A natural image segmentation method and system based on Zhang quantum space clustering

Abstract

The present invention belongs to the field of natural image segmentation, and specifically relates to a natural image segmentation method and system based on tensor subspace clustering. The natural image segmentation method based on tensor subspace clustering includes obtaining an original natural image; preprocessing the original natural image to obtain multiple super pixel blocks; extracting feature data from the super pixel blocks to obtain a feature data matrix; substituting the feature data matrix into a preset tensor subspace clustering model to calculate a representation coefficient matrix; according to the representation coefficient matrix, according to a preset formula, calculating an affinity matrix; importing the affinity matrix into a spectral clustering algorithm model to obtain a segmented foreground image and background image. The present invention can achieve that when using tensor subspace clustering for image segmentation, the features obtained by tensor subspace clustering are more robust, so that the accuracy of the foreground image and background image segmented from the natural image is effectively improved.

Description

[0001] The present invention belongs to the field of natural image segmentation, and in particular relates to a natural image segmentation method and system based on tensor subspace clustering. Background Art

[0002] Traditional clustering methods encounter the following two problems when clustering high-dimensional data sets: the presence of a large number of irrelevant attributes in high-dimensional data sets makes the possibility of clusters in all dimensions almost zero; the data in high-dimensional space is sparser than that in low-dimensional space, and it is a common phenomenon that the distances between data are almost equal. Because it is impossible to build clusters based on distance in high-dimensional space, subspace clustering (SC) has become an effective way to achieve clustering of high-dimensional data sets and is applied in the field of natural image segmentation.

[0003] At present, in the field of natural image segmentation, segmentation is usually performed based on certain features of natural images. However, natural images often suffer from data nonlinearity and complex noise pollution problems, and data points may not strictly follow the subspace structure, which affects the difficulty and accuracy of natural image segmentation. Summary of the invention

[0004] The technical problem to be solved by the present invention is that the existing natural image segmentation is difficult and has low precision.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a natural image segmentation method based on Zhang quantum space clustering, comprising:

[0006] S1, obtain the original natural image;

[0007] S2, preprocessing the original natural image to obtain multiple super-pixel blocks;

[0008] S3, extracting feature data from the super pixel block to obtain a feature data matrix;

[0009] S4, substituting the characteristic data matrix into a preset Zhang quantum space clustering model to calculate a representation coefficient matrix;

[0010] S5, calculating an affinity matrix according to the representation coefficient matrix and a preset formula;

[0011] S6, importing the affinity matrix into a preset spectral clustering algorithm model to obtain segmented foreground image and background image.

[0012] Furthermore, in S2, the preprocessing uses a linear iterative clustering method to segment the original natural image into super-pixel blocks.

[0013] Further, in S3, feature data is extracted from the super pixel block to obtain a feature data matrix, including:

[0014] For the superpixel block obtained by preprocessing, the color space in the superpixel block is divided into color intervals, each interval is a bin of a color histogram, and the color histogram is extracted to form a feature data matrix; wherein the color space is a chromaticity range corresponding to a chromaticity set formed according to the chromaticity corresponding to all pixels in the superpixel block.

[0015] Furthermore, in S4, the feature data matrix is ​​mapped into a high-dimensional Hilbert space using a kernel.

[0016] Furthermore, in S4, the Zhang quantum space clustering model is:

[0017]

[0018] stZ≥0,diag(Z)=0,Z=Z T ,Μ=θ+ω

[0019] Among them, Z is the coefficient matrix, I is the identity matrix, and M is composed of K i The damaged tensor is accumulated, and K i =φ i (X) T φ i (X) denotes the kernel matrix, θ denotes the recovered tensor, and ω denotes the noise error tensor.

[0020] Further, in S5, the preset formula is to substitute the representation coefficient matrix into the following formula:

[0021]

[0022] Among them, Z is the coefficient matrix and A is the affinity matrix.

[0023] Furthermore, in S5, the spectral clustering algorithm model cuts the image composed of data points in the affinity matrix so that the sum of edge weights between different subgraphs after cutting reaches the lowest value, while the sum of edge weights within the subgraph reaches the highest value.

