JP2002024794A - Method for detecting different feature information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve a method for detecting a different feature information from a peripheral information, as well as achieve detection of the boundary line between an object and its background. SOLUTION: The method for detecting a different feature information from a peripheral information is achieved by a network consisting of the steps S11, S12 and S13. In the step S11, there is spatially disposed an element which transits from a stable motion to a chaos motion by an input to the element defined by a function which generates the chaos motion. In the step S12, an element is disposed such that only a portion relating to the space where a different feature is disposed transits from the stable motion to the chaos motion by performing a differential calculation between spatially adjacent elements with regard to the output signal from the step S11. In the step S13, an element is disposed such that only a spatial portion relating to the different feature generates the chaos motion from the stable motion by integrating the output from the step S12.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the application of chaotic phenomena to a feature detection method for extracting time-series signals having different characteristics from a large number of time-series signals.

[0002]

2. Description of the Related Art Heretofore, in order to construct a method for extracting information having different characteristics from a large number of characteristics, a method of imitating a human ability related thereto has been used.

[0003] A plurality of objects are arranged in an image, and a feature extraction ability of a human is measured using a visual search task that requires searching for an object having only one characteristic from among the objects. Models that represent capabilities have been proposed (eg, Scientific America 254, 114-124 (1
986) (Scientific America, 254, 114-124 (1986)).

[0004] Further, it has been pointed out that an electrical signal obtained from the brain of an organism exhibits chaotic behavior, and information is processed when shifting from stable motion to chaotic motion ( For example, "The world where complex systems open" Nikkei Science 42-51 (1997)).

In recent years, a method relating to a chaotic neural network using such chaos motion has been widely used for associative memory, a combination optimization problem, and the like (for example,
JP-A-06-259483, JP-A-09-146908).

As a conventional technique using chaotic motion, a signal detecting method using a chaotic neural network (Japanese Patent Laid-Open No. 11-311647) has been proposed. JP
In 11-311647, the threshold value is not set for the input signal, the input signal is directly input to the neuron element, the signal of the frequency of the periodic signal is amplified by the resonance phenomenon of chaotic motion, and We have proposed a signal detection method that detects weak signals by using the enhancement of resonance phenomena due to the interaction of.

[0007]

However, in the construction of a method for extracting different information from a large number of features, a method of imitating human ability has been attempted, despite the fact that a method for realizing the information by imitating human ability has been attempted. There has not been proposed any device that realizes a human visual search function by applying a chaotic electric signal obtained from the brain.

Further, in the conventional application of chaotic motion, application to associative memory, combination optimization problem, short-term prediction, etc. has been made, but application to human visual search function has not been made. It has not been.

Further, the conventional method relating to the extraction of information having different characteristics from a large number of characteristics has not been applied to the application to an image processing method using the same. In image processing techniques, it is important to extract the boundary between the object and the background (for example, "Image Processing Engineering", Kyoritsu Shuppan, 69-77 (1996)).

Therefore, the present invention realizes a technique for detecting different characteristic information from peripheral information by applying chaotic motion. In addition, by using the technique of detecting different characteristic information from peripheral information implemented in the present invention, detection of a boundary between an object and a background is realized.

[0011]

According to the present invention, FIG.
As shown in (1), a step (S11) of spatially arranging elements that transition from stable motion to chaotic motion by input to an element defined by a function that generates chaotic motion is performed.

Next, with respect to the output signal from S11, the difference between the spatially adjacent elements is calculated so that only the portion related to the space where the different features are arranged transitions from the stable motion to the chaotic motion. The step of arranging the elements (S12) is performed.

Next, the step of arranging the elements such that only the spatial portion relating to the different feature generates the chaotic motion from the stable motion by integrating the outputs from S12 (S1).
Perform 3).

In the present invention, a method for detecting different characteristic information from surrounding information is realized by a network including these steps.

