Fourier transform device, fourier transform method, and program

The Fourier transform device and method allow for the first time graph Fourier transform on directed graphs by combining undirected and directed graph signal processing, addressing the challenge of noise removal in directed graph signals.

WO2024247061A1PCT designated stage expired Publication Date: 2024-12-05NT T INC
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2023/019969
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-29
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

There is no known method to perform graph Fourier transform on directed graph signals, making it difficult to remove noise from such signals.

Method used

A Fourier transform device and method that acquires and processes both undirected and directed graph signals, using a directed graph signal Fourier transform unit to perform graph Fourier transform based on the results from an undirected graph signal Fourier transform, enabling the combination and processing of both types of signals to achieve graph Fourier transform on directed graphs.

Benefits of technology

Enables graph Fourier transform on directed graphs, allowing for frequency analysis, filtering, and noise removal, which was previously not possible.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023019969_05122024_PF_FP_ABST
    Figure JP2023019969_05122024_PF_FP_ABST
Patent Text Reader

Abstract

This Fourier transform device comprises: a directed graph signal acquisition unit for acquiring a directed graph signal; an undirected graph signal Fourier transform unit for performing a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal; and a directed graph signal Fourier transform unit for performing a graph Fourier transform of the directed graph signal on the basis of the directed graph signal acquired by the directed graph signal acquisition unit and the result of the graph Fourier transform performed by the undirected graph signal Fourier transform unit.
Need to check novelty before this filing date? Find Prior Art

Description

Fourier transform device, Fourier transform method and program

[0001] The present invention relates to a Fourier transform device, a Fourier transform method, and a program.

[0002] Graph signal processing is a data processing technique that extends conventional signal processing to numerical sequences with irregular connections (graph signals). In particular, the graph Fourier transform is an important technique for frequency analysis, filtering, and noise reduction of graph signals. A high-speed calculation method for graph signals is known for certain undirected graph signals, and is disclosed in, for example, Non-Patent Document 1.

[0003] Yuichi Tanaka, "Graph Signal Processing", IEICE Fundamentals Review Vol.8 No.1, pp. 15-29, 2014

[0004] However, while it has been possible to perform graph Fourier transform processing on undirected graph signals, methods for performing graph Fourier transform on directed graph signals have not been known, making it difficult to perform frequency analysis, filtering, and noise removal on directed graph signals.

[0005] In view of the above circumstances, an object of the present invention is to provide a technique that can perform a graph Fourier transform on a directed graph.

[0006] One aspect of the present invention is a Fourier transform device comprising: a directed graph signal acquisition unit that acquires a directed graph signal; an undirected graph signal Fourier transform unit that performs a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal; and a directed graph signal Fourier transform unit that performs a graph Fourier transform of the directed graph signal based on the directed graph signal acquired by the directed graph signal acquisition unit and the result of the graph Fourier transform performed by the undirected graph signal Fourier transform unit.

[0007] One aspect of the present invention is a Fourier transform method that acquires a directed graph signal, performs a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal, and performs a graph Fourier transform of the directed graph signal based on the acquired directed graph signal and the result of performing the graph Fourier transform on the undirected graph signal.

[0008] One aspect of the present invention is a program that causes a computer of a Fourier transform device to acquire a directed graph signal, perform a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal, and perform a graph Fourier transform of the directed graph signal based on the acquired directed graph signal and a result of performing the graph Fourier transform on the undirected graph signal.

[0009] The present invention allows for the graph Fourier transform to be performed on directed graphs.

[0010] 1 is a diagram showing an example of an undirected graph according to an embodiment of the present invention. FIG. 2 is a diagram showing an example of a directed graph according to an embodiment of the present invention. FIG. 3 is a schematic block diagram showing a configuration of a Fourier transform device according to a first embodiment of the present invention. FIG. 4 is a diagram showing an example of an undirected path graph signal acquired by an undirected graph signal acquisition unit (FIG. 3) in the first embodiment of the present invention. FIG. 5 is a diagram showing an example of a directed path graph signal acquired by a directed graph signal acquisition unit (FIG. 3) in the first embodiment of the present invention. FIG. 6 is a flowchart showing an outline of processing by a Fourier transform device according to the first embodiment of the present invention. FIG. 7 is a diagram showing an example of an undirected ring graph signal acquired by an undirected graph signal acquisition unit (FIG. 3) in the second embodiment of the present invention. FIG. 8 is a diagram showing an example of a directed ring graph signal acquired by a directed graph signal acquisition unit (FIG. 3) in the second embodiment of the present invention. FIG. 9 is a diagram showing another example of a directed ring graph signal acquired by a directed graph signal acquisition unit (FIG. 3) in the second embodiment of the present invention. FIG. 10 is a flowchart showing an outline of processing by a Fourier transform device according to the second embodiment of the present invention. FIG. 11 is a diagram showing an example of a first undirected graph acquired by an undirected graph signal acquisition unit (FIG. 3) in the third embodiment of the present invention. FIG. 12 is a diagram showing an example of a second undirected graph acquired by an undirected graph signal acquisition unit (FIG. 3) in the third embodiment of the present invention. FIG. 10 is a diagram showing an example of an undirected Cartesian product graph generated by the undirected graph signal acquisition unit ( FIG. 3 ) in the third embodiment of the present invention. FIG. 11 is a diagram showing an example of a first directed graph acquired by the directed graph signal acquisition unit ( FIG. 3 ) in the third embodiment of the present invention. FIG. 12 is a diagram showing an example of a second directed graph acquired by the directed graph signal acquisition unit ( FIG. 3 ) in the third embodiment of the present invention. FIG. 13 is a flowchart showing an overview of the processing of the Fourier transform device according to the third embodiment of the present invention.

