Late fusion object image clustering method, device and computer equipment

Abstract

The application relates to a late-stage fused object image clustering method, device and computer equipment. The method comprises the following steps: in response to an image clustering request, acquiring object image data to be clustered; determining a basic partition matrix, an initial rotation disturbance matrix and an initial weight coefficient of the object image data under different views, and an initial consistent clustering partition matrix corresponding to all views; constructing an object image clustering objective function according to the basic partition matrix, the initial rotation disturbance matrix, the initial weight coefficient and the initial consistent clustering partition matrix; alternately optimizing the parameters to be optimized in the object image clustering objective function, and obtaining a target consistent clustering partition matrix of a basic partition matrix fused with each view when the function value of the object image clustering objective function is a preset function value; and performing clustering processing on the target consistent clustering partition matrix to obtain a clustering result of the object image data. The method can improve the clustering accuracy of images.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus and computer device for post-fusion object image clustering. Background Technology

[0002] With the development of multimedia technology, data is experiencing explosive growth, and massive amounts of data have become a prominent characteristic of today's data. We can easily obtain feature information from multiple different views of a data sample. Data with feature information from multiple views is called multi-view data, and the method for clustering multi-view data is called multi-view clustering method. Since a data sample possesses feature information from multiple views, effectively fusing information from multiple views is a key issue in multi-view clustering methods.

[0003] However, existing multi-view clustering directly fuses the feature information and representations of different views to obtain a unified representation, and then applies a clustering method to the unified representation to obtain the final clustering result, resulting in low accuracy of image clustering results. Summary of the Invention

[0004] Therefore, it is necessary to provide a post-fusion object image clustering method, apparatus, computer device, computer-readable storage medium, and computer program product that can improve the clustering accuracy of images, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a post-processing fusion method for object image clustering. The method includes:

[0006] In response to an image clustering request, obtain the image data of the objects to be clustered;

[0007] Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views;

[0008] Based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix, a clustering objective function for object images is constructed.

[0009] The parameters to be optimized in the objective function of the object image clustering are alternately optimized. When the function value of the objective function of the object image clustering is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0010] Clustering is performed on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0011] In one embodiment, constructing the object image clustering objective function based on the base partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix includes:

[0012] Based on the initial consistent clustering partition matrix and the initial rotation perturbation matrix of each view, the corresponding basic partition matrix is ​​rotated to obtain the rotated basic partition matrix of each view;

[0013] The object image clustering objective function is obtained by weighting and fusing the corresponding rotated basic partitioning matrix according to the initial weight coefficients of each view, and determining the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix as parameters to be optimized.

[0014] In one embodiment, the method involves alternately optimizing the parameters to be optimized in the object image clustering objective function, and obtaining a consistent clustering partition matrix when the function value of the object image clustering objective function is a preset function value. The method further includes:

[0015] Obtain optimization strategy data for the parameters to be optimized in the objective function of the object image clustering;

[0016] Based on the optimization strategy data, the parameters to be optimized in the object image clustering objective function are optimized. If the function value of the object image clustering objective function after optimization is not the preset function value, the step of optimizing the parameters to be optimized in the object image clustering objective function based on the optimization strategy data is continued.

[0017] When the function value of the target function for clustering the object image is a preset function value, a target-consistent clustering partition matrix is ​​obtained.

[0018] In one embodiment, the optimization process for the parameters to be optimized in the object image clustering objective function based on the optimization strategy data includes:

[0019] Based on the initial rotation perturbation matrix and the object image clustering objective function, the initial weight coefficients are minimized to obtain intermediate weight coefficients, and the initial rotation perturbation matrix is ​​maximized to obtain an intermediate consistent clustering partition matrix.

[0020] Based on the intermediate weight coefficients, the intermediate consistent clustering partition matrix, and the object image clustering objective function, the initial rotation perturbation matrix is ​​subjected to maximum optimization processing to obtain the intermediate rotation perturbation matrix.

[0021] In one embodiment, the step of performing minimum optimization on the initial weight coefficients based on the initial rotation perturbation matrix and the object image clustering objective function to obtain intermediate weight coefficients, and performing maximum optimization on the initial rotation perturbation matrix to obtain an intermediate consistent clustering partition matrix, includes:

[0022] While keeping the initial rotation perturbation matrix unchanged, the objective function for clustering the object image is processed to obtain a first objective optimization function for optimizing the initial consistent clustering partition matrix and the initial weight coefficients;

[0023] Based on the first objective optimization function, calculate the simplified gradient of the initial weight coefficients;

[0024] If the sum of the initial weight coefficients for each view is a preset value, gradient descent is performed on the object image clustering objective function to update the simplified gradient;

[0025] When the objective function for clustering the object image has its minimum extremum in the gradient descent direction, the minimum weight coefficient and the optimal consistent clustering partition matrix are obtained. The minimum weight coefficient is determined as the intermediate weight coefficient, and the maximum consistent clustering partition matrix is ​​determined as the intermediate consistent clustering partition matrix.