[0015]

[Equation 1]

O | _{t + 1} = f (O | _t ) In the layer (S21) for receiving the input signal shown in FIG. 2, by adding a coupling term (Equation 2) to the chaotic element shown in Eq. The chaotic motion can be generated by the transient vibration of the chaotic element indicating the motion. [ _Equation 2] O _1xy | _{t + 1} = f (O _1xy | _t ) + w ₀₁ × I _xy | _{t + 1} Here, O ₁ is the output of the chaotic element arranged in the layer (S 21) that receives the input signal. coupling coefficient of the signal, w ₀₁ and chaotic element disposed in the signal and S21 to be inputted to S21,
I is an input signal. Also, x and y indicate the spatial positions of the chaotic elements arranged in the space shown in FIG. In the chaotic element shown in Expression 2, a certain spatial position (x, y), time t
The output signal O _1xy | _{t + 1} from S21 at +1 is at time t
It is determined by the _{t + 1} | input signal I _xy at _t and time t + 1 | output signal O _1Xy of S21 to the.

The layer for receiving the output signal from S21 (S2
In each of the chaotic elements shown in 2), the adjacent
Chaotic motion by the difference between the output signals of the chaotic elements
appear. Output signal of chaotic element arranged in S22
Is determined by Equations 3 and 4. What
Note that Equation 3 is a chaotic element arranged in the x direction shown in FIG.
Equation 4 is the difference between the chaotic elements arranged in the y direction.
The difference between the output signals of the slaves is detected. [Equation 3] O_{2 (2x) (2y-1)}|_{t + 1}= F (O_{2 (2x) (2y-1)}|_t) + W₁₂× | (O_{1 (x + 1) y}|_{t + 1}-O_1xy|_{t + 1}|  [Equation 4] O_{2 (2x-1) (2y)}|_{t + 1}= F (O_{2 (2x-1) (2y)}|_t) + W₁₂× | (O_{1x (y + 1)}|_{t + 1}-O_1xy|_{t + 1}| Where O_TwoIs the output signal of the chaotic element arranged in S22.
Number, w₁₂Is between the chaotic elements arranged in S21 and S22.
Coupling coefficient. In Equations 3 and 4, it is arranged in S21
The difference between the output signals of adjacent chaotic elements
The chaotic element arranged in S21 is performing synchronous oscillation
The output signal of S22 has the same motion as before the signal input.
Will continue. Therefore, stable motion before signal input
The chaotic element has stable motion after signal input.
Continue.

On the other hand, when the difference between the output signals of the adjacent chaotic elements arranged in S21 is not 0, the output signal of S22 is affected by S21. Therefore, a chaotic motion is generated by the transient vibration of the chaotic element due to the influence from S21.

Each chaotic element arranged in S23 is S2
Signals having different characteristics are extracted by integrating the output signals from the two. Each chaotic element arranged in S23 is determined by Expression 5. Here, O ₃ is placed on the output signal of the chaotic elements arranged in S23, w ₂₃ the coupling coefficient between the chaos elements arranged to S22 and S23, n is the spatial position of within S23 (x, y) Is a variable indicating the number of couplings from the chaotic element arranged in S22 to the chaotic element. Therefore, in the chaotic element shown in Expression 5, when the spatial position (x, y) of each chaotic element arranged in S22 in the coupling term from S22 becomes a negative value, there is no coupling related to the chaotic element. And is reflected in the variable n indicating the number of connections from S22.

In the chaotic element expressed by Equation 5, when the output signal of S22 makes a stable motion, the output signal of the chaotic element arranged in S23 also makes a stable motion.

On the other hand, when there is a chaotic motion in the output signal of S22, the output signal of S23, which has performed a stable motion before the signal input, generates a chaotic motion.

According to the present invention, only the portion to which a signal having a characteristic different from that of the surroundings is input generates chaotic motion by the above method.

