Robust image clustering method suitable for abnormal image interference environment

Abstract

The present application relates to a kind of robust image clustering method suitable for abnormal image interference environment, belong to image clustering technical field, comprising the following steps: S1: the image data input robust clustering model with adaptive learning ability;S2: given domain parameter, weight parameter;S3: learning space threshold value;S4: solving robust adjacency matrix;S5: solving approximate regular term, obtain robust embedding;S6: solving robust structure learning model, obtain new robust adjacency matrix;S7: judge whether the difference of the robust adjacency matrix of before and after two times satisfies threshold value, if satisfy then output the new robust adjacency matrix and robust embedding F, if not satisfy then return step S5, and the robust adjacency matrix in step S5 is replaced as the new robust adjacency matrix S8: robust adjacency matrix is used as the input of traditional spectral clustering algorithm, and the label information of real sample is obtained.

Description

Technical Field

[0001] This invention belongs to the field of image clustering technology and relates to a robust image clustering method suitable for abnormal image interference environments. Background Technology

[0002] For clustering problems under outlier interference, current research mainly focuses on the following types of algorithms: Density-based robust clustering algorithms: These models identify cluster structures by analyzing the local density distribution of data points and treat data points in low-density regions as outliers. Representative models include the density-based noise-applied spatial clustering model proposed by Ester et al., the density-based density clustering model based on reachability ranking proposed by Ankerst et al., and the hierarchical density clustering model proposed by Campello et al. They do not require a pre-defined number of clusters and are suitable for complex distributed data; however, parameter selection has a significant impact on the results, and it is difficult to achieve an ideal balance between robustness and clustering performance. Robust clustering models based on nonnegative matrix factorization: These models assume that the data exists in a low-dimensional subspace. By introducing a robust loss function, they decompose the data matrix into a coordinate matrix and a coefficient matrix, effectively capturing the partial representation of the data and reducing the impact of outliers on the clustering results. Representative models include those proposed by Kong et al. Non-negative matrix factorization (NMF) models based on the loss function, the NMF model based on the Hx loss function proposed by Wang et al., and their derivative models. Although they can extract the low-dimensional structure of data, they have high computational complexity. Robust clustering models based on graph models: These models construct similarity graphs or adjacency matrices between data points and utilize graph regularization or sparse constraints to suppress outliers. Representative models include sparse subspace clustering proposed by Elhamifar et al., robust graph clustering proposed by Shi et al., and low-rank representation models proposed by Liu et al. While they can handle non-convex data, the construction of similarity graphs (or adjacency matrices) is complex and sensitive to initial parameters.

[0003] Although the above methods improve robustness to some extent, they still have problems such as high parameter sensitivity, high computational cost, and limited clustering performance, making it difficult to obtain stable and reliable results in various application scenarios. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a robust image clustering method suitable for abnormal image interference environments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A robust image clustering method suitable for abnormal image interference environments includes the following steps: S1: Input the image data into a robust clustering model with adaptive learning capabilities; S2: Given domain parameters Weight parameters ; S3: Learning Space Threshold ; S4: Will As input to the robust clustering model, solve for the robust adjacency matrix. ; S5: Based on Solving for the approximate regularization term yields the robust embedding. ; S6: Based on and Solve the robust structure learning model to obtain a new robust adjacency matrix. ; S7: Determine the difference between two robust adjacency matrices. If the threshold is met, output the new robust adjacency matrix. and robust embedding If the condition is not met, return to step S5 and change the robust adjacency matrix from step S5. Replace with this new robust adjacency matrix ; S8: Use the robust adjacency matrix As input to the traditional spectral clustering algorithm, the label information of the image data is obtained.

[0006] Furthermore, the robust structure learning model is represented as follows:

[0007]

[0008] in Sample points The probability of adjacency with other sample points; ; It is a field parameter; These are the regularization parameters to be determined.