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. First, a graph Fourier transform of an undirected graph signal will be described. Note that the graph Fourier transform is a Fourier transform of a graph signal.

[0012] FIG. 1 is a diagram showing an example of an undirected graph according to an embodiment of the present invention. An undirected graph is a graph in which the edges connecting the vertices that make up the graph have no direction. The undirected graph shown in FIG. 1 is composed of five vertices: 0, 1, 2, 3, and 4. Vertex 0 is connected to vertices 1, 2, 3, and 4. Vertex 1 is connected to vertices 0 and 2. Vertex 2 is connected to vertices 0, 1, and 3. Vertex 3 is connected to vertices 0, 2, and 4. Vertex 4 is connected to vertices 0 and 3.

[0013] If an undirected graph is denoted by G, it is expressed by the following formula (1): In the following formula (1), v is a vertex set indicating the vertices that make up the undirected graph G, and A is an adjacency matrix.

[0014] ...(1)

[0015] The degree matrix D is expressed by the following equation (2).

[0016] ... (2)

[0017] order d m is expressed by the following equation (3).

[0018] ...(3)

[0019] The graph Laplacian matrix L is expressed by the following equation (4). ...(4)

[0020] The graph Fourier transform is performed based on the following equation (5).

[0021] ...(5)

[0022] Here, the spectrum of the graph is expressed as follows:

[0023]

[0024] v is a vertex set indicating the vertices that make up the graph. A is an adjacency matrix, whose components take values ​​of 1 or 0. The vertex set v, adjacency matrix A, and degree matrix of the undirected graph shown in Figure 1 are expressed by the following equation (6).

[0025] ...(6)

[0026] The graph Laplacian matrix L of the undirected graph shown in FIG. 1 is expressed by the following equation (7).

[0027] ... (7)

[0028] The undirected graph U shown in FIG. 1 is expressed by the following formula (8).

[0029] ...(8)

[0030] The Λ of the undirected graph shown in FIG. 1 is expressed by the following formula (9).

[0031] ...(9)

[0032] Next, the graph Fourier transform of a directed graph signal will be described.

[0033] FIG. 2 is a diagram showing an example of a directed graph according to an embodiment of the present invention. A directed graph is a graph in which the edges connecting the vertices that make up the graph have directionality. The directed graph shown in FIG. 2 is composed of five vertices: 0, 1, 2, 3, and 4. Vertex 0 is connected to vertices 1, 2, 3, and 4. Vertex 1 is connected to vertices 0 and 2. Vertex 2 is connected to vertices 0, 1, and 3. Vertex 3 is connected to vertices 0, 2, and 4. Vertex 4 is connected to vertices 0 and 3. The edge connecting vertices 0 and 1 points from vertex 0 to vertex 1. The edge connecting vertex 0 and 2 points from vertex 2 to vertex 0. The edge connecting vertex 0 and 3 points from vertex 0 to vertex 3. The edge connecting vertex 0 and vertex 4 points from vertex 4 to vertex 0. The edge connecting vertex 1 and vertex 2 points from vertex 1 to vertex 2. The edge connecting vertex 2 and vertex 3 is directed from vertex 3 to vertex 2. The edge connecting vertex 3 and vertex 4 is directed from vertex 3 to vertex 4.

[0034] If the directed graph is G, it is expressed by the following formula (10): In the following formula (10), v is a vertex set indicating the vertices that make up the directed graph G, and A indicates an adjacency matrix.

[0035] ...(10)

[0036] If the near-adjacency matrix is ​​C, it is expressed by the following equation (11).

[0037] ...(11)

[0038] Here, A + represents the symmetric component and is expressed by the following equation (12).

[0039] ...(12)

[0040] Also, A - represents an alternating component and is expressed by the following equation (13).

[0041] ...(13)

[0042] If the degree matrix of the directed graph is D, it is expressed by the following equation (14).

[0043] ...(14)

[0044] Let the degree of the directed graph be d. m Then, it is expressed by the following equation (15).

[0045] ...(15)

[0046] If the quasi-graph Laplacian matrix of the directed graph is L, it is expressed by the following equation (16).

[0047] ...(16) The quasi-graph Fourier transform of the directed graph is performed by the following equation (17).