[0024] A natural image segmentation system based on tensor quantum space clustering, comprising:

[0025] An information receiving unit, used for receiving an original natural image;

[0026] A storage unit, used to store the Zhang quantum space clustering model, the preset formula and the spectral clustering algorithm model;

[0027] A preprocessing unit, used for segmenting the original natural image into super-pixel blocks by using a linear iterative clustering method;

[0028] A feature extraction unit is used to divide the color space in the super pixel block sent by the preprocessing unit into color intervals, each interval is a bin of a color histogram, and extract the color histogram to form a feature data matrix;

[0029] A representation coefficient calculation unit, used for substituting the feature data matrix sent by the feature extraction unit into the Zhang quantum space clustering model to obtain a representation coefficient matrix;

[0030] An affinity calculation unit, used for substituting the representation coefficient matrix sent by the representation coefficient calculation unit into a preset formula to calculate an affinity matrix;

[0031] A segmentation unit, used for segmenting the affinity matrix sent by the affinity calculation unit into a foreground image and a background image by using a spectral clustering algorithm model;

[0032] The output unit is used to output the foreground image and the background image sent by the segmentation unit.

[0033] A network-side server, comprising:

[0034] at least one processor; and,

[0035] a memory communicatively connected to the at least one processor; wherein,

[0036] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any of the natural image segmentation methods based on tensor subspace clustering described above.

[0037] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the natural image segmentation method based on tensor subspace clustering.

[0038] The beneficial effects of the present invention are as follows: the present invention first uses linear iterative clustering to obtain multiple super pixel blocks; extracts feature data from the super pixel blocks by extracting color histograms to obtain a feature data matrix; substitutes the feature data matrix into a preset tensor subspace clustering model to calculate a representation coefficient matrix; calculates an affinity matrix according to the representation coefficient matrix and a preset formula; imports the affinity matrix into a spectral clustering algorithm model to obtain segmented foreground images and background images. A more effective low-rank constraint method is proposed in the tensor subspace clustering model: non-convex kernel tensor low-rank learning. Not only can the potential relationship between each kernel be fully explored, but also mapping can be performed to ensure that the mapped feature space has multiple low-dimensional subspace structures. The present invention can obtain more robust features by using tensor subspace clustering for image segmentation, so that the accuracy of the foreground image and background image segmented from the natural image is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0040] Figure 1 is a flow chart of a natural image segmentation method based on Zhang quantum space clustering according to a first embodiment of the present invention;

[0041] Figure 2 is a schematic structural diagram of a natural image segmentation system based on Zhang quantum space clustering according to a second embodiment of the present invention;

[0042] Figure 3 It is a structural diagram of a network-side server provided according to a third embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] The first embodiment of the present invention relates to a natural image segmentation method based on Zhang quantum space clustering. In this embodiment, the original natural image is firstly subjected to a linear iterative clustering method to obtain a plurality of super pixel blocks; the super pixel blocks are subjected to feature data extraction by extracting color histograms to obtain a feature data matrix; the feature data matrix is ​​substituted into a preset Zhang quantum space clustering model to calculate a representation coefficient matrix; according to the representation coefficient matrix, an affinity matrix is ​​calculated according to a preset formula; the affinity matrix is ​​introduced into a spectral clustering algorithm model to obtain a segmented foreground image and background image. The present invention can achieve that when using Zhang quantum space clustering for image segmentation, the features obtained by Zhang quantum space clustering are more robust, so that the accuracy of segmentation is effectively improved.

[0045] The following is a detailed description of the implementation details of the natural image segmentation method based on tensor subspace clustering in this embodiment. The following content is only for the convenience of understanding the implementation details, which is not necessary for the implementation of this solution. The specific process of this embodiment is as follows: Figure 1 As shown, this implementation is applied to the server on the network side.

[0046] Step S1, obtaining an original natural image.

[0047] Specifically, the original natural images are imported into the image dataset through MATLAB.

[0048] Step S2: preprocess the original natural image to obtain multiple super pixel blocks.

[0049] Specifically, linear iterative clustering is used to perform superpixel segmentation on the original image. The specific implementation steps are as follows:

[0050] Initialize the seed points (cluster centers) and evenly distribute some seed points in the image according to the set number of superpixel blocks. Assuming that the image contains a total of N pixels, which are pre-divided into K superpixels, then the size of each superpixel is N / K, and the distance or step size between adjacent seed points can be approximately expressed as S = sqrt(N / K).

[0051] Reselect the seed point in the n×n neighborhood of the seed point (usually n=3). First, calculate the gradient values ​​of all pixels in the neighborhood; then move the seed point to the place with the smallest gradient in the neighborhood. The purpose of this is to avoid the seed point falling on the contour boundary with a large gradient, so as not to affect the subsequent clustering effect.

[0052] Assign a class label to each pixel in the neighborhood around each seed point (i.e., which cluster center it belongs to). Unlike the standard k-means search in the entire image, SLIC limits the search range to 2S×2S, which can accelerate the convergence of the algorithm. The expected superpixel size is S×S, but the search range is 2S×2S.