[0026] In one embodiment, the step of performing maximum optimization processing on the initial rotation perturbation matrix based on the intermediate weight coefficients, the intermediate consistent clustering partition matrix, and the object image clustering objective function to obtain the intermediate rotation perturbation matrix includes:

[0027] The initial weight coefficients are updated based on the intermediate weight coefficients, and the initial consistent clustering partition matrix is ​​updated based on the intermediate consistent clustering partition matrix. The object image clustering objective function is then processed to obtain a second objective optimization function for optimizing the initial rotation perturbation matrix.

[0028] Based on the second objective optimization function, the initial rotation perturbation matrix is ​​maximized using the singular value decomposition method to obtain the intermediate rotation perturbation matrix.

[0029] Secondly, this application also provides a post-fusion object image clustering apparatus. The apparatus includes:

[0030] The image data acquisition module is used to acquire image data of objects to be clustered in response to image clustering requests;

[0031] The data determination module is used to determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views;

[0032] The function construction module is used to construct an object image clustering objective function based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix;

[0033] The optimization processing module is used to alternately optimize the parameters to be optimized in the object image clustering objective function. When the function value of the object image clustering objective function is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0034] The clustering module is used to perform clustering processing on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0035] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0036] In response to an image clustering request, obtain the image data of the objects to be clustered;

[0037] Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views;

[0038] Based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix, a clustering objective function for object images is constructed.

[0039] The parameters to be optimized in the objective function of the object image clustering are alternately optimized. When the function value of the objective function of the object image clustering is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0040] Clustering is performed on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0041] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0042] In response to an image clustering request, obtain the image data of the objects to be clustered;

[0043] Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views;

[0044] Based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix, a clustering objective function for object images is constructed.

[0045] The parameters to be optimized in the objective function of the object image clustering are alternately optimized. When the function value of the objective function of the object image clustering is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0046] Clustering is performed on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0047] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0048] In response to an image clustering request, obtain the image data of the objects to be clustered;

[0049] Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views;

[0050] Based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix, a clustering objective function for object images is constructed.

[0051] The parameters to be optimized in the objective function of the object image clustering are alternately optimized. When the function value of the objective function of the object image clustering is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0052] Clustering is performed on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0053] The aforementioned post-processing fusion object image clustering method, apparatus, computer equipment, storage medium, and computer program product, in response to an image clustering request, acquires the object image data to be clustered; constructs an object image clustering objective function based on the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views; by alternately optimizing the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partitioning matrix in the object image clustering objective function, the object image clustering objective function is prevented from converging to a local optimum, resulting in a target consistent clustering partitioning matrix that fuses the basic partitioning matrix of each view; and by performing clustering processing on the target consistent clustering partitioning matrix, the clustering result of the object image data is obtained, thus improving the accuracy of image clustering. Attached Figure Description

[0054] Figure 1 This is an application environment diagram of a post-fusion object image clustering method in one embodiment;

[0055] Figure 2 This is a flowchart illustrating a late-stage fusion object image clustering method in one embodiment;

[0056] Figure 3 This is a flowchart illustrating a method for constructing an object image clustering objective function in one embodiment;

[0057] Figure 4 This is a flowchart illustrating an object image clustering objective function optimization method in one embodiment;

[0058] Figure 5 This is a flowchart illustrating the optimization steps for weight coefficients and the consistent clustering partition matrix in one embodiment;

[0059] Figure 6 This is a flowchart illustrating the rotational perturbation matrix optimization steps in one embodiment;

[0060] Figure 7 This is a flowchart illustrating the late-stage fusion of object image clustering methods in another embodiment;

[0061] Figure 8 This is a schematic diagram illustrating the change of the object image clustering objective function in each iteration of one embodiment;

[0062] Figure 9 This is a performance comparison chart of algorithms on the Caltech dataset in one example;

[0063] Figure 10This is a structural block diagram of a late-stage fusion object image clustering device in one embodiment;

[0064] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0066] The post-fusion object image clustering method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Responding to an image clustering request, the terminal retrieves the object image data to be clustered from the server; determines the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views; constructs an object image clustering objective function based on the basic partitioning matrix, initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partitioning matrix; alternately optimizes the parameters to be optimized in the object image clustering objective function, and when the function value of the object image clustering objective function is a preset function value, a target consistent clustering partitioning matrix that fuses the basic partitioning matrices of each view is obtained; the parameters to be optimized include the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partitioning matrix; and performs clustering processing on the target consistent clustering partitioning matrix to obtain the clustering result of the object image data. The terminal 102 can be, but is not limited to, various personal computers, laptops, IoT devices, and portable wearable devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0067] In one embodiment, such as Figure 2 As shown, a post-fusion object image clustering method is provided, which is then applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0068] Step 202: In response to the image clustering request, obtain the image data of the objects to be clustered.

[0069] Image clustering requests can be tailored to different application scenarios. For example, in logistics, to prevent damage to goods during sorting and ensure the safety of sorting, image clustering is needed to detect abnormal behavior. Similarly, in security scenarios, facial images need to be clustered for face recognition. And in autonomous driving scenarios, to ensure driving safety, images captured during driving need to be clustered to identify obstacles and plan appropriate paths.