[0009] Furthermore, the approximate regularization term specifically includes: Constrained Assignment Probabilities This allows it to have an adaptive learning process; Let the solution obtained by the robust structure learning model be... and order ; Assuming each sample point Assigned a low-dimensional function value If neighboring sample point pairs have similar low-dimensional function values, and unrelated sample point pairs have mutually orthogonal low-dimensional function values, this constraint can be expressed as:

[0010]

[0011] The above constraints are transformed into a symmetric nonnegative matrix factorization problem:

[0012]

[0013] Furthermore, the robust clustering model with adaptive learning capability, combined with a robust structure learning model and an approximate regularization term, is achieved through the following optimization objective:

[0014]

[0015] in These are weight parameters used to regulate the robust structure learning model and the approximate regularization term.

[0016] Furthermore, the learning space threshold mentioned in step S3 Specifically, it includes: Assume the input data matrix is Its corresponding distance matrix is ,in Representing data points and Euclidean distance between them; for matrices Sort the elements in each row in descending order to get the corresponding sorting matrix ,in express The Given column vectors, then,

[0017] in Representing vectors the median of Representing vectors The mean.

[0018] Furthermore, step S4 describes... As input to a robust image clustering method, the robust adjacency matrix is ​​solved. Specifically, it includes: For any , row vectors The calculation formula is:

[0019] in express The largest number between 0 and 0; For matrix The Okay, number Column element values; To meet The corresponding serial number.

[0020] Furthermore, the step S5 based on Solving for the approximate regularization term yields the robust embedding. Specifically, it includes: For any ,matrix element value It can be obtained through the following iterative formula: make

[0021]

[0022]

[0023] in Indicates the number of categories required; This indicates that the data was generated using the software's built-in uniform distribution function. OK Column matrix; and Represent matrices respectively and The Okay, number Column element values.

[0024] Furthermore, the step S6 based on and Solve the robust structure learning model to obtain a new robust adjacency matrix. Specifically, it includes: For any , row vectors The calculation formula is

[0025] in express The largest number between 0 and 0; For matrix The Okay, number Column element values; To meet The corresponding serial number; Representation matrix Zhongyu The element values ​​for the corresponding row and column.

[0026] Furthermore, in step S7, a threshold is set; if the difference between the two robust adjacency matrices is... If the value is less than or equal to the threshold, the algorithm terminates and outputs the result. and If it is greater than the threshold, then let Return to step S5.

[0027] The beneficial effects of this invention are as follows: Addressing the clustering problem under outlier interference environments, this project proposes a robust clustering model with adaptive learning capabilities. This model achieves accurate identification and effective isolation of outliers through a robust structure learning model, and combines an approximate regularization strategy to efficiently extract robust embedding features from the data, ensuring the stability and reliability of the clustering results. This technological breakthrough not only reduces the dependence of existing robust clustering models on parameters and reduces computational burden, but also provides accurate and efficient data analysis support for scenarios with frequent outliers, such as financial risk assessment, clinical medical image analysis, and image classification.

[0028] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 The structure diagram of a robust clustering model with adaptive learning capabilities; Figure 2 This is the raw facial data; Figure 3 For robust adjacency matrix ; Figure 4 For robust embedding ; Figure 5 For the new robust adjacency matrix ; Figure 6 This represents the final image clustering result. Detailed Implementation

[0030] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0032] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0033] Example 1: To address the clustering problem under outlier interference, this invention constructs an innovative robust clustering model. This model integrates spatial thresholding and adaptive structure learning strategies to automatically identify outliers while simultaneously learning the distribution information of true values. Furthermore, drawing on the idea of ​​nonnegative matrix factorization, it transforms the eigenvector calculation process of the Laplacian matrix into a regularization term, significantly reducing the computational burden. The specific technical approach is as follows: (1) Robust Structure Learning Model: In order to effectively capture the spatial relationships between samples and suppress the interference of outliers on the spatial network, this project proposes a robust structure learning model. The model assumes: ① The number of neighbors of a sample point is inversely proportional to its distance from the cluster center; that is, the smaller the distance, the more neighbors it has.