[0048] ...(17)

[0049] The spectrum of a directed graph is expressed as follows:

[0050]

[0051] A vertex set v indicating the vertices that make up the directed graph shown in FIG. 2 is expressed by the following formula (18).

[0052] ...(18)

[0053] Symmetric component A of the directed graph shown in FIG. + and alternating component A - is expressed by the following equation (19).

[0054] ...(19)

[0055] The quasi-adjacency matrix C and the degree matrix of the directed graph shown in FIG. 2 are expressed by the following equation (20).

[0056] ...(20)

[0057] The quasi-graph Laplacian matrix L of the directed graph shown in Fig. 2 is expressed by the following formula (21): In the following formula (21), * denotes an arbitrary complex number.

[0058] ...(21)

[0059] U of the directed graph shown in FIG. 2 is expressed by the following formula (22).

[0060] ...(22)

[0061] The directed graph Λ shown in FIG. 2 is expressed by the following formula (23).

[0062] ...(23)

[0063] In this way, the graph Laplacian matrix L of a directed graph without backward edges is d is the graph Laplacian matrix L of the corresponding undirected graph, as shown in the following equation (24). u and can be expressed as a Hadamard product of a matrix V.

[0064] ...(24)

[0065] Here, the diagonal components v of the matrix V m,m can be expressed as the following equation (25).

[0066] ...(25)

[0067] Also, the off-diagonal components v of the matrix V m,n can be expressed as the following equation (26).

[0068] ...(26)

[0069] [First Embodiment] Next, a Fourier transform device according to a first embodiment of the present invention will be described. The Fourier transform device according to the first embodiment performs a graph Fourier transform on a directed path graph. A path graph is a graph in which multiple vertices are linearly connected.

[0070] 3 is a schematic block diagram showing the configuration of a Fourier transform device according to a first embodiment of the present invention. The Fourier transform device 100 shown in FIG. 3 includes an undirected graph signal acquisition unit 11, an undirected graph signal Fourier transform unit 12, a directed graph signal acquisition unit 13, a predetermined vector calculation unit 14, a Hadamard product calculation unit 15, a directed graph signal Fourier transform unit 16, and a storage unit 17.

[0071] The undirected graph signal acquisition unit 11 acquires an undirected path graph signal from the outside and outputs it to the undirected graph signal Fourier transform unit 12. The undirected graph signal Fourier transform unit 12 performs a graph Fourier transform on the undirected path graph signal input from the undirected graph signal acquisition unit 11 and outputs it to the directed graph signal Fourier transform unit 16. The directed graph signal acquisition unit 13 acquires a directed path graph signal from the outside and outputs it to the Hadamard product calculation unit 15. The predetermined vector calculation unit 14 calculates a predetermined vector used in the Hadamard product calculation unit 15 and outputs it to the Hadamard product calculation unit 15.

[0072] The Hadamard product calculation unit 15 calculates a Hadamard product based on the directed path graph signal input from the directed graph signal acquisition unit 13 and the predetermined vector input from the predetermined vector calculation unit 14, and outputs the calculated product to the directed graph signal Fourier transform unit 16. The directed graph signal Fourier transform unit 16 performs a Fourier transform on the directed path graph signal based on the Fourier transform result of the undirected path graph signal input from the undirected graph signal Fourier transform unit 12 and the Hadamard product input from the Hadamard product calculation unit 15, and outputs the Fourier transform result to the outside. The storage unit 17 is composed of a memory or the like, and stores data processed by the undirected graph signal acquisition unit 11, the undirected graph signal Fourier transform unit 12, the directed graph signal acquisition unit 13, the predetermined vector calculation unit 14, the Hadamard product calculation unit 15, and the directed graph signal Fourier transform unit 16.

[0073] 4 is a diagram showing an example of an undirected path graph signal acquired by the undirected graph signal acquisition unit 11 (FIG. 3) in the first embodiment of the present invention. The undirected path graph shown in FIG. 4 is composed of eight vertices: 0, 1, 2, 3, 4, 5, 6, and 7. Vertex 0 is connected to vertex 1. Vertex 1 is connected to vertex 2. Vertex 2 is connected to vertex 3. Vertex 3 is connected to vertex 4. Vertex 4 is connected to vertex 5. Vertex 5 is connected to vertex 6. Vertex 6 is connected to vertex 7.

[0074] The Laplacian matrix L of the undirected path graph in Figure 4 upath is expressed by the following equation (27).

[0075] ...(27)

[0076] This Laplacian matrix L upath can be diagonalized by the discrete cosine transform matrix. Therefore, the graph Fourier transform of an undirected path graph can be performed quickly using the fast discrete cosine transform.