[0070] The object image data can include benchmark image datasets and large-scale datasets. The number of benchmark image datasets and large-scale datasets can be determined according to actual needs. For example, the object image data may include 6 benchmark image datasets and 6 large-scale image datasets. The acquired object image data may include different objects, such as a chessboard, different kinds of animals, etc.

[0071] Step 204: Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views.

[0072] The basic partitioning matrix for each view can be obtained by processing the object image data using a kernel function. The kernel matrix corresponding to each view is then obtained by performing kernel K-means clustering on the kernel matrix and extracting the eigenvectors corresponding to the top K eigenvalues ​​to determine the basic partitioning matrix of the object image data under different views.

[0073] Specifically, based on multiple views of the object image data, corresponding kernel matrices are generated using kernel functions. K-means clustering is performed on each kernel matrix to extract the eigenvectors corresponding to the top K eigenvalues. The basic partitioning matrix of the object image data under different views is determined. The rotation perturbation matrix and weight coefficients of each view, as well as the consistent clustering partitioning matrix corresponding to all views, are initialized to obtain the corresponding initial rotation perturbation matrix and initial weight coefficients, as well as the initial consistent clustering partitioning matrix corresponding to all views.

[0074] Step 206: Construct the object image clustering objective function based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix.

[0075] Understandably, traditional clustering methods incur significant time and space overhead and computational complexity. This leads to low image clustering accuracy when dealing with large datasets, failing to meet the demands of practical applications. Therefore, in the clustering process, an object image clustering objective function is constructed based on the fundamental partitioning matrix, initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partitioning matrix for each view.

[0076] The constructed object image clustering objective function can be expressed as:

[0077]

[0078] H is the basic partitioning matrix for each view; H is the initial consistent clustering partitioning matrix and belongs to It is the initial rotation perturbation matrix corresponding to each view, used to align H. p The cluster correspondence between H and H; γ is the initial weight coefficient, representing the weighting of each view, and belongs to

[0079] Step 208: Alternately optimize the parameters to be optimized in the object image clustering objective function. When the function value of the object image clustering objective function is a preset function value, the target consistent clustering partition matrix that fuses the basic partition matrix of each view is obtained.

[0080] The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix. Alternating optimization involves sequentially optimizing each parameter in a set order. This alternating optimization process allows the object image clustering objective function to converge to the global optimum.

[0081] Taking the parameters to be optimized, including the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partition matrix, as an example, the first round of parameters to be optimized includes the initial rotation perturbation matrix and initial weight coefficients, and the second round of parameters to be optimized includes the initial consistent clustering partition matrix. The first round of parameters to be optimized and the second round of parameters to be optimized are optimized alternately.

[0082] The objective function for object image clustering is a preset function value, meaning that the objective function for object image clustering converges to the preset function value. At this point, the optimal rotation perturbation matrix, weight coefficients, and consistent clustering partition matrix are obtained.

[0083] Step 210: Perform clustering processing on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0084] Clustering can be, but is not limited to, K-means clustering.

[0085] Specifically, K-means clustering is performed on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0086] In the aforementioned post-fusion object image clustering method, in response to an image clustering request, the object image data to be clustered is acquired. Based on the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views, an object image clustering objective function is constructed. By alternately optimizing the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partitioning matrix in the object image clustering objective function, the object image clustering objective function is prevented from converging to a local optimum. The target consistent clustering partitioning matrix that fuses the basic partitioning matrix of each view is obtained. By performing clustering processing on the target consistent clustering partitioning matrix, the clustering result of the object image data is obtained, thus improving the accuracy of image clustering.

[0087] In one embodiment, such as Figure 3 As shown, a method for constructing an objective function for object image clustering is provided, including:

[0088] Step 302: Rotate the corresponding basic partition matrix based on the initial consistent clustering partition matrix and the initial rotation perturbation matrix of each view to obtain the rotated basic partition matrix of each view.

[0089] To ensure the effectiveness and interpretability of clustering, the basic partitioning matrix and the corresponding rotation perturbation matrix are always kept orthogonal.

[0090] Step 304: Perform weighted fusion of the corresponding rotated basic partition matrix according to the initial weight coefficients of each view, and determine the initial rotation perturbation matrix, initial weight coefficients and initial consistent clustering partition matrix as parameters to be optimized, so as to obtain the object image clustering objective function.

[0091] The above-mentioned method for constructing the objective function for object image clustering can solve the alignment problem of non-corresponding clusters between different views due to the unsupervised nature of clustering tasks by rotating the clustering partition matrix of each view, and reduce the time and space computational complexity by weighting and fusing the new clustering partition matrices corresponding to each view. Furthermore, by determining the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partition matrix as parameters to be optimized, the objective function for object image clustering can be constructed, which can meet the requirements of clustering processing of large-scale datasets and improve the accuracy of image clustering.

