Weighted rough membership clustering RGB-D image segmentation method based on composite distance

Abstract

The invention discloses a weighted rough membership clustering RGB-D image segmentation method based on a composite distance. Comprising the steps that image features of an RGB-D image are extracted, and the image features comprise color features, point cloud features and normal features; selecting an initial class center based on a density method; the clustering algorithm is optimized based on adaptive weighting and a composite distance measurement method, and a weighted rough membership clustering algorithm is obtained; the self-adaptive weighting comprises the step of carrying out self-adaptive weighting on each feature dimension through a local feature weighting method; the composite distance measurement method comprises the following steps: calculating the similarity between a sample and a clustering center through a method of fusing an Euclidean distance and a cosine distance; and clustering the RGB-D image based on a weighted rough membership clustering algorithm and carrying out region merging to generate a segmented image. According to the method, feature weighting is introduced and a similarity measurement method is improved, so that the method can better adapt to the characteristics of different types of features, and finally, a higher-quality segmentation effect is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a weighted rough membership clustering RGB-D image segmentation method based on composite distance. Background Art

[0002] In 2010, Microsoft launched the Kinect camera, which can capture depth information within a scene. Incorporating depth information into RGB images has enhanced computer vision capabilities. Depth information is unaffected by factors such as lighting and shadows. Combining RGB and depth images to generate RGB-D images can better handle uncertain areas in color images. Currently, RGB-D image segmentation has been widely used in scenarios such as autonomous driving, robot navigation, smart homes, and agricultural automation. Existing RGB-D image segmentation methods primarily include deep learning-based methods and traditional segmentation techniques. Deep learning-based RGB-D image segmentation methods offer high segmentation accuracy, but their training requires a large amount of labeled sample data and places high demands on hardware performance. In contrast, traditional image segmentation methods have been widely used due to their convenience and speed. Traditional image segmentation algorithms primarily include threshold-based methods, edge detection-based methods, and clustering-based methods. RGB-D image segmentation algorithms based on cluster analysis utilize only information from the image itself, have lower algorithmic complexity, and can achieve satisfactory segmentation results even in the absence of labeled data or prior knowledge.

[0003] Existing clustering-based RGB-D image segmentation algorithms directly segment images without considering the multimodal features of RGB-D images. The direct segmentation process does not take into account the differences between different modal features, and it is difficult to effectively fuse the information between the features. In addition, existing methods use Euclidean distance to calculate the similarity between pixels. Since the features of different modalities may have different properties and distributions, the Euclidean distance cannot accurately measure the similarity between them. Based on the above problems, in order to improve the segmentation accuracy of RGB-D images, the present invention proposes a weighted rough membership clustering (Weighted Composite Distance Rough Membership C-means Clustering, WCD-RMCM) RGB-D image segmentation method based on composite distance. A composite distance for RGB-D multimodal features is constructed, and an improved weighted rough membership clustering algorithm is proposed. Summary of the Invention

[0004] The present invention proposes a weighted rough membership clustering RGB-D image segmentation method based on composite distance to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a weighted rough membership clustering RGB-D image segmentation method based on composite distance, comprising the following steps:

[0006] Extracting image features of the RGB-D image, wherein the image features include color features, point cloud features, and normal features;

[0007] The density-based method selects the class center and uses the high-density area as the initial class center;

[0008] The clustering algorithm is optimized based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm.

[0009] Adaptive weighting involves adaptively weighting each feature dimension through local feature weighting method;

[0010] The composite distance measurement method includes calculating the similarity between samples and cluster centers by fusing Euclidean distance and cosine distance;

[0011] The RGB-D image is clustered based on the weighted rough membership clustering algorithm, and the clustering results are merged to generate a segmented image.

[0012] Preferably, when selecting the initial class center, the neighborhood density of each point is calculated and a high-density point is selected as the initial class center.

[0013] Preferably, during adaptive weighting, different weights are assigned to color features, point cloud features, and normal features according to the importance of each feature in the cluster, and the weights are dynamically adjusted according to the intra-cluster distance during the iteration process.

[0014] Preferably, in the composite distance measurement method, Euclidean distance is used to calculate similarity for color features and point cloud features, cosine distance is used to calculate similarity for normal features, and different distance metrics are normalized by a linear normalization method.

[0015] Preferably, the expression of the weighted rough membership clustering algorithm is:

[0016]

[0017] In the formula, n, M and C represent the number of data samples, the spatial dimension of the feature and the number of cluster centers respectively, and W cm Represents the feature weight of the mth dimension in the cth class, and q is an exponential parameter used to adjust the weight W cm The strength of the effect on the feature, For sample x i To cluster center x cBased on the rough membership of the grid neighborhood binary relation R, d f Represents the sample point x i To cluster center b c The composite distance, x im represents the m-th dimension feature of the i-th sample, b cm Represents the m-th dimension feature of the c-th cluster center.

[0018] Preferably, the method updates the rough membership by Lagrangian function and constraint conditions, and comprehensively considers the local structure and global characteristics of the data.

[0019] Preferably, the region merging includes: treating each label region as a node, merging nodes that meet the conditions by checking the candidate eligibility and merging eligibility of the node and verifying the merging consistency, until there are no more candidate nodes to be merged.