[Embodiment 1] The mapping function shown in Equation 1 is replaced by Equation 6.
The extraction of the different feature information was performed as shown by. [Equation 6] O | _{t + 1} = a × O | _t × (1−O | _t ) where a is a coefficient that determines the characteristics of the mapping, and a>
At 3.570, the map generates chaotic motion. At this time, as shown in Expressions 7 and 8, an input signal whose signal intensity changes with time was used. In this embodiment, the input signal shown in Expression 7 is used as a signal intended for extraction by the present method, that is, different feature information, and the input signal shown in Expression 8 is used as other signals. The input signal at this time is as shown in FIG.

Each of the chaotic elements arranged in S21 shown in FIG. 2 determines the characteristic coefficient a of the chaotic element so as to perform a stable motion before signal input. Therefore, in this embodiment, S
The test was performed with the characteristic coefficient a = 2.5 of the chaotic element arranged at 21. When the characteristic coefficient of the chaotic element is set as described above and the signals shown in Equations 7 and 8 are input, w ₀₁
When> 0.80, the output signal of the chaotic element arranged in S21 diverges, and no chaotic motion occurs. Therefore, in this embodiment, w ₀₁ = 0.5.
At this time, the output signal of the chaotic element arranged in S21 is as shown in FIG.

Similarly, the characteristic coefficient of each chaotic element arranged in S22 is set to a = 2.5 so that the chaotic element moves stably before signal input. At this time, the coupling coefficient between S21 and S22 is obtained by inputting the different feature information and other signals shown in Equations 7 and 8, respectively, and by using the coefficients of S21 and the characteristics of each chaotic element arranged in S22. When the coefficient is determined as described above, when w ₁₂ > 2.8, S2
The output signal of the chaotic elements arranged in 2 diverges and no longer generates chaotic motion. Therefore, in this embodiment,
The coupling coefficient between the chaotic elements arranged in S21 and S22 was set as w ₁₂ = 2.0. At this time, the output signal of the chaotic element arranged in S22 is as shown in FIG.

Each chaotic element arranged in S23 is given by
When n = 2 to 4 shown in (a), a = 2.5 and w ₂₃ = so that a chaotic element which exhibits a stable motion before the signal input and a different characteristic signal after the signal input generates a chaotic motion. 0.
When it is determined to be 6, the output shown in FIG. 6 is performed. Also, S2
FIG. 7 shows a case where the change amount Δ of the output value between the time t and the time t−1 is calculated from Expression 9 with respect to the output signal value of each chaotic element of No. 3. [ _Equation 9] Δ = | O _3xy | _t −O _3xy | _t-1 | As shown in FIG. 7, the signal input start time, that is, time t = 0
Regarding the change amount of the output signal value in the above, the difference between the different feature information and other signals does not appear on the plane coordinates.

At time t = 1, the difference between the different feature information of the output signal value and other signals starts to appear and develops until time t = 3. Further, the difference between the different feature information of the output signal value and other signals may decrease with time as shown at time t = 5.

As described above, by using the present invention, it is possible to detect different characteristic information.

Example 2 By using the present invention,
It is possible to extract a part having a different feature from the surroundings.

For example, when x and y shown in the plane coordinates are considered as a pixel unit, the signal which moves periodically as shown in Expression 7 is expressed as x = 5-10.
(Pixels) and y = 5-10 (pixels).
When the signal shown in Equation 8 is input to other areas, it is possible to extract the boundary line and each vertex of the area where the signal shown in Equation 7 is input as shown in FIG.

As described above, by using the present invention, it is possible to extract a boundary between different feature information.

[Embodiment 3] With respect to the boundary line extraction described in Embodiment 2, even when white noise is presented to the input signal, the boundary line can be extracted in the present invention.

For example, the characteristics and the coupling coefficient of the chaotic element shown in the first and second embodiments are used, and the input signal performed in the second embodiment is compared with the signal intensity difference at the boundary of the input signal by 25 minutes. When white noise having a signal strength of 1 is presented, the amount of change in the output signal from S23 (Equation 9) is as shown in FIG. As described above, according to the present invention, even when white noise is present in an input signal, it is possible to extract a boundary line of a region where different feature information exists.