[0092] In one embodiment, such as Figure 4 As shown, a method for optimizing the objective function of object image clustering is provided, and this method is applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0093] Step 402: Obtain optimization strategy data for the parameters to be optimized in the objective function of object image clustering.

[0094] The optimization strategy data includes: First, while keeping the rotation perturbation matrix unchanged, optimize the weight coefficients and consistent clustering partition matrix of each view to obtain the optimized weight coefficients and consistent clustering partition matrix of each view; Second, based on the optimized weight coefficients and consistent clustering partition matrix of each view, optimize the rotation perturbation matrix of each view. If the function value of the object image clustering objective function does not converge, then the first and second steps are executed repeatedly, that is, the first and second steps are performed alternately until the function value of the object image clustering objective function converges, and the optimization ends.

[0095] Step 404: Based on the optimization strategy data, optimize the parameters to be optimized in the object image clustering objective function. If the function value of the optimized object image clustering objective function is not the preset function value, continue to execute the step of optimizing the parameters to be optimized in the object image clustering objective function based on the optimization strategy data.

[0096] The optimization of the parameters to be optimized in the object image clustering objective function can be divided into optimization of the weight coefficients and the consistent clustering partition matrix, and optimization of the rotation perturbation matrix. That is, based on the initial rotation perturbation matrix and the object image clustering objective function, the initial weight coefficients are minimized to obtain intermediate weight coefficients, and the initial rotation perturbation matrix is ​​maximized to obtain an intermediate consistent clustering partition matrix. In one embodiment, such as... Figure 5 As shown, the optimization of the weight coefficients and the consistent clustering partition matrix includes the following steps:

[0097] Step 502: While keeping the initial rotation perturbation matrix unchanged, process the object image clustering objective function to obtain the first objective optimization function used to optimize the initial consistent clustering partition matrix and the initial weight coefficients.

[0098] Step 504: Calculate the simplified gradient of the initial weight coefficients based on the first objective optimization function.

[0099] Step 506: With the sum of the initial weight coefficients of each view being a preset value, perform gradient descent on the object image clustering objective function to update the simplified gradient.

[0100] Step 508: When the extremum of the objective function for clustering the object image is minimized in the gradient descent direction, the minimum weight coefficients and the optimal consistent clustering partition matrix are obtained. The minimum weight coefficients are determined as intermediate weight coefficients, and the optimal consistent clustering partition matrix is ​​determined as the intermediate consistent clustering partition matrix.

[0101] Wherein, the initial rotation perturbation matrix is ​​maintained. While keeping them unchanged, optimize the initial weight coefficients and the initial consistent clustering partition matrix (γ,H):

[0102] At this point, the objective expression for optimizing the weight coefficients and the clustering partition matrix (γ,H) can be written in the following form:

[0103]

[0104] This problem is a second-order mini-maximum optimization problem, which can be optimized using a simplified gradient descent algorithm, ensuring that the sum of the weight coefficients is always 1 during optimization. First, this problem can be equivalently rewritten as a maximization problem of an optimal value function, as shown in the first objective function below:

[0105]

[0106]

[0107] At this point, for any given weight coefficient γ, an optimal value of H determined by γ can be obtained. Therefore, by calculating the reduced gradient of the weight coefficient γ, while constraining the sum of the weight coefficients to 1, we perform gradient descent on the entire object image clustering objective function, thereby ensuring the global monotonicity of the objective function. The reduced gradient of the weight coefficient γ is calculated as follows:

[0108]

[0109]

[0110] Let represent the reduced gradient of the p-th component, and u be the index of the component with the largest weight among the selected weight coefficients. After calculating the gradient, we automatically select the step size according to the Armijo criterion and continuously update the weight coefficients γ using the gradient descent method to achieve the optimal value. That is, when the weight coefficients γ are minimized and the function value of the object image clustering objective function is maximized, the optimal consistent clustering partition matrix can be determined.

[0111] Based on intermediate weight coefficients, an intermediate consistent clustering partition matrix, and an object image clustering objective function, the initial rotation perturbation matrix is ​​optimized to obtain an intermediate rotation perturbation matrix. In one embodiment, such as... Figure 6 As shown, optimizing the rotational perturbation matrix includes the following steps:

[0112] Step 602: Update the initial weight coefficients based on the intermediate weight coefficients and update the initial consistent clustering partition matrix based on the intermediate consistent clustering partition matrix. Process the object image clustering objective function to obtain the second objective optimization function used to optimize the initial rotation perturbation matrix.

[0113] Step 604: Based on the second objective optimization function, the initial rotation perturbation matrix is ​​maximized using the singular value decomposition method to obtain the intermediate rotation perturbation matrix.

