as a tablet, and a sample point sequence of handwritten characters or figures is displayed or transmitted to be displayed on a display device of the other party. In a system for character recognition or figure recognition, a series of sample points obtained from a coordinate input device is configured to be retained and used almost as it is in a process of removing consecutive identical coordinate points. . [Problems to be Solved by the Invention] However, as described above, when the sample point sequence of handwriting is stored as it is, an extremely large storage capacity is required and the processing time is increased. For this reason, in a system for recognizing handwriting, for example, a pen is input as a break until the pen separates from the tablet, and feature points are extracted from a series of sample points in this section, and are representatively represented for subsequent processing. Or
However, the entire sample point sequence must be saved before extracting the feature points, and a large storage capacity is still necessary to cope with long handwriting, such as a single stroke. And This is a case of inputting a figure anyway in the case of handwritten character recognition, and is particularly problematic when editing is possible in the form of the input figure. [Means and Actions for Solving the Problems] In order to solve the above problems, the present invention provides a straight-line component extraction method for extracting a straight-line component from input coordinate data. Determine, determine a linearization range for extracting a linear component from the coordinate data, detect whether the coordinate data input following the start point is within the linearization range, and determine the result of the detection. If it is determined that the input coordinate data is within the linearization range, then re-determine a linearization range to determine whether the next input coordinate data is within the linearization range, As a result of the detection, when it is determined that the input coordinate data is out of the linearization range, the coordinate data before the coordinate data and the start point data are extracted as a linear component, and the extracted linear component is stored. A straight line characterized by A method for extracting components is provided. Embodiment FIG. 1 shows an embodiment of the present invention. An input terminal 100 is used to sequentially input coordinate values from a coordinate input device such as a tablet. Reference numerals 101 and 102 denote delay means for holding the input state of the delay means in synchronization with the input timing of the coordinate value input to the input terminal, and for keeping the output in this state until the next timing. Numeral 103 denotes a conditional delay means for re-holding the input state at that time only when the above-mentioned input timing is reached when the value of the E- input is a signal indicating true. Coordinate data above
P _i is represented by two-dimensional coordinates (x _i , y _i ). Assuming that the x-coordinate and the y-coordinate are each represented by a 16-bit binary number, each delay means can be realized by 32 D-type flip-flops. Simply referred to as P _i coordinate data for later easy. Numeral 104 denotes vectorization means for outputting, for example, a displacement vector from the start point to the end point. In this case, the result obtained by subtracting the x coordinate y coordinate of the from input from the x coordinate y coordinate of the to input is x
The component and the y component are collectively output from the delta-output. Numeral 105 denotes a range departure judging means. If the direction from the from-input point to the to-input point falls within the angle range of the "direction" input from the zone-input, the false is indicated from the out-terminal. It outputs a signal that indicates true if not included. Details of the realization of this will be described later. Reference numeral 106 denotes a linearization range calculation unit that determines a line segment connecting the from-input point to the to-input point as a representative line of the linear approximation, and determines whether or not the subsequent input point is located at the same linear component. For this purpose, the direction range is calculated based on the old direction range of the zone-input, and the new direction range is output from the Nzone-output. Details of the realization of this will be described later. Ten
A switch 7 outputs a direction range of the O-input to the z-output when a signal representing true is given to the S-input, and outputs a direction range of the I-input to the Z-output when a signal representing falseness is given to the S-input. And can be realized by using a data selector of a standard ethics IC or the like. 108 is a FIFO. This means that when the value of the E-input indicates true,