[0077] FIG. 5 is a diagram showing an example of a directed path graph signal acquired by the directed graph signal acquisition unit 13 ( FIG. 3 ) in the first embodiment of the present invention. The directed path graph shown in FIG. 5 is composed of eight vertices: 0, 1, 2, 3, 4, 5, 6, and 7. Vertex 0 is connected to vertex 1. Vertex 1 is connected to vertex 2. Vertex 2 is connected to vertex 3. Vertex 3 is connected to vertex 4. Vertex 4 is connected to vertex 5. Vertex 5 is connected to vertex 6. Vertex 6 is connected to vertex 7. The edge connecting vertex 0 and vertex 1 is oriented from vertex 0 to vertex 1. The edge connecting vertex 1 and vertex 2 is oriented from vertex 1 to vertex 2. The edge connecting vertex 2 and vertex 3 is oriented from vertex 3 to vertex 2. The edge connecting vertex 3 and vertex 4 is oriented from vertex 3 to vertex 4. The edge connecting vertex 4 and vertex 5 is oriented from vertex 5 to vertex 4. The edge connecting vertex 5 and vertex 6 is directed from vertex 6 to vertex 5. The edge connecting vertex 6 and vertex 7 is directed from vertex 6 to vertex 7.

[0078] The Laplacian matrix L of the directed path graph in FIG. dpath is expressed by the following equation (28).

[0079] ...(28)

[0080] Here, the vector v is defined as follows:

[0081] ...(29)

[0082] For example, in the case of the directed path graph shown in FIG. 5, the vector v may be expressed as in the following equation (30).

[0083] ...(30)

[0084] In this case, the matrix V = vv H Considering this, the following equation (31) is obtained.

[0085] ...(31)

[0086] Here, an arbitrary matrix X and vectors a and b are expressed as in the following equation (32).

[0087] ...(32)

[0088] In this case, the relationship of the following equation (33) holds between an arbitrary matrix X and vectors a and b.

[0089] ...(33)

[0090] Therefore, the following equation (34) is obtained.

[0091] ...(34)

[0092] Laplacian matrix L of the undirected path graph upath The eigenvalue decomposition of is expressed as in the following equation (35).

[0093] ...(35)

[0094] In this case, the Laplacian matrix L of the directed path graph dpath can be decomposed into eigenvalues ​​as shown in the following equation (36).

[0095] ...(36)

[0096] As a result, the graph Fourier transform of the directed path graph shown in FIG. 5 can be performed by the following equation (37).

[0097] ...(37)

[0098] That is, the graph Fourier transform of a directed path graph (see Figure 5) is the vector (also referred to as a predetermined vector) and a graph Fourier transform (i.e., a fast discrete cosine transform) of the undirected path graph (see FIG. 4).

[0099] 6 is a flowchart showing an outline of the processing of the Fourier transform device according to the first embodiment of the present invention. First, the undirected graph signal acquisition unit 11 (FIG. 3) acquires an undirected path graph signal (e.g., FIG. 4) from outside the Fourier transform device 100 (step S11). Next, the undirected graph signal Fourier transform unit 12 (FIG. 3) performs a Fourier transform on the undirected path graph signal acquired in step S11 using a known method (step S12). Next, the directed graph signal acquisition unit 13 (FIG. 3) acquires a directed path graph signal (e.g., FIG. 5) from outside the Fourier transform device 100 (step S13).

[0100] Next, the predetermined vector calculation unit 14 (FIG. 3) calculates the predetermined vector by using the above-mentioned equation (30) (step S14). Next, the Hadamard product calculation unit 15 (FIG. 3) calculates the Hadamard product of the directed path graph signal acquired in step S13 and the predetermined vector calculated in step S14 (step S15). Next, the directed graph signal Fourier transform unit 16 (FIG. 3) performs a Fourier transform of the directed path graph signal by using the above-mentioned equation (37) based on the undirected path graph signal Fourier transformed in step S12 and the Hadamard product calculated in step S15 (step S16).

[0101] 6, the case where an undirected path graph signal is acquired in step S11 and a directed path graph signal is acquired in step S13 has been described, but the present invention is not limited to this. For example, the directed path graph signal may be acquired before the undirected path graph signal, or the undirected path graph signal and the directed path graph signal may be acquired simultaneously.

[0102] The position of each vertex in the undirected path graph signal corresponds to the position of each vertex in the directed path graph signal. Therefore, instead of separately acquiring the undirected path graph signal and the directed path graph signal, the undirected path graph signal may be generated and used by removing information about the directionality between vertices from the directed path graph signal. In this way, there is no need to provide the undirected graph signal acquisition unit 11 in FIG. 3, and the device configuration can be simplified.

[0103] According to the first embodiment, a graph Fourier transform can be performed on a directed path graph such as that shown in FIG. 5, and frequency analysis, filtering, and noise removal can be performed on directed path graph signals as well as undirected path graph signals.

[0104] Second Embodiment Next, a Fourier transform device according to a second embodiment of the present invention will be described. The Fourier transform device according to the second embodiment performs a graph Fourier transform on a directed ring graph. A ring graph is a graph in which multiple vertices are connected in a ring shape.