[0114] Specifically, after the initial optimization of the initial weight coefficients and the initial consistent clustering partition matrix, the intermediate weight coefficients are obtained to update the initial weight coefficients. At this time, while keeping the intermediate weight coefficients and the intermediate consistent clustering partition matrix (γ,H) unchanged, the initial rotation perturbation matrix is ​​adjusted. Optimize:

[0115] At this point, regarding the rotational perturbation matrix The optimization objective can be written in the form of the second objective optimization function as follows:

[0116]

[0117]

[0118] At this point, the optimization process can be equivalent to solving m independent subproblems. Since this step does not directly affect different views, for each view's W... p The problems can be solved separately, and the p-th subproblem can be written as:

[0119]

[0120] At this point, we use the classic singular value decomposition (SVD) method for optimization. Under the condition that the function value of the object image clustering objective function is maximized, and other parameters are fixed, we obtain the solution of the optimal rotation perturbation matrix.

[0121] Step 406: When the loss function of the object image clustering objective function is a preset function value, the target consistent clustering partition matrix is ​​obtained.

[0122] The aforementioned object image clustering objective function optimization method employs two steps: first, minimizing the initial weight coefficients based on the initial rotation perturbation matrix and the object image clustering objective function to obtain intermediate weight coefficients; and second, maximizing the initial rotation perturbation matrix to obtain an intermediate consistent clustering partition matrix. These two steps are performed alternately, ensuring that the sum of the weight coefficients is always 1 during optimization, thus determining the global optimal solution to the mini-maximum optimization problem. Furthermore, the optimization of the rotation perturbation matrix yields the optimal rotation perturbation matrix given other parameters, thereby improving the accuracy and robustness of clustering while maintaining high efficiency and low computational complexity.

[0123] In another embodiment, such as Figure 7 As shown, a post-fusion object image clustering method is provided, which is then applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0124] Step 702: In response to the image clustering request, obtain the image data of the objects to be clustered.

[0125] Among them, the object image data can be multi-core standard data.

[0126] Step 704: Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views.

[0127] Step 706: Construct the object image clustering objective function based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix.

[0128] Step 708: While keeping the initial rotation perturbation matrix unchanged, process the object image clustering objective function to obtain the first objective optimization function used to optimize the initial consistent clustering partition matrix and the initial weight coefficients.

[0129] Step 710: Calculate the simplified gradient of the initial weight coefficients based on the first objective optimization function.

[0130] Step 712: With the sum of the initial weight coefficients of each view being a preset value, perform gradient descent on the object image clustering objective function to update the simplified gradient.

[0131] Step 714: When the objective function of object image clustering has the minimum extremum in the gradient descent direction, the minimum weight coefficient and the optimal consistent clustering partition matrix are obtained. The minimum weight coefficient is determined as the intermediate weight coefficient, and the optimal consistent clustering partition matrix is ​​determined as the intermediate consistent clustering partition matrix.

[0132] Step 716: Update the initial weight coefficients based on the intermediate weight coefficients and update the initial consistent clustering partition matrix based on the intermediate consistent clustering partition matrix. Process the object image clustering objective function to obtain the second objective optimization function used to optimize the initial rotation perturbation matrix.

[0133] Step 718: Based on the second objective optimization function, the initial rotation perturbation matrix is ​​maximized using the singular value decomposition method to obtain the intermediate rotation perturbation matrix.

[0134] Step 720: If the loss function value of the clustering objective function of the optimized object image is not the preset function value, then continue to execute step 708.

[0135] Step 722: When the loss function of the objective function for clustering object images is a preset function value, the objective consistent clustering partition matrix is ​​obtained.

[0136] Step 724: Perform clustering processing on the target consistent clustering partition matrix to obtain the clustering results of the object image data.

[0137] It is understood that the construction and optimization of the object image clustering objective function in this embodiment can be achieved in the manner described above, and will not be elaborated upon here.

[0138] The following is an example of clustering processing applied to the method of this application in one embodiment. The acquired object image data is clustered based on the above-described post-fusion object image clustering method. The acquired object image data includes 6 benchmark datasets and 6 large-scale datasets, namely Football, Liver, Olympic, BBCSport, Caltech, Cora, Flower102, YoutubeFace, SUNRGBD, ALOI, Reuters, and NUS-WIDE, as shown in Table 1, which contains relevant information about the datasets.

[0139] Table 1. Relevant information about the dataset.

[0140] Dataset Name Number of samples Number of nuclei Number of clusters Football 248 9 20 Liver 345 6 2 Olympics 464 3 29 BBCSport 544 6 5 Caltech 1530 25 102 Cora 2708 2 7 Flower102 8189 4 102 YouTube Face 10153 5 31 SUNRGBD 10335 2 45 ALOI 10800 4 100 Reuters 18758 5 6 NUS-WIDE 23593 5 31

[0141] The aforementioned post-fusion object image clustering employs the following methods: Merging Kernels via Kernel Machines (MKKM), Local Kernel Kernels (LMKKM), Online Nonnegative Kernel Clustering (ONKC), Kernel Kernels with Matrix Induced Regularization (MKKM-MiR), Large Kernel Approximate Mean Shift Clustering (LKAM), Low-rank Factorization-based Multi-view Clustering (LF-MVC), Robust Kernel Spectral Clustering via Matrix Manifolds (MKKM-MM), Simple Kernel Kernels (SMKKM), and Hierarchical Low Rank Matrix Sketch (HLRMSC). (Clustering). In all experiments, all benchmark kernels were first centered and regularized. For all datasets, it was assumed that the number of classes was known and set as the number of clustering classes. The comparison algorithms used in this experiment were all configured with parameters according to the relevant literature; this method does not require adjustment of any hyperparameters.