-The value of the input is added to the end of the already held value column, and when data is extracted therefrom, the values are sequentially output from the first value of the column. This is for example MM
It can be realized by using C57401 of Company I (Monolithic Memories Inc.). (There is no equivalent to the E-input in this IC, but it can be realized by generating a SHIFTIN signal by a logical product of a clock signal for taking input timing and the E input, etc.) FIGS. 2 and 3 The operation of the present invention will be described with reference to FIG. Dealing with the time of the rising edge of the input clock CLK ₁ and data determination. That is, coordinate data inputted to the input terminal 100 in synchronization with the rising edge of the input clock CLK _1, the change to the next data begins. In this description, as shown in Figure 2, taken up the operation when the coordinate data is input is sampled as _{P 1, P 2, P 3} , P 4, P 5, P 6, already delayed the means 101 are the coordinates P ₂ is latched, the delay means 102 signals representative of the range of forward (representing in figure ○)
There is latched, the delay means 103 shall coordinate P ₁ is latched. Although any method may be used to record the direction range, for example, as shown in FIG. 4, the right direction is set to 0 °, the counterclockwise direction is counted as 360 °, and the first direction θ _{i of the} range is counted. And the angle S _i representing the range. Hereinafter, for simplicity, a pair of the direction θ _i and the angle δ _i is defined as a direction range α _i . In addition, when all directions are defined as 360 °, they are represented by ○. Straightening range calculation unit 106 inputs from the delay unit 103 coordinates P ₁ as the starting point, enter coordinates P ₂ as the end point from the delay unit 101, the vector ▲ ▼ ₂ orientation theta _'2 and length gamma' ₂ Ask for. Assuming that the linearization threshold is length d,
The one-sided displacement angle ε ′ ₂ is calculated by ε ′ ₂ = sin ⁻¹ d / γ ′ ₂ . Then, the direction θ ′ ₂ + ε ′ ₂ is set as a direction range candidate that can be linearized from the direction θ ′ ₂ −ε ′ _{2 in the} counterclockwise direction. Then, a common range with the immediately preceding direction range obtained from the delay unit 102 is output from the Nzone-output as a direction range that can be linearized. In this case, since the ○ represents the entire range from the delay means 102 is inputted, the orientation range alpha ₁ is in the range from the direction θ _'2 -ε' ₂ orientation θ _{'2 +} ε' ₂ counterclockwise times. The range deviation determination means 105 calculates the direction range α ₁ just calculated.
Is input from the zone− input, and the coordinate P ₁ obtained from the delay means 103 is input as a start point, and the coordinate P ₃ obtained from the input terminal 100 is input as an end point. The direction of the vector ▲ ▼ ₃ enters the direction range α ₁ Out signal f
-Output from output. The vectorization means 104 inputs the coordinates P ₂ and P ₁ of the end point and the start point from the delay means 101 and 103, respectively, and outputs the vector ▲ ▼ ₂
Is output from the delta-output. Switch means 107, since the input signal f from the S-, and outputs I- input or the orientation range alpha ₁ output linearization range calculation means 106 Z- from the output. System or more states is stabilized until the rising of the next input clock CLK _1. So the input clock CLK ₁ rising, the input coordinates describing the operation when it is the time point of changing from P ₃ to P _4. Since the output signal f of the range deviation determining means 105, the state of the delay means 103, do not change the rising of the next input clock CLK _1, a FIFO 108, vector ▲ ▼ ₂ is not added. Delay means 101 outputs the coordinates P ₃ is before the input, the delay means 102 outputs the orientation range alpha _1. This condition persists until the next rising edge of the input clock CLK _1. The linearization range calculating means 106 inputs the coordinates P ₁ and P ₃ , obtains the direction θ ′ ₃ and the one-side displacement angle ε ′ ₃ based on the vector ▲ ▼ ₃ , and obtains the direction θ ′ ₃ −ε ′ _3. From the direction θ ' ₃
And range up to + epsilon _'3, and outputs the orientation range alpha ₂ in the common range, i.e. second view of the orientation range alpha _1. Range deviation determining means 105 determines whether the coordinates P ₄ from the direction range alpha ₂ has it. In this case, from-input, to
-Input the coordinates P ₁ and P ₄ from the input, and
▼ Since ₄ orientation is in the range alpha ₂ outer directions, the signal t
out-output from output. In response to this signal t, switch 1