[0053] Distance measurement. Including color distance and spatial distance. For each searched pixel, calculate the distance between it and the seed point. The distance calculation method is as follows:

[0054]

[0055]

[0056]

[0057] Among them, d c Represents the color distance, d s Represents the spatial distance, N s is the maximum spatial distance within the class, defined as N s = S = sqrt(N / K), applicable to each cluster. The maximum color distance N c It varies from image to image and from cluster to cluster. Since each pixel is searched by multiple seed points, each pixel has a distance from the surrounding seed points, and the seed point corresponding to the minimum value is taken as the cluster center of the pixel.

[0058] The above steps are iterated and optimized until the error converges, that is, the cluster center of each pixel no longer changes.

[0059] Step S3, extracting feature data from the superpixel block to obtain a feature data matrix.

[0060] Specifically, a color histogram is extracted from the superpixel block obtained by preprocessing, specifically: the chromaticity range corresponding to the chromaticity set formed by the chromaticity corresponding to all pixels in the superpixel block is divided into several small color intervals, each interval is a bin of the color histogram, and a color histogram is obtained.

[0061] This process is called color quantization. Color quantization uses a clustering method. By calculating the number of pixels whose colors fall within each small interval, a color histogram can be obtained. The clustering algorithm takes into account the distribution of the image color features in the entire space, thereby avoiding the situation where the number of pixels in some bins is very sparse, making quantization more effective. In addition, if the image is in RGB format and the histogram is in LAB space, a look-up table from the quantized RGB space to the quantized LAB space is pre-established to speed up the histogram calculation process. By extracting the chromaticity of each pixel in the color histogram, a feature data matrix is ​​obtained.

[0062] Step S4, substituting the characteristic data matrix into a preset Zhang subspace clustering model to calculate a representation coefficient matrix.

[0063] Specifically, in order to solve the low-dimensional linear inseparable problem, the feature data matrix is ​​mapped to the high-dimensional Hilbert space using the kernel; at the same time, because the high-dimensional feature space has the problem of feature non-flatness, the multi-kernel learning method is used to divide X into i kernel representations (i<12), that is, K i =φ i (X) T φ i (X), one kernel representation corresponds to one kernel tensor. However, due to noise interference, K i The accumulated kernel tensor M is damaged.

[0064] The damaged kernel tensor M needs to be restored and then input into the objective function. In order to keep the coefficient representation matrix Z block diagonal in the underlying subspace, a block diagonalization constraint is imposed on the objective function. The objective function is iteratively optimized and convergence is determined by setting convergence conditions. When convergence occurs, the coefficient matrix representing the low-rank subspace is output.

[0065] The Zhang Quantum Space Clustering Model uses the tensor low-rank representation combined with the non-convex characteristics of the Schatten-p norm to propose a more effective low-rank constraint method: non-convex kernel tensor low-rank learning. This method can not only fully explore the potential relationship between each kernel, but also map to ensure that the mapped feature space has multiple low-dimensional subspace structures. In addition, this Zhang Quantum Space Clustering Model also introduces a block diagonal regularizer (BDR)||·|| F To encourage the coefficient matrix Z to maintain the block diagonal property when the underlying subspaces are independent, thereby improving the effect of subspace clustering.

[0066] The formula of Zhang quantum space clustering model is as follows:

[0067]

[0068] stZ≥0,diag(Z)=0,Z=Z T ,Μ=θ+ω

[0069] Among them, Z is the coefficient matrix, I is the identity matrix, and M is composed of K i The damaged tensor is accumulated, and K i =φ i (X) T φ i (X) represents the kernel matrix, θ represents the restored tensor, ω represents the noise error tensor, T represents the matrix transpose, i represents a positive integer, and r represents a positive integer.

[0070] Step S5: Calculate the affinity matrix according to the representation coefficient matrix and a preset formula.

[0071] Specifically, substitute the obtained representation coefficient matrix into the following formula:

[0072]

[0073] Among them, Z is the coefficient matrix and A is the affinity matrix.

[0074] S6, importing the affinity matrix into the spectral clustering algorithm model to obtain the segmented foreground image and background image.

[0075] Specifically, the step of inputting the affinity matrix into the spectral clustering algorithm model is to first treat all the data as points in space, which can be connected by edges. The edge weight value between two points that are far away is low, while the edge weight value between two points that are close is high. By cutting the graph composed of all data points, the sum of the edge weights between different subgraphs after cutting is as low as possible, and the sum of the edge weights within the subgraph is as high as possible, so as to perform clustering and segmentation, and obtain the segmented foreground image and background image.