[0142] This experiment used common clustering accuracy (ACC) and purity metrics to demonstrate the clustering performance of each method. To reduce the randomness caused by k-means, we randomly initialized all methods and repeated the experiment a preset number of times, reporting the mean and standard deviation. The preset number of trials could be 50. Table 2 shows the clustering accuracy (ACC) of different algorithms on six benchmark datasets. Table 3 shows the purity (Purity) of different algorithms on six benchmark datasets.

[0143] Table 2. Clustering accuracy of different algorithms on six benchmark datasets.

[0144]

[0145] Table 3. Purity of different algorithms on six benchmark datasets.

[0146]

[0147]

[0148] As shown in Tables 2 and 3, the post-fusion object image clustering method of this application and the comparison algorithms demonstrate the clustering performance on six benchmark datasets. Under both evaluation criteria, the post-fusion object image clustering method of this application outperforms all comparison algorithms. Furthermore, the clustering accuracy (ACC) of the post-fusion object image clustering method of this application is higher than the average by 6.33%, 5.59%, 6.98%, 10.78%, 2.71%, and 12.68% on the six benchmark datasets, respectively; and the purity is higher than the average by 6.46%, 1.87%, 6.34%, 4.16%, 2.84%, and 7.74%, respectively.

[0149] Table 4 shows the clustering accuracy (ACC) of different algorithms on six large-scale datasets, and Table 5 shows the purity of different algorithms on six large-scale datasets.

[0150] Table 4. Clustering accuracy of different algorithms on six large-scale datasets.

[0151]

[0152]

[0153] Table 5. Purity of different algorithms on six large-scale datasets.

[0154]

[0155]

[0156] Tables 4 and 5 present the clustering performance of the proposed post-fusion object image clustering method on large-scale datasets, including clustering accuracy and purity. A "-" indicates a memory overflow, preventing the algorithm from running. Tables 4 and 5 show that the proposed method achieves top-two clustering accuracy on five out of six large-scale datasets and best performance on three. Specifically, the proposed method outperforms the average clustering accuracy (ACC) on the six large-scale datasets by 7.26%, 5.41%, -0.30%, 13.93%, 5.00%, and 0.86%, respectively; and the average purity (Purity) by 7.87%, 3.38%, 0.77%, 13.89%, 1.29%, and 0.54%, respectively. Therefore, the proposed method demonstrates good clustering performance and high accuracy on large-scale datasets.

[0157] Given Figure 8The change of the objective function for object image clustering in each iteration is based on Figure 8 The schematic diagram shows that the objective function for clustering object images increases monotonically and can reach convergence within 30 iterations. Figure 9 This is a performance comparison chart of algorithms on the Caltech dataset in the examples. The metrics include Purity, Clustering Accuracy (ACC), Normalized Mutual Information (NMI), and Rand Index. The Caltech102 dataset includes Cal-5, Cal-10, Cal-15, Cal-20, Cal-25, and Cal-30 datasets.

[0158] The aforementioned method for constructing an object image clustering objective function, in response to an image clustering request, acquires the object image data to be clustered. By rotating the clustering partition matrix of each view, it solves the alignment problem of mismatched clusters between different views caused by the unsupervised nature of the clustering task. Furthermore, it weights and fuses the new clustering partition matrices corresponding to each view, reducing the computational complexity in time and space. Further, it determines the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partition matrix as parameters to be optimized, constructing the object image clustering objective function. By alternately optimizing the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partition matrix in the object image clustering objective function, it avoids the object image clustering objective function only converging to a local optimum, obtaining the target consistent clustering partition matrix that fuses the basic partition matrices of each view. By performing clustering processing on the target consistent clustering partition matrix, it obtains the clustering results of the object image data, improving the accuracy of image clustering.

[0159] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0160] Based on the same inventive concept, this application also provides a post-fusion object image clustering apparatus for implementing the post-fusion object image clustering method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more post-fusion object image clustering apparatus embodiments provided below can be found in the limitations of the post-fusion object image clustering method described above, and will not be repeated here.

[0161] In one embodiment, such as Figure 10 As shown, a post-fusion object image clustering device is provided, including: an image data acquisition module 1002, a data determination module 1004, a function construction module 1006, an optimization processing module 1008, and a clustering module 1010, wherein:

[0162] The image data acquisition module 1002 is used to acquire image data of objects to be clustered in response to an image clustering request.

[0163] The data determination module 1004 is used to determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views.

[0164] Function construction module 1006 is used to construct the object image clustering objective function based on the basic partition matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0165] The optimization processing module 1008 is used to perform alternating optimization processing on the parameters to be optimized in the object image clustering objective function. When the function value of the object image clustering objective function is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix.