[0105] The configuration of the Fourier transform device according to the second embodiment is similar to the configuration of the Fourier transform device 100 (FIG. 3) described in the first embodiment, and therefore the second embodiment will also be described using the Fourier transform device 100 shown in FIG. 3.

[0106] 7 is a diagram showing an example of an undirected ring graph signal acquired by the undirected graph signal acquisition unit 11 (FIG. 3) in the second embodiment of the present invention. The undirected ring graph shown in FIG. 7 is composed of eight vertices: 0, 1, 2, 3, 4, 5, 6, and 7. Vertex 0 is connected to vertex 1. Vertex 1 is connected to vertex 2. Vertex 2 is connected to vertex 3. Vertex 3 is connected to vertex 4. Vertex 4 is connected to vertex 5. Vertex 5 is connected to vertex 6. Vertex 6 is connected to vertex 7. Furthermore, vertex 7 is connected to vertex 0.

[0107] The Laplacian matrix L of the undirected ring graph in FIG. uring is expressed by the following equation (38).

[0108] ...(38)

[0109] This Laplacian matrix L uring can be diagonalized by the discrete Fourier transform matrix. Therefore, the graph Fourier transform of an undirected ring graph can be performed quickly using the fast discrete Fourier transform.

[0110] 8A and 8B are diagrams showing an example of a directed ring graph signal acquired by the directed graph signal acquisition unit 13 (FIG. 3) in the second embodiment of the present invention. Here, the directed ring graph signal in FIG. 8A will be described, and an explanation of the directed ring graph signal in FIG. 8B will be omitted.

[0111] The directed ring graph shown in FIG. 8A consists of eight vertices: 0, 1, 2, 3, 4, 5, 6, and 7. Vertex 0 is connected to vertex 1. Vertex 1 is connected to vertex 2. Vertex 2 is connected to vertex 3. Vertex 3 is connected to vertex 4. Vertex 4 is connected to vertex 5. Vertex 5 is connected to vertex 6. Vertex 6 is connected to vertex 7. Furthermore, vertex 7 is connected to vertex 0. The edge connecting vertex 0 and vertex 1 is oriented from vertex 0 to vertex 1. The edge connecting vertex 1 and vertex 2 is oriented from vertex 1 to vertex 2. The edge connecting vertex 2 and vertex 3 is oriented from vertex 3 to vertex 2. The edge connecting vertex 3 and vertex 4 is oriented from vertex 3 to vertex 4. The edge connecting vertex 4 and vertex 5 is oriented from vertex 5 to vertex 4. The edge connecting vertex 5 and vertex 6 is oriented from vertex 6 to vertex 5. The edge connecting vertex 6 and vertex 7 is directed from vertex 6 to vertex 7. Furthermore, the edge connecting vertex 7 and vertex 0 is directed from vertex 0 to vertex 7.

[0112] The Laplacian matrix L of the directed ring graph in FIG. dring is expressed by the following equation (39).

[0113] ...(39)

[0114] Here, the vector v is defined as follows:

[0115] ...(40)

[0116] Here, the matrix v H Considering this, the following equation (41) holds true.

[0117] ...(41)

[0118] However, in general, teeth, The condition for these to match is the following equation (42).

[0119] ...(42)

[0120] Only when this congruence holds true, the following equation (43) is obtained.

[0121] ...(43)

[0122] Laplacian matrix L of an undirected ring graph uring The eigenvalue decomposition of is expressed as in the following equation (44).

[0123] ...(44)

[0124] In this case, the Laplacian matrix L of the directed ring graph dring can be decomposed into eigenvalues ​​as shown in the following equation (45).

[0125] ...(45)

[0126] As a result, the graph Fourier transform of the directed ring graph shown in FIG. 8 can be performed by the following equation (46).

[0127] ...(46)

[0128] That is, the graph Fourier transform of a directed ring graph (see Figure 8) is the vector (also referred to as a predetermined vector) and a graph Fourier transform (i.e., a fast discrete Fourier transform) of an undirected ring graph (see FIG. 7).

[0129] 9 is a flowchart showing an outline of the processing of the Fourier transform device according to the second embodiment of the present invention. First, the undirected graph signal acquisition unit 11 (FIG. 3) acquires an undirected ring graph signal (e.g., FIG. 7) from outside the Fourier transform device 100 (step S21). Next, the undirected graph signal Fourier transform unit 12 (FIG. 3) performs a Fourier transform on the undirected ring graph signal acquired in step S21 using a known method (step S22). Next, the directed graph signal acquisition unit 13 (FIG. 3) acquires a directed ring graph signal (e.g., FIG. 8) from outside the Fourier transform device 100 (step S23).

[0130] Next, the predetermined vector calculation unit 14 (FIG. 3) calculates the predetermined vector by using the above-mentioned equation (40) (step S24). Next, the Hadamard product calculation unit 15 (FIG. 3) calculates the Hadamard product of the directed ring graph signal acquired in step S23 and the predetermined vector calculated in step S24 (step S25). Next, the directed graph signal Fourier transform unit 16 (FIG. 3) performs a Fourier transform of the directed ring graph signal by using the above-mentioned equation (46) based on the undirected ring graph signal Fourier transformed in step S22 and the Hadamard product calculated in step S25 (step S26).