[0076] In the first embodiment of the present invention, a more effective low-rank constraint method is proposed by combining the non-convex characteristics of the Schatten-p norm with the low-rank representation of the tensor: non-convex kernel tensor low-rank learning. Not only can the potential relationship between the kernels be fully explored, but also the mapped feature space can be mapped to ensure that it has multiple low-dimensional subspace structures. In addition, the block diagonal regularizer (BDR) is introduced. F To encourage the coefficient matrix Z to maintain the block diagonal property when the underlying subspace is independent, so as to improve the effect of subspace clustering. The present invention can achieve that when using tensor subspace clustering for image segmentation, the features obtained by tensor subspace clustering are more robust, so that the accuracy of segmentation is effectively improved.

[0077] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this patent; adding insignificant modifications to the process or introducing insignificant designs without changing the core design of its algorithm and process are all within the scope of protection of this patent.

[0078] like Figure 2 As shown, the second embodiment of the present invention relates to a natural image segmentation system based on tensor subspace clustering, including: an information receiving unit 201, a storage unit 202, a preprocessing unit 203, a feature extraction unit 204, a representation coefficient calculation unit 205, an affinity calculation unit 206, a segmentation unit 207, and an output unit 208.

[0079] Specifically, the information receiving unit 201 is used to receive the imported original natural image; the storage unit 202 is used to store the Zhang quantum space clustering model, the preset formula and the spectral clustering algorithm model; the preprocessing unit 203 is used to segment the original natural image into super pixel blocks by using the linear iterative clustering method; the feature extraction unit 204 is used to divide the color space in the super pixel block sent by the preprocessing unit 203 into color intervals, each interval is a bin of the color histogram, and the color histogram is extracted to form a feature data matrix; the representation coefficient calculation unit 205 is used to substitute the feature data matrix sent by the feature extraction unit 204 into the Zhang quantum space clustering model to obtain the representation coefficient matrix; the affinity calculation unit 206 is used to substitute the representation coefficient matrix sent by the representation coefficient calculation unit 205 into the preset formula to calculate the affinity matrix; the segmentation unit 207 is used to segment the affinity matrix sent by the affinity calculation unit 206 into a foreground image and a background image by using the spectral clustering algorithm model; the output unit 208 is used to output the foreground image and the background image sent by the segmentation unit 207.

[0080] It is not difficult to find that this embodiment is a system embodiment corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the first embodiment.

[0081] Compared with the prior art, the present embodiment provides a natural image segmentation system of Zhang quantum space clustering. During the working process, the Zhang quantum space clustering model, the preset formula and the spectral clustering algorithm model are stored in advance by the storage unit 202. The imported original natural image is received by the information receiving unit 201. The acquired original natural image is input into the preprocessing unit 203, and the simple linear iterative clustering (SLIC) method is used to segment the super pixel blocks to form super pixel blocks. The super pixel blocks are input into the feature extraction unit 204 to extract the color histogram to form a feature data matrix. The feature coefficient matrix is ​​then input into the representation coefficient calculation unit 205 and substituted into the Zhang quantum space clustering model to obtain the representation coefficient matrix. The representation coefficient matrix is ​​input into the affinity calculation unit 206 and the affinity matrix is ​​calculated by the preset formula. Finally, the affinity matrix is ​​input into the segmentation unit 207 and the spectral clustering algorithm model is used to segment it into a foreground image and a background image. The segmented foreground image and background image are output through the output unit 208.

[0082] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of the present invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by the present invention, but this does not mean that there are no other units in this embodiment.

[0083] A third embodiment of the present invention relates to a network side server, such as Figure 3 As shown, it includes at least one processor 301; and a memory 302 that is communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions that can be executed by the at least one processor 301, and the instructions are executed by the at least one processor 301 so that the at least one processor 301 can execute the above-mentioned data processing method.

[0084] The memory 302 and the processor 301 are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors 301 and the memory 302 together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor 301 is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor 301.

[0085] The processor 301 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 302 can be used to store data used by the processor 301 when performing operations.

[0086] The fifth embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the automatic clothing image segmentation mechanism based on deep learning in the first embodiment is implemented.