07 outputs the data from the O-input from the Z-output. That ignores the linearized range alpha _2, and outputs the ○ orientation range representing the entire range. This can be done by specifying the direction as 0 ° and the angle as 360 °. Also, the vectorization means 104
₃ will be output. Furthermore, at the next rising edge of the input clock CLK _1, the input coordinate is explained when the change from P ₄ to P _5. Since the output of the range deviation determining means 105 is the signal t, the delay means 102, 1
03 and the state of FIFO 108 change here. That is, the output of the delay means 103 becomes the coordinate P ₃ of the immediately preceding input, the output of the delay means 102 becomes the direction range ○, and the vector ▼▼ ₃ is added to the FIFO 108. And the delay means 101 has the coordinates P
₄ is output, and this state indicates that the input clock CLK ₁
Lasts until the next rise. This means that the preparation for extracting the point sequence of P ₁ P ₂ P ₃ as a straight line component, performing linear approximation with the vector ▲ ▼ ₃ , and extracting the straight line component starting from the coordinate P ₃ is completed. By continuing concerns a similar operation to the coordinates inputted have further subsequent from P _6, part sequence of points is observed with linear is accumulated in FIFO108 in the form of a straight line connected vectorized. For example to confirm the coordinates P ₆ is, when the input clock CLK ₁ rises, the input point sequence _{P 1 P 2 P 3 P 4} P 5 P 6, linear P ₁ P _3, linearly P ₃ P
₅ two lines are summarized output, will have been determined further linearize the coordinates P ₅ as the starting point. Linearization range calculation means 106 and range deviation determination means 105
May be configured with hardware logic,
Each of them can be easily realized in the form of an embedded program using a one-chip microcomputer or the like. 6 and 5 show an embodiment of a flowchart of a program for realizing the linearized range calculating means 106 and the range deviation determining means 105, respectively. An implementation method of the linearization range calculation means 106 will be described. In step S601 from- input, to- input, Zone- each coordinate from the input _{(x f, y f) (} x t, y t) as and orientation range, starting orientation theta, inputs the angular [delta]. Then, in step S602, the length γ ' Is calculated by In the direction θ ′, the displacement of x and the displacement of y are (x _t −x _f ) and (y _t −y _f ), respectively.
Tan ⁻¹ ((x _t −x _f ), (y _t −y _f )). Subsequently, in step S603, the one-sided displacement angle ε ′ is
Calculated as n ^-1 (d / γ '). Here, d is a limit distance of one component from an approximate straight line when a straight line component is extracted, and is used as a threshold. Subsequently, in step S604, when a straight line connecting one point following the input start point to the input end point is set as an approximate straight line, the direction range of the approximate straight line is a condition for the input end point to be a distance less than d from this straight line. Is calculated with θ′−ε ′ as the start direction and 2ε ′ as the interior angle. And
Input direction range (θ, δ) Calculate this common range,
Let this be α. An example of the calculation is as follows. Start angle θ '
_f is normalized to θ'-ε'-θ (from 0 ° to less than 360 °)
It is assumed that the end angle θ ′ _t is obtained by normalizing θ ′ + ε′−θ. The theta _'the theta _f when _f ≦ _{δ θ' f}
Otherwise, θ _{f is set} to 0, and θ _{t is set} to θ ′ when θ ′ _t ≦ δ.
To δ the θ _t when it is not _t likely. Then the common range α
Can be calculated by setting the orientation to θ + θ _f and the inner angle to θ ′ _t −θ ′ _f . Subsequently, in step S605, the common orientation range α is output. Then, the process returns to step S601 to repeat the above processing. A method of implementing the range deviation determination means 105 will be described with reference to the flowchart of FIG. After step S501, step S502 is executed, and these steps are the same as the above-described steps S601 and S602. However, step S5
In 02, the length γ 'need not be calculated. And step
If θ ′ deviates from θ in the direction of θ + δ in S503, the process proceeds to step S504, and if θ ′ falls within the range of this direction, the process proceeds to step S505. Then, the signal t is output in step S504, the signal f is output in step S505, and the process returns to step S501 to repeat the above processing. As described above, the linearization range is calculated, and it is determined whether or not a straight line ending at the point following the range is determined. Data can be output sequentially. 6 and 5, the processing is simply described as a loop. However, in synchronization with the input timing, after waiting for a time for the input data to stabilize, step S6 is performed.