[0131] 9, the case where an undirected ring graph signal is acquired in step S21 and a directed ring graph signal is acquired in step S23 has been described, but this is not limiting. For example, the directed ring graph signal may be acquired before the undirected ring graph signal, or the undirected ring graph signal and the directed ring graph signal may be acquired simultaneously.

[0132] The position of each vertex of the undirected ring graph signal corresponds to the position of each vertex of the directed ring graph signal. Therefore, instead of separately acquiring the undirected ring graph signal and the directed ring graph signal, the undirected ring graph signal may be generated and used by removing information about the directionality between vertices from the directed ring graph signal. In this way, there is no need to provide the undirected graph signal acquisition unit 11 in FIG. 3, and the device configuration can be simplified.

[0133] According to the second embodiment, a graph Fourier transform can be performed on a directed ring graph as shown in FIG. 8, and frequency analysis, filtering, and noise removal can be performed on directed ring graph signals as well as undirected ring graph signals.

[0134] Third Embodiment Next, a Fourier transform device according to a third embodiment of the present invention will be described. The Fourier transform device according to the third embodiment performs a graph Fourier transform on a directed product graph. A direct product graph is a graph created by multiplying one graph by another.

[0135] The configuration of the Fourier transform device according to the third embodiment is similar to the configuration of the Fourier transform device 100 (FIG. 3) described in the first embodiment, and therefore the third embodiment will also be described using the Fourier transform device 100 shown in FIG. 3.

[0136] The graph G = (ν, A) is the first graph G 1 =(ν 1 , A 1 ) and the second graph G 2 =(ν 2 , A 2 ) is defined as a direct product graph when the relationship of the following equation (47) holds.

[0137] ...(47)

[0138] where the matrix operator is the Kronecker sum. Two square matrices X and Y are expressed as in the following equation (48).

[0139] ...(48)

[0140] In this case, the definition as in the following equation (49) holds.

[0141] ...(49)

[0142] where the matrix operator is the Kronecker product. A specific example of a Cartesian product graph and a method for fast calculation of the graph Fourier transform will be described later.

[0143] FIG. 10A shows the first undirected graph G acquired by the undirected graph signal acquisition unit 11 (FIG. 3) in the third embodiment of the present invention. 11 10A is a diagram illustrating an example of the first undirected graph G 11 is the same as the undirected graph explained in FIG. 1, so its explanation will be omitted.

[0144] FIG. 10B shows the second undirected graph G acquired by the undirected graph signal acquisition unit 11 (FIG. 3) in the third embodiment of the present invention. 12 1 is a diagram illustrating an example of a second undirected graph G 12 is composed of four vertices: 0, 1, 2, and 3. Vertex 0 is connected to vertex 1. Vertex 1 is connected to vertex 2. Vertex 2 is connected to vertex 3.

[0145] FIG. 10C shows the undirected Cartesian product graph G generated by the undirected graph signal acquisition unit 11 (FIG. 3) in the third embodiment of the present invention. 13 The undirected graph signal acquisition unit 11 acquires a first undirected graph G consisting of five vertices. 11 and a second undirected graph G consisting of four vertices 12 By multiplying these graphs, an undirected Cartesian product graph G consisting of 20 vertices (5×4=) is obtained as shown in FIG. 10C. 13 Generate.

[0146] First undirected graph G 11 The Laplacian matrix of L 1 Let the second undirected graph G 12 The Laplacian matrix of L 2 Then, the undirected Cartesian product graph G 13 The Laplacian matrix L of is expressed as a Kronecker sum as shown in the following equation (50).

[0147] ...(50)

[0148] From this property and the property of the eigenvalue decomposition of the matrix exponential, the undirected Cartesian product graph G 13 The eigenvalue decomposition of the Laplacian matrix L is expressed by the following equation (51).

[0149] ...(51)

[0150] Therefore, the graph Fourier transform of an undirected graph is equivalent to performing the graph Fourier transform of each factor graph vertically and horizontally. Next, a directed product graph will be described.

[0151] FIG. 11A shows the first directed graph G acquired by the directed graph signal acquisition unit 13 (FIG. 3) in the third embodiment of the present invention. 21 11A is a diagram illustrating an example of the first directed graph G 21 is the same as the directed graph explained in FIG. 2, and therefore its explanation will be omitted.

[0152] FIG. 11B shows the second directed graph G acquired by the directed graph signal acquisition unit 13 (FIG. 3) in the third embodiment of the present invention. 22 1 is a diagram illustrating an example of a second directed graph G 22 is composed of four vertices: 0, 1, 2, and 3. Vertex 0 is connected to vertex 1. Vertex 1 is connected to vertex 2. Vertex 2 is connected to vertex 3. The edge connecting vertices 0 and 1 points from vertex 0 to vertex 1. The edge connecting vertex 1 and vertex 2 points from vertex 1 to vertex 2. The edge connecting vertex 2 and vertex 3 points from vertex 3 to vertex 2.