[0034] ② Sample point pairs adjacency probability Its spatial distance Negative correlation. If the distance between two points... Exceeding the preset space threshold If they do not constitute neighbors, then they do not have a neighbor relationship.

[0035] ③If a certain sample point The distance to all other sample points is greater than If it is an outlier, it will be marked as an outlier and excluded from the neighbor relationships.

[0036] The above mechanism can be formally expressed by the following model:

[0037]

[0038] in Sample points The adjacency probability with other sample points. . It is a domain parameter. These are the regularization parameters to be determined.

[0039] (2) Approximate Regularization: Although the robust structure learning model can effectively construct the adjacency matrix, it still has a certain disconnect from the data clustering process. This may lead to the final low-dimensional embedding representation failing to fully capture the geometric structure information of the data space. To overcome this deficiency, this invention constrains the allocation probability. This allows it to have an adaptive learning process. Specifically, let the solution obtained by the robust structure learning model be denoted as . and order Assuming each sample point Assigned a low-dimensional function value If the condition is met, then neighboring sample point pairs should have similar low-dimensional function values, and unrelated sample point pairs should have mutually orthogonal low-dimensional function values. This constraint can be expressed as:

[0040]

[0041] However, orthogonal constraints The existence of this significantly increases the computational cost of solving the model. Therefore, this invention further transforms the above model into a symmetric nonnegative matrix factorization problem:

[0042]

[0043] (3) Robust clustering model with adaptive learning capability: Combining a robust structure learning model and an approximate regularization term, this project proposes a robust clustering model with adaptive learning capability to effectively address clustering tasks contaminated by outliers. Its structure is as follows: Figure 1 As shown. This model is achieved through the following optimization objectives:

[0044]

[0045] in These are weight parameters used to regulate the robust structure learning model and the approximate regularization term.

[0046] Example 2: This embodiment provides a robust image clustering method suitable for abnormal image interference environments. Taking face image clustering as an example, it includes the following steps: Step 1: Input facial data from a publicly available dataset, for example: ,in Belonging to the same person Belonging to the same person Belonging to the same person It belongs to noise, such as Figure 2 As shown.

[0047] Step 2: Given the domain parameters Weight parameters ,For example: ; Step 3: Learning Space Threshold The input data matrix is Its corresponding distance matrix is ,in Representing data points and Euclidean distance between them; for matrices Sort the elements in each row in descending order to get the corresponding sorting matrix ,in express The Given column vectors, then,

[0048] in Representing vectors the median of Representing vectors The mean.

[0049] For example: ,in yes The sequence of numbers sorted from smallest to largest. ; Step 4: As input to the robust clustering model, solve for the robust adjacency matrix. ,like Figure 3 As shown, for any , row vectors The calculation formula is:

[0050] in express The largest number between 0 and 0; For matrix The Okay, number Column element values; To meet The corresponding serial number.

[0051] Step 5: Based on Solving for the approximate regularization term yields the robust embedding. ,like Figure 4 As shown. For any ,matrix element value It can be obtained through the following iterative formula: make

[0052]

[0053]

[0054] in Indicates the number of categories required; This indicates that the data was generated using the software's built-in uniform distribution function. OK Column matrix; and Represent matrices respectively and The Okay, number Column element values.

[0055] Step 6: Based on and Solve the robust structure learning model to obtain a new robust adjacency matrix. ,like Figure 5 As shown. For any , row vectors The calculation formula is

[0056] in express The largest number between 0 and 0; For matrix The Okay, number Column element values; To meet The corresponding serial number; Representation matrix Zhongyu The element values ​​for the corresponding row and column.

[0057] Step 7: Determine the difference between the two robust adjacency matrices. Does it meet the threshold? For example, if the threshold is set to... .like The algorithm then terminates and outputs the new robust adjacency matrix. and robust embedding If the condition is not met, return to step 5 and change the robust adjacency matrix from step 5. Replace with this new robust adjacency matrix .