The program may be programmed to operate once from step 01 to step S605 and steps S501 to S504 and S505. If the strictness is allowed to be relaxed, the one-sided displacement angle ε 'should be
Although n ⁻¹ (d / γ ′) is used, it may be simply set to C · d / γ ′ and the maximum value may be suppressed to 90 °. C is a constant corresponding to this characteristic, and uses a radian-degree conversion coefficient such as 360 / 3.14. Further, in this embodiment, the direction range is handled in units of 0 ° to 360 °, but this may be set in a radian unit of 0 to 2π, or in a range of 0 to 1 or divided into n. Approximately in the quantized direction of integers from 1 to n, it may be handled. In this case, the calculation method of the direction range by the range deviation determination means 105 and the linearization range calculation means 106 may be changed according to these units. [Other Embodiments] FIG. 7 is a flowchart showing another embodiment of the present invention. In the above-described embodiment, an example in which the respective units operate in parallel has been described. In this embodiment, a method of realizing a sequential procedure using a program of a microprocessor or a CPU of a computer system that handles a point locus will be described. The program in FIG. 7 is executed in synchronization with the input point sequence. That is, every time one point coordinates are determined from the coordinate input device or the like, steps S711 to S719 are executed. Similar to the foregoing description embodiments in the description of the program, a description will be given assuming that the coordinates P ₁ as in the Figure _{2, P 2, P 3, P} 4, P 5, P 6 are input. In the figure, reference numeral 701 denotes a start point, 702 denotes an end point, and 703 denotes a storage area for statically storing a range. Specifically, a memory area, a register, and the like are used. The start point memory 701 has the coordinates P ₁ , the end point memory 702 has the coordinates P ₂ , the range memory 70
The 3 shall orientation range alpha ₁ is stored in the coordinate P ₃ now assume that input confirmation run from step S711 starts. Storing the coordinates P ₃ as a sample point in step S711. Then, in step S712, the starting point memory 701
It determines whether the direction from the start point P ₁ in the sample point P ₂ is in the orientation range alpha ₁ in range storage 703. In the FIG. 2 because the vector than the start point P ₁ to a sample point P ₂ is within alpha ₁ proceeds to step S716, and stores the return value as nil. Here, nil is a value indicating that linearization could not be performed, and for example, a vector value of length O may be substituted.
The process advances to step S717, where, rewrites the value of the end point storage 702 at the sample point P _3. Then, in step S718, the direction range for determining the linearization of the next point is calculated from the value of the start point memory 701 and the vector to the value of the end point memory 702, and the intersection of the result and the value of the range memory 703 is calculated. Then, this common orientation range is set as the value of the new range storage 703. It is rewritten in this case facing range alpha _2. And step S719
Then, the return value set to nil as described above is returned, and this processing ends. When the next coordinate P ₄ is confirmed, this program is called. Storing sample points as P ₄ at step S711. In step S712, since the vector ▲ ▼ ₄ is not in the orientation range alpha ₂ value in the range storage 703, now it proceeds to step S713. Here, the return value is stored as a vector ▲ ▼ ₃ from the point stored in the start point memory 701 to the point stored in the end point memory 702, and the flow advances to step S714. Then, the direction range of the range memory 703 is rewritten so as to represent the entire range. When the orientation range is represented by the initial orientation θ and the interior angle δ of the orientation range, the interior angle δ only needs to be 360 °. Then, in step S715, the coordinates P ₃ of the value of the end point memory 702 are stored in the start point memory 7.