[0153] FIG. 11C shows the directed product graph G generated by the directed graph signal acquisition unit 13 (FIG. 3) in the third embodiment of the present invention. 23 The digraph signal acquisition unit 13 acquires a first digraph G consisting of five vertices. 21 and a second directed graph G consisting of four vertices 22 By multiplying these graphs, a Cartesian product graph G consisting of 20 vertices (5×4=) is obtained as shown in FIG. 11C. 23 Generate.

[0154] First directed graph G 21 The Laplacian matrix of L 1 Let the second directed graph G 22 The Laplacian matrix of L 2 Then, the directed product graph G 23 The Laplacian matrix L of is expressed as a Kronecker sum as shown in the following equation (52).

[0155] ...(52)

[0156] Therefore, the eigenvalue decomposition of the Laplacian matrix L of the directed product graph G23 is expressed as in the following equation (53).

[0157] ...(53)

[0158] Therefore, the graph Fourier transform of a directed graph is equivalent to performing the graph Fourier transform of each factor graph vertically and horizontally.

[0159] When holds, To prove this, we present the following three propositions (Proposition 1), (Proposition 2), and (Proposition 3).

[0160] (Proposition 1) (Proposition 2) (Proposition 3)

[0161] The above (Proposition 1) is proved by the following (Proof 1).

[0162] (Proof 1)

[0163] Furthermore, the above (Proposition 2) is proven by the following (Proof 2).

[0164] (Proof 2)

[0165] Furthermore, the above (Proposition 3) is proven by the following (Proof 3).

[0166] (Proof 3) Focusing on the diagonal elements, It is sufficient to show that

[0167] From the above, the following equation (54) holds true.

[0168] ...(54) holds true.

[0169] 12 is a flowchart showing an outline of the processing of the Fourier transform device according to the third embodiment of the present invention. First, the undirected graph signal acquirer 11 (FIG. 3) acquires a first undirected graph (e.g., FIG. 10A) and a second undirected graph (e.g., FIG. 10B) from outside the Fourier transform device 100 (step S31). Next, the undirected graph signal acquirer 11 (FIG. 3) generates an undirected Cartesian product graph (e.g., FIG. 10C) by multiplying the first undirected graph and the second undirected graph acquired in step S31 (step S32). Next, the undirected graph signal Fourier transformer 12 (FIG. 3) performs a Fourier transform on the undirected Cartesian product graph generated in step S32 using a known method (step S33).

[0170] Next, the digraph signal acquirer 13 (FIG. 3) acquires a first digraph (e.g., FIG. 11A) and a second digraph (e.g., FIG. 11B) from outside the Fourier transform device 100 (step S34). Next, the digraph signal acquirer 13 (FIG. 3) generates a directed product graph (e.g., FIG. 11C) by multiplying the first digraph and the second digraph acquired in step S34 (step S35).

[0171] Next, the predetermined vector calculation unit 14 ( FIG. 3 ) calculates the predetermined vector by using the above-mentioned formula (30) (or the above-mentioned formula (40)) (step S36). Next, the Hadamard product calculation unit 15 ( FIG. 3 ) calculates the Hadamard product of the directed Cartesian product graph acquired in step S35 and the predetermined vector calculated in step S36 (step S37). Next, the directed graph signal Fourier transform unit 16 ( FIG. 3 ) performs a Fourier transform of the directed Cartesian product graph signal by using the above-mentioned formula (37) (or the above-mentioned formula (46)) based on the undirected Cartesian product graph signal Fourier transformed in step S33 and the Hadamard product calculated in step S37 (step S38).

[0172] 12 illustrates a case in which the first undirected graph signal and the second undirected graph signal are acquired in step S31, and the first directed graph signal and the second directed graph signal are acquired in step S34, but this is not limiting. For example, the first directed graph signal and the second directed graph signal may be acquired before the first undirected graph signal and the second undirected graph signal, or the first undirected graph signal and the second undirected graph signal may be acquired simultaneously with the first directed graph signal and the second directed graph signal.

[0173] The position of each vertex of the undirected Cartesian product graph signal corresponds to the position of each vertex of the directed Cartesian product graph signal. Therefore, instead of separately acquiring the first undirected graph signal and the second undirected graph signal and the first directed graph signal and the second directed graph signal, an undirected Cartesian product graph signal may be generated and used by removing information about the directionality between vertices from the directed Cartesian product graph signal. This eliminates the need for the undirected graph signal acquisition unit 11 in FIG. 3, thereby simplifying the device configuration.

[0174] According to the third embodiment, a graph Fourier transform can be performed on a directed Cartesian product graph as shown in FIG. 11C, and frequency analysis, filtering, and noise removal can be performed on directed Cartesian product graph signals as well as undirected Cartesian product graph signals.