[0166] Clustering module 1010 is used to perform clustering processing on the target consistent clustering partition matrix to obtain the clustering results of object image data.

[0167] The aforementioned post-fusion object image clustering device, in response to an image clustering request, acquires the object image data to be clustered; constructs an object image clustering objective function based on the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views; by alternately optimizing the initial rotation perturbation matrix, initial weight coefficients, and initial consistent clustering partitioning matrix in the object image clustering objective function, it avoids the object image clustering objective function from only converging to a local optimum, and obtains the target consistent clustering partitioning matrix that fuses the basic partitioning matrix of each view; by performing clustering processing on the target consistent clustering partitioning matrix, it obtains the clustering result of the object image data, thereby improving the accuracy of image clustering.

[0168] In another embodiment, a post-fusion object image clustering apparatus is provided, which, in addition to including an image data acquisition module 1002, a data determination module 1004, a function construction module 1006, an optimization processing module 1008, and a clustering module 1010, further includes a rotation processing module, wherein:

[0169] The rotation processing module is used to rotate the corresponding basic partition matrix based on the initial consistent clustering partition matrix and the initial rotation perturbation matrix of each view to obtain the rotated basic partition matrix of each view.

[0170] The function construction module 1006 is also used to perform weighted fusion of the corresponding rotated basic partition matrix according to the initial weight coefficients of each view, and to determine the initial rotation perturbation matrix, initial weight coefficients and initial consistent clustering partition matrix as parameters to be optimized, so as to obtain the object image clustering objective function.

[0171] The optimization processing module 1008 is also used to obtain optimization strategy data for the parameters to be optimized in the object image clustering objective function;

[0172] Based on the optimization strategy data, the parameters to be optimized in the object image clustering objective function are optimized. If the function value of the object image clustering objective function after optimization is not the preset function value, the step of optimizing the parameters to be optimized in the object image clustering objective function based on the optimization strategy data is continued.

[0173] When the objective function for clustering object images is set to a preset function value, a consistent clustering partition matrix is ​​obtained.

[0174] The optimization processing module 1008 is also used to perform minimum optimization processing on the initial weight coefficients based on the initial rotation perturbation matrix and the object image clustering objective function to obtain intermediate weight coefficients, and to perform maximum optimization processing on the initial rotation perturbation matrix to obtain an intermediate consistent clustering partition matrix.

[0175] Based on the intermediate weight coefficients, the intermediate consistent clustering partition matrix, and the object image clustering objective function, the initial rotation perturbation matrix is ​​optimized to obtain the intermediate rotation perturbation matrix.

[0176] The optimization processing module 1008 is also used to process the object image clustering objective function while keeping the initial rotation perturbation matrix unchanged, to obtain the first objective optimization function used to optimize the initial consistent clustering partition matrix and the initial weight coefficients;

[0177] Based on the first objective function, calculate the simplified gradient of the initial weight coefficients;

[0178] Given that the sum of the initial weight coefficients for each view is a preset value, gradient descent is applied to the object image clustering objective function to update the simplified gradient;

[0179] When the objective function for clustering object images has its minimum extremum in the gradient descent direction, we obtain the minimum weight coefficients and the optimal consistent clustering partition matrix. The minimum weight coefficients are determined as the intermediate weight coefficients, and the maximum consistent clustering partition matrix is ​​determined as the intermediate consistent clustering partition matrix.

[0180] The optimization processing module 1008 is also used to update the initial weight coefficients according to the intermediate weight coefficients and update the initial consistent clustering partition matrix according to the intermediate consistent clustering partition matrix, and process the object image clustering objective function to obtain a second objective optimization function for optimizing the initial rotation perturbation matrix;

[0181] Based on the second objective optimization function, the initial rotation perturbation matrix is ​​maximized using the singular value decomposition method to obtain the intermediate rotation perturbation matrix.

[0182] Each module in the aforementioned post-fusion object image clustering device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0183] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 11As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a post-fusion object image clustering method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0184] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0185] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0186] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0187] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0188] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0189] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0190] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0191] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A post-processing fusion method for object image clustering, characterized in that, The method includes: In response to an image clustering request, obtain the image data of the objects to be clustered; Determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views; Based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix, a clustering objective function for object images is constructed. The parameters to be optimized in the objective function of the object image clustering are alternately optimized. When the function value of the objective function of the object image clustering is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix. Clustering is performed on the target consistent clustering partition matrix to obtain the clustering results of the object image data; The step of constructing the object image clustering objective function based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix includes: Based on the initial consistent clustering partition matrix and the initial rotation perturbation matrix of each view, the corresponding basic partition matrix is ​​rotated to obtain the rotated basic partition matrix of each view; The initial weight coefficients of each view are used to perform weighted fusion on the corresponding rotated basic partition matrix, and the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix are determined as parameters to be optimized to obtain the object image clustering objective function; The constructed objective function for clustering the object images is expressed as follows: ； It is the basic partitioning matrix for each view; It is the initial consistent clustering partition matrix and belongs to ; It is the initial rotation perturbation matrix for each view, used for alignment. and Cluster correspondence between them; The initial weight coefficients represent the weighting of each view and belong to... .