Write to 01. The write the coordinates P ₄ of the current sample point as the value of the end point storage 702 in step S717, step S7
In 18, the direction range for linearization determination of the next point is calculated from the vector ▲ ▼ ₄ , and the resulting direction range α
₃ and the value of the range storage 703, that is, the common part with the entire range is set as a new value of the range storage 703. That is, the orientation range α
_{Make 3} Then, the process proceeds to a step S719, where the vector ▲ ▼ ₃ is returned as a return value. Then at the time of execution at the time of input of coordinates P ₅ is, vector ▲
▼ Because ₅ is within the range alpha ₃ orientation, the end point storage 702 is updated to the coordinate P _5, range alpha ₄ range storage 703 orientation, nil is returned. And when running on input coordinates P ₆ are vector ▲ ▼ because ₆ is in the range alpha ₄ outer directions, the starting point storage
701 coordinates P _5, the end point storage 702 is updated to the coordinate P _6, after the range storage 703 also been updated, the returned vector P _{3 5.} As described above, when the coordinate strings P ₁ , P ₂ , P ₃ , P ₄ , P ₅ , and P ₆ are sequentially input, the vectors ▲ ▼ ₃ , ▲ ▼ ₅ are returned in this order, respectively, and the linear component in a state where the extraction has begun the coordinates P ₅ as a starting point. As described above, even if one point subsequent to the subsequence of the sample point series is added, the linearization range calculation means for calculating the position range satisfying the condition of the linear component in advance is provided. At the time when is input, it is possible to extract information representing an approximate straight line. This has the following effects. To solve the drawback that a linear approximation result cannot be obtained until all sample point sequences have been input. ◎ You can receive straight line information immediately when you can no longer approximate the straight line. That is, the response becomes faster. ◎ Since it becomes possible to handle the locus data of the linear approximation and the amount of information is reduced, the memory can be saved and the processing time can be shortened. ◎ By converting the stepwise trajectory pattern due to quantum error into a linear vector when treating the coordinates of the point sequence with digital values,
Unnecessary folding information can be removed. ◎ Since simple trajectory data (linear approximation) can be obtained,
It is easy to extract this feature point (folding point etc.), and the recognition process becomes easy. Since there is no need to store all input sample points, temporary storage is almost unnecessary. ◎ Since the processing can be performed in a pipeline, the present method can be used on the input device side, and the CPU and the processing of the system main unit can be distributed with almost no reduction in response speed. (The overall processing capacity is improved.) Ε In the case of ε ′ = sin ⁻¹ d / γ ′, an ideal approximation that every point of the linear component is within a distance d from the approximate line can be obtained. When the maximum value is kept at 90 ° by ε ′ = Cd / γ ′, the calculation is simple, and it can be realized by an inexpensive circuit and an inexpensive microprocessor. [Effects of the Invention] As described above, according to the present invention, when a linear component is extracted from an input coordinate point, a linearization range for extracting the linear component is sequentially determined, so that higher accuracy is achieved. A simple linear component can be extracted.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a straight-line component extraction method embodying the present invention. FIG. 2 is an explanatory diagram of an input coordinate point sequence. FIG. 3 is a timing chart of the present invention. FIG. 4 is an explanatory diagram of an expression of a direction range. FIG. 5 is a diagram showing a program embodiment of a range deviation determining means. FIG. 6 is a diagram showing a program embodiment of a linearization range calculating means. FIG. 7 is a diagram showing another embodiment of the present invention, which is a program embodiment in the case where all the components are realized by a program. 101, 102 and 103 delay means, 104 vectorization means, range deviation determining means 105, the linearization range calculation means 106, 107 _{switch, P 1, P 2, P} 3, P 4, P 5, P 6 is input coordinates , Α ₁ , α ₂ , α ₃ ,
alpha ₄ is linearized range, 701 starting storage, 702 endpoint storage, 70
3 is range memory.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int. Cl. ⁶ , DB name) G06T 5/00

Claims (1)

(57) [Claims] A linear component extraction method for extracting a linear component from input coordinate data, comprising: determining a starting point from the coordinate data; determining a linearization range for extracting a linear component from the coordinate data; Detecting whether or not the input coordinate data is within the linearization range, and determining that the input coordinate data is within the linearization range as a result of the detection, The linearization range for determining whether the data is within the linearization range is re-determined. If the result of the detection is that the input coordinate data is determined to be outside the linearization range, the coordinate data And extracting the coordinate data and the starting point data before the step (c) as a straight line component, and storing the extracted straight line component.