[0175] At least some of the functions of the Fourier transform device 100 ( FIG. 3 ) in each of the above-described embodiments may be implemented by a computer. In this case, a program for implementing these functions may be recorded on a computer-readable recording medium, and the program may be loaded and executed by a computer system. Note that the term "computer system" as used herein includes hardware such as an operating system (OS) and peripheral devices. Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, read-only memories (ROMs), and CD-ROMs, as well as storage devices such as hard disks built into computer systems. Furthermore, the term "computer-readable recording medium" may also include media that dynamically store programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or telephone lines, or media that store programs for a fixed period of time, such as volatile memory within a computer system that serves as a server or client. The program may be designed to implement some of the above-described functions, or may be capable of implementing the above-described functions in combination with a program already stored in the computer system, or may be implemented using a programmable logic device such as an FPGA.

[0176] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention.

[0177] The present invention can be applied to a Fourier transform device, a Fourier transform method, a program, and the like that need to perform a graph Fourier transform on a directed graph.

[0178] 11: Undirected graph signal acquisition unit, 12: Undirected graph signal Fourier transform unit, 13: Directed graph signal acquisition unit, 14: Predetermined vector calculation unit, 15: Hadamard product calculation unit, 16: Directed graph signal Fourier transform unit, 17: Storage unit, 100: Fourier transform device

Claims

1. A Fourier transform device comprising: a directed graph signal acquisition unit that acquires a directed graph signal; an undirected graph signal Fourier transform unit that performs a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal; and a directed graph signal Fourier transform unit that performs a graph Fourier transform of the directed graph signal based on the directed graph signal acquired by the directed graph signal acquisition unit and a result of the graph Fourier transform performed by the undirected graph signal Fourier transform unit.

2. The Fourier transform device according to claim 1, wherein the directed graph signal acquisition unit acquires a directed path graph signal as the directed graph signal, the undirected graph signal Fourier transform unit performs a graph Fourier transform of an undirected path graph signal corresponding to the directed path graph signal, the Fourier transform device further comprises a Hadamard product calculation unit that calculates a Hadamard product between a predetermined vector and the directed path graph signal acquired by the directed graph signal acquisition unit, and the directed graph signal Fourier transform unit performs a graph Fourier transform of the directed path graph signal based on the Hadamard product calculated by the Hadamard product calculation unit and a result of the graph Fourier transform performed by the undirected graph signal Fourier transform unit.

3. The Fourier transform device according to claim 1, wherein the directed graph signal acquisition unit acquires a directed ring graph signal as the directed graph signal, the undirected graph signal Fourier transform unit performs a graph Fourier transform of an undirected ring graph signal corresponding to the directed ring graph signal, the Fourier transform device further comprises a Hadamard product calculation unit that calculates a Hadamard product between a predetermined vector and the directed ring graph signal acquired by the directed graph signal acquisition unit, and the directed graph signal Fourier transform unit performs a graph Fourier transform of the directed ring graph signal based on the Hadamard product calculated by the Hadamard product calculation unit and a result of the Fourier transform performed by the undirected graph signal Fourier transform unit.

4. The Fourier transform device according to claim 1, wherein the directed graph signal acquisition unit acquires a directed cartesian product graph signal as the directed graph signal, the undirected graph signal Fourier transform unit performs a graph Fourier transform of an undirected cartesian product graph signal corresponding to the directed cartesian product graph signal, the Fourier transform device further comprises a Hadamard product calculation unit that calculates a Hadamard product between a predetermined vector and the directed cartesian product graph signal acquired by the directed graph signal acquisition unit, and the directed graph signal Fourier transform unit performs a graph Fourier transform of the directed cartesian product graph signal based on the Hadamard product calculated by the Hadamard product calculation unit and a result of the Fourier transform performed by the undirected graph signal Fourier transform unit.

5. The Fourier transform device according to claim 2, wherein the Hadamard product calculation section uses a vector based on the following formula (A) as the predetermined vector: ... (A) 6. The Fourier transform device according to claim 1, wherein the undirected graph signal Fourier transform unit acquires the undirected graph signal based on the directed graph signal acquired by the directed graph signal acquisition unit, and performs a graph Fourier transform of the undirected graph signal.

7. A Fourier transform method comprising: acquiring a directed graph signal; performing a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal; and performing a graph Fourier transform of the directed graph signal based on the acquired directed graph signal and a result of performing the graph Fourier transform on the undirected graph signal.

8. A program for causing a computer of a Fourier transform device to acquire a directed graph signal, perform a graph Fourier transform of an undirected graph signal corresponding to the directed graph signal, and perform a graph Fourier transform of the directed graph signal based on the acquired directed graph signal and a result of performing the graph Fourier transform on the undirected graph signal.

Citation Information

Patent Citations

  • Graph analysis device, graph analysis method, and graph analysis program

    WO2021005805A1

  • Detection device, detection method, and detection program

    WO2022059107A1

  • Detection device, detection method, and detection program

    WO2022059108A1