Each numerical value in the above embodiment is an example of the embodiment of the present invention. In the present invention, if the operation is performed within the above-described range, the object of the present invention can be implemented.

[0034]

According to the present invention, as described above, a chaotic element, which is performing a stable motion before a signal is input, generates a chaotic motion due to the transient vibration of the input signal.

Further, according to the present invention, a difference between signals undergoing chaotic motion is calculated, and a chaotic motion is generated by a transient vibration of a chaotic element performing stable motion by using the difference as an input signal. Can be.

Further, according to the present invention, a chaotic element that is performing a stable motion can be caused to perform a chaotic motion with respect to an output signal that has generated a chaotic motion due to a difference.

According to the present invention, it is possible to detect the presence of information having different characteristics from signals input by the above-described method. Further, the boundary line of the different feature information can be extracted by the above-described method according to the present invention.

[Brief description of the drawings]

FIG. 1 is a flowchart for implementing the present invention.

FIG. 2 is a schematic diagram of a chaotic device for implementing the present invention.

FIG. 3 is a time-output signal strength characteristic diagram of a signal input in the first embodiment.

FIG. 4 is a time-output signal strength characteristic diagram of the chaotic element arranged in S21 shown in FIG. 2 when the input signal shown in the first embodiment is executed, and (1) is an input signal shown in Expression 7; Time-output signal strength characteristics for the signal, (2)
Is a time-output signal strength characteristic for the input signal indicated by.

FIG. 5 is a time-output signal strength characteristic diagram of the chaotic element arranged in S22 shown in FIG. 2 when the input signal shown in the first embodiment is implemented. The time-output signal strength characteristic of the chaotic element arranged in S22 when the difference calculation of the output signal of S21 is performed based on the input signal shown, (2) is the output signal of S21 due to the input signals shown in Formula 8 S22 when the difference calculation is performed
5 is a time-output signal strength characteristic of the chaotic element arranged in FIG.

FIG. 6 is a time-output signal strength characteristic diagram of the chaotic element arranged in S23 shown in FIG. 2 when the input signal shown in the first embodiment is executed, and (1) is an input signal shown in Expression 7; Time-output signal strength characteristics for the signal, (2)
Is a time-output signal strength characteristic for the input signal indicated by.

FIG. 7 shows a calculation result of a signal intensity difference shown in Expression 9 when the input signal shown in the first embodiment is performed, and (1)
Is the result immediately after presenting the input signals shown in Equations 7 and 8 (t = 0), and (2) t = 1, (3) t = 2
(4) Calculation results at t = 3, (5) t = 4, (6) t = 5.

FIG. 8 is a calculation result of a signal strength difference shown in Expression 9 when the input signal shown in Embodiment 2 is performed, and (1) is a result immediately after the input signals shown in Expressions 7 and 8 are presented. (T = 0), and (2) t = 1, (3) Calculation results at t = 2.

FIG. 9 is a calculation result (t = 2) of a signal intensity difference shown in Expression 9 when the signal obtained by adding the white noise shown in Embodiment 3 to the input signal shown in Embodiment 2 is input.

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Kage Ryu 2217-20 Hayashi-cho, Takamatsu-shi, Kagawa F-term (reference) 5L096 FA06 HA03 HA11 Faculty of Engineering, Kagawa University

Claims (5)

[Claims] 1. A method of spatially arranging elements that transition from a stable motion to a chaotic motion by an input to an element defined by a function that generates a chaotic motion. 2. A method according to claim 1, wherein a difference between spatially adjacent elements is calculated so that only a portion related to a space where different features are arranged transitions from a stable motion to a chaotic motion. Method of arranging elements in 3. A method of integrating the outputs from claim 2 so as to arrange elements such that only a spatial portion related to a different feature generates a chaotic motion from a stable motion. 4. A method for detecting only different feature information from a large number of input signals by the combination of claim 1, claim 2, and claim 3. 5. A method for detecting a boundary between an object and a background by the method described in claim 4.