2. The method according to claim 1, characterized in that, The process of alternately optimizing the parameters to be optimized in the objective function of object image clustering, and obtaining a consistent clustering partition matrix when the function value of the objective function of object image clustering is a preset function value, includes: Obtain optimization strategy data for the parameters to be optimized in the objective function of the object image clustering; Based on the optimization strategy data, the parameters to be optimized in the object image clustering objective function are optimized. If the function value of the object image clustering objective function after optimization is not the preset function value, the step of optimizing the parameters to be optimized in the object image clustering objective function based on the optimization strategy data is continued. When the function value of the target function for clustering the object image is a preset function value, a target-consistent clustering partition matrix is ​​obtained.

3. The method according to claim 2, characterized in that, The optimization process, based on the optimization strategy data, involves optimizing the parameters to be optimized in the objective function of object image clustering, including: Based on the initial rotation perturbation matrix and the object image clustering objective function, the initial weight coefficients are minimized to obtain intermediate weight coefficients, and the initial rotation perturbation matrix is ​​maximized to obtain an intermediate consistent clustering partition matrix. Based on the intermediate weight coefficients, the intermediate consistent clustering partition matrix, and the object image clustering objective function, the initial rotation perturbation matrix is ​​subjected to maximum optimization processing to obtain the intermediate rotation perturbation matrix.

4. The method according to claim 3, characterized in that, The step of performing minimum optimization on the initial weight coefficients based on the initial rotation perturbation matrix and the object image clustering objective function to obtain intermediate weight coefficients, and performing maximum optimization on the initial rotation perturbation matrix to obtain an intermediate consistent clustering partition matrix, includes: While keeping the initial rotation perturbation matrix unchanged, the objective function for clustering the object image is processed to obtain a first objective optimization function for optimizing the initial consistent clustering partition matrix and the initial weight coefficients; Based on the first objective optimization function, calculate the simplified gradient of the initial weight coefficients; If the sum of the initial weight coefficients for each view is a preset value, gradient descent is performed on the object image clustering objective function to update the simplified gradient; When the objective function for clustering the object image has its minimum extremum in the gradient descent direction, the minimum weight coefficient and the optimal consistent clustering partition matrix are obtained. The minimum weight coefficient is determined as the intermediate weight coefficient, and the optimal consistent clustering partition matrix is ​​determined as the intermediate consistent clustering partition matrix.

5. The method according to claim 3, characterized in that, The process of performing maximum optimization on the initial rotation perturbation matrix based on the intermediate weight coefficients, the intermediate consistent clustering partition matrix, and the object image clustering objective function to obtain the intermediate rotation perturbation matrix includes: The initial weight coefficients are updated based on the intermediate weight coefficients, and the initial consistent clustering partition matrix is ​​updated based on the intermediate consistent clustering partition matrix. The object image clustering objective function is then processed to obtain a second objective optimization function for optimizing the initial rotation perturbation matrix. Based on the second objective optimization function, the initial rotation perturbation matrix is ​​maximized using the singular value decomposition method to obtain the intermediate rotation perturbation matrix.

6. A post-fusion object image clustering device, characterized in that, The device includes: The image data acquisition module is used to acquire image data of objects to be clustered in response to image clustering requests; The data determination module is used to determine the basic partitioning matrix, initial rotation perturbation matrix, and initial weight coefficients of the object image data under different views, as well as the initial consistent clustering partitioning matrix corresponding to all views; The function construction module is used to construct an object image clustering objective function based on the basic partitioning matrix, the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partitioning matrix; The optimization processing module is used to alternately optimize the parameters to be optimized in the object image clustering objective function. When the function value of the object image clustering objective function is a preset function value, a target consistent clustering partition matrix that fuses the basic partitioning matrix of each view is obtained. The parameters to be optimized include the initial rotation perturbation matrix, the initial weight coefficients, and the initial consistent clustering partition matrix. The clustering module is used to perform clustering processing on the target consistent clustering partition matrix to obtain the clustering results of the object image data; The rotation processing module is used to rotate the corresponding basic partition matrix based on the initial consistent clustering partition matrix and the initial rotation perturbation matrix of each view to obtain the rotated basic partition matrix of each view. The function construction module is further configured to perform weighted fusion on the corresponding rotated basic partition matrix according to the initial weight coefficients of each view, and to determine the initial rotation perturbation matrix, the initial weight coefficients and the initial consistent clustering partition matrix as parameters to be optimized, so as to obtain the object image clustering objective function; The constructed objective function for clustering the object images is expressed as follows: ； It is the basic partitioning matrix for each view; It is the initial consistent clustering partition matrix and belongs to ; It is the initial rotation perturbation matrix for each view, used for alignment. and Cluster correspondence between them; The initial weight coefficients represent the weighting of each view and belong to... .

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images