[0087] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0088] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the scope of the present invention through the above description. The technical scope of this invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A natural image segmentation method based on Zhang quantum space clustering, characterized in that: include: S1, obtain the original natural image; S2, preprocessing the original natural image to obtain multiple super-pixel blocks; S3, extracting feature data from the super pixel block to obtain a feature data matrix; S4, substituting the feature data matrix into a preset Zhang quantum space clustering model to calculate a representation coefficient matrix, including: mapping the feature data matrix into a high-dimensional Hilbert space using a kernel, dividing X into i kernel representations (i<12) by a multi-kernel learning method, i.e., K i =φ i (X) T φ i (X), one kernel representation corresponds to one kernel tensor; the damaged kernel tensor M is restored and then input into the objective function, a block diagonalization constraint is imposed on the objective function, the objective function is iteratively optimized, and convergence is determined by setting convergence conditions. When convergence occurs, the representation coefficient matrix of the low-rank subspace is output; The Zhang quantum space clustering model is: stZ≥0,diag(Z)=0,Z=Z T ,Μ=θ+ω Among them, Z is the coefficient matrix, I is the identity matrix, and M is composed of K i The damaged tensor is accumulated, and K i =φ i (X) T φ i (X) represents the kernel matrix, θ represents the restored tensor, and ω represents the noise error tensor; S5, calculating an affinity matrix according to the representation coefficient matrix and a preset formula; S6, importing the affinity matrix into a preset spectral clustering algorithm model to obtain segmented foreground image and background image.

2. The natural image segmentation method based on Zhang quantum space clustering according to claim 1, characterized in that: In S2, the preprocessing uses a linear iterative clustering method to segment the original natural image into super-pixel blocks.

3. The natural image segmentation method based on Zhang quantum space clustering according to claim 2 is characterized in that: In S3, feature data is extracted from the super pixel block to obtain a feature data matrix, including: For the superpixel block obtained by preprocessing, the color space in the superpixel block is divided into color intervals, each interval is a bin of a color histogram, and the color histogram is extracted to form a feature data matrix; wherein the color space is a chromaticity range corresponding to a chromaticity set formed according to the chromaticity corresponding to all pixels in the superpixel block.

4. The natural image segmentation method based on Zhang quantum space clustering according to claim 1, characterized in that: In S5, the preset formula is to substitute the representation coefficient matrix into the following formula: Among them, Z is the coefficient matrix and A is the affinity matrix.

5. The natural image segmentation method based on Zhang quantum space clustering according to claim 1, characterized in that: In S5, the spectral clustering algorithm model cuts the image composed of data points in the affinity matrix so that the sum of edge weights between different subgraphs after cutting reaches the lowest value, while the sum of edge weights within the subgraph reaches the highest value.

6. A natural image segmentation system based on Zhang quantum space clustering, characterized in that: include: An information receiving unit, used for receiving an original natural image; A storage unit, used to store the Zhang quantum space clustering model, the preset formula and the spectral clustering algorithm model; A preprocessing unit, used for segmenting the original natural image into super-pixel blocks by using a linear iterative clustering method; A feature extraction unit is used to divide the color space in the super pixel block sent by the preprocessing unit into color intervals, each interval is a bin of a color histogram, and extract the color histogram to form a feature data matrix; The representation coefficient calculation unit is used to substitute the feature data matrix sent by the feature extraction unit into the Zhang quantum space clustering model to obtain the representation coefficient matrix, which specifically includes: mapping the feature data matrix to the high-dimensional Hilbert space using the kernel, dividing X into i kernel representations (i<12) through the multi-kernel learning method, that is, K i =φ i (X) T φ i (X), one kernel representation corresponds to one kernel tensor; the damaged kernel tensor M is restored and then input into the objective function, a block diagonalization constraint is imposed on the objective function, the objective function is iteratively optimized, and convergence is determined by setting convergence conditions. When convergence occurs, the representation coefficient matrix of the low-rank subspace is output; The Zhang quantum space clustering model is: stZ≥0,diag(Z)=0,Z=Z T ,Μ=θ+ω Among them, Z is the coefficient matrix, I is the identity matrix, and M is composed of K i The damaged tensor is accumulated, and K i =φ i (X) T φ i (X) represents the kernel matrix, θ represents the restored tensor, and ω represents the noise error tensor; An affinity calculation unit, used for substituting the representation coefficient matrix sent by the representation coefficient calculation unit into a preset formula to calculate an affinity matrix; A segmentation unit, used for segmenting the affinity matrix sent by the affinity calculation unit into a foreground image and a background image by using a spectral clustering algorithm model; The output unit is used to output the foreground image and the background image sent by the segmentation unit.

7. A network side server, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the natural image segmentation method based on tensor subspace clustering as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the natural image segmentation method based on tensor subspace clustering according to any one of claims 1 to 5 is implemented.

Images