[0058] Step 8: Convert the robust adjacency matrix As input to a traditional spectral clustering algorithm, the label information of the image data is obtained. The final result is as follows: Figure 6 As shown.

[0059] In the above embodiments, the reference to "this embodiment" in the specification indicates that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple appearances of "this embodiment" do not necessarily refer to the same embodiment.

[0060] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0061] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods in this embodiment.

[0062] This embodiment also provides an electronic terminal, including: a processor and a memory; The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to cause the terminal to perform any of the methods in this embodiment.

[0063] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0064] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0065] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0066] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0067] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0068] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A robust image clustering method suitable for abnormal image interference environments, characterized in that: Includes the following steps: S1: Input the image data into a robust clustering model with adaptive learning capability; the robust clustering model with adaptive learning capability is obtained by combining a robust structure learning model and an approximate regularization term; the robust structure learning model is expressed as: in Sample points The probability of adjacency with other sample points; ; It is a field parameter; These are the regularization parameters to be determined; The approximate regularization term specifically includes: Constrained Assignment Probabilities This allows it to have an adaptive learning process; Let the solution obtained by the robust structure learning model be... and order ; Assuming each sample point Assigned a low-dimensional function value If neighboring sample point pairs have similar low-dimensional function values, and unrelated sample point pairs have mutually orthogonal low-dimensional function values, this constraint can be expressed as: The sample points For image sample points; The above constraints are transformed into a symmetric nonnegative matrix factorization problem: The robust clustering model with adaptive learning capability is achieved through the following optimization objective: in These are weight parameters used to regulate the robust structure learning model and the approximate regularization term; S2: Given domain parameters Weight parameters ; S3: Learning Space Threshold Specifically, it includes: Assume the input data matrix is Its corresponding distance matrix is ,in Represents sample points and Euclidean distance between them; for matrices Sort the elements in each row in descending order to get the corresponding sorting matrix ,in express The Given column vectors, then, in Representing vectors the median of Representing vectors The mean; S4: Will As input to the robust clustering model, solve for the robust adjacency matrix. ; S5: Based on Solving for the approximate regularization term yields the robust embedding. ; S6: Based on and Solve the robust structure learning model to obtain a new robust adjacency matrix. Specifically, it includes: For any , row vectors The calculation formula is: in express The largest number between 0 and 0; For matrix The Okay, number Column element values; To meet The corresponding serial number; Representation matrix Zhongyu The element values ​​corresponding to the row and column; S7: Determine the difference between two robust adjacency matrices. If the threshold is met, output the new robust adjacency matrix. and robust embedding If the condition is not met, return to step S5 and change the robust adjacency matrix from step S5. Replace with this new robust adjacency matrix ; S8: Use the robust adjacency matrix As input to the traditional spectral clustering algorithm, the label information of the image data is obtained.

2. The robust image clustering method for abnormal image interference environments according to claim 1, characterized in that: Step S4 describes the following: As input to a robust image clustering method, the robust adjacency matrix is ​​solved. Specifically, it includes: For any , row vectors The calculation formula is: in express The largest number between 0 and 0; For matrix The Okay, number Column element values; To meet The corresponding serial number.

3. The robust image clustering method applicable to abnormal image interference environments according to claim 1, characterized in that: Step S5 is based on Solving for the approximate regularization term yields the robust embedding. Specifically, it includes: For any ,matrix element value It can be obtained through the following iterative formula: make in Indicates the number of categories required; This indicates that the data was generated using the software's built-in uniform distribution function. OK Column matrix; and Represent matrices respectively and The Okay, number Column element values.

4. The robust image clustering method applicable to abnormal image interference environments according to claim 1, characterized in that: In step S7, a threshold is set; if the difference between the two robust adjacency matrices is... If the value is less than or equal to the threshold, the algorithm terminates and outputs the result. and If it is greater than the threshold, then let Return to step S5.

Images