[0020] The present invention also provides a weighted rough membership clustering RGB-D image segmentation system based on composite distance, comprising:

[0021] A feature extraction module is used to extract image features of the RGB-D image, wherein the image features include color features, point cloud features, and normal features;

[0022] Initialization module, used to improve the traditional random initial class center selection method and select points in high-density areas as initial class centers;

[0023] An optimization module is used to optimize the clustering algorithm based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm;

[0024] Among them, adaptive weighting includes adaptively weighting each feature dimension through local feature weighting method;

[0025] Among them, the composite distance measurement method includes calculating the similarity between the sample and the cluster center by integrating the composite distance measurement method of Euclidean distance and cosine distance;

[0026] The image segmentation module is used to cluster RGB-D images based on the weighted rough membership clustering algorithm, and merge the clustering results into regions to generate segmented images.

[0027] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0028] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0029] Compared with the prior art, the present invention has the following advantages and technical effects:

[0030] The present invention discloses a weighted rough membership clustering RGB-D image segmentation method based on composite distance, comprising: extracting image features of the RGB-D image, wherein the image features include color features, point cloud features, and normal features; selecting initial cluster centers based on density; optimizing the clustering algorithm based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm; adaptive weighting includes adaptively weighting each feature dimension through a local feature weighting method; the composite distance measurement method includes calculating the similarity between samples and cluster centers through a composite distance measurement method that combines Euclidean distance and cosine distance; clustering the RGB-D image based on the weighted rough membership clustering algorithm, and performing regional merging on the clustering results to generate a segmented image. The present invention adopts a feature adaptive weighting method to assign different weights to different features based on their importance and contribution. This strategy aims to help the algorithm fully explore the potential relationships between features, enable different features to complement each other, adapt to the characteristics of different features, and achieve effective information fusion. To address the differences in expression and data distribution characteristics between different features, the present invention designs differentiated similarity metrics: for RGB and point cloud features, Euclidean distance is used to measure their spatial distribution similarity; for normal features, cosine distance is used to evaluate geometric structure similarity based on their directional characteristics. This method effectively preserves the inherent characteristics of features and enhances the generalization ability for complex tasks. In summary, by introducing feature weighting and improving similarity metrics, the present invention effectively captures the inherent characteristics of different modal data, enhances adaptability to complex data, and achieves higher-quality segmentation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0032] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0033] Figure 2 Schematic diagram of the effect of feature weighting on image cluster segmentation quality according to an embodiment of the present invention;

[0034] Figure 3 Schematic diagram of the effect of composite distance on image cluster segmentation quality according to an embodiment of the present invention;

[0035] Figure 4 Schematic diagram of the time consumption of the segmentation results of each method in the embodiment of the present invention;

[0036] Figure 5 The original images 01 and 08 in the NYUD2 RGB-D dataset;

[0037] Figure 6 1 and 2 are the segmentation results of the comparative method 1 (hard C-means clustering algorithm) of the embodiment of the present invention on the NYUD2 RGB-D images 01 and 08.

[0038] Figure 7 1 and 2 are the segmentation results of comparative method 2 (fast robust fuzzy C-means clustering algorithm) of an embodiment of the present invention on NYUD2 RGB-D images 01 and 08.

[0039] Figure 8 3 (multi-layer clustering algorithm) of the comparative method according to the embodiment of the present invention on the NYUD2 RGB-D images 01 and 08.

[0040] Figure 9 4 (joint color-space-direction clustering and region merging algorithm) of the embodiment of the present invention on the NYUD2 RGB-D images 01 and 08.

[0041] Figure 10 1 and 2 are the segmentation results of the comparative method 5 (chaotic kbest gravitational search algorithm + fuzzy c-means clustering algorithm) of the embodiment of the present invention on the NYUD2 RGB-D images 01 and 08.

[0042] Figure 11 3. The segmentation results of comparative method 6 (random Henry gas solubility optimization algorithm + fuzzy c-means clustering algorithm) of an embodiment of the present invention on NYUD2 RGB-D images 01 and 08.

[0043] Figure 12 1 and 2 are segmentation results of the method of the present invention on NYUD2 RGB-D images 01 and 08 according to an embodiment of the present invention.

[0044] Figure 13 These are the original images 41 and 43 in the SUN RGB-D dataset.

[0045] Figure 14 4 is the segmentation result of the comparative method 1 (hard C-means clustering algorithm) of the embodiment of the present invention on the SUN RGB-D images 41 and 43.

[0046] Figure 15 4 is the segmentation result of the comparative method 2 (fast robust fuzzy C-means clustering algorithm) of the embodiment of the present invention on the SUN RGB-D images 41 and 43.

[0047] Figure 164 is the segmentation result of the comparative method 3 (multi-layer clustering algorithm) of the embodiment of the present invention on the SUN RGB-D images 41 and 43.

[0048] Figure 17 4 are the segmentation results of the comparative method 4 (joint color-space-direction clustering and region merging algorithm) of the embodiment of the present invention on SUN RGB-D images 41 and 43.

[0049] Figure 18 4 is the segmentation result of the comparative method 5 (chaotic kbest gravitational search algorithm + fuzzy c-means clustering algorithm) of the embodiment of the present invention on the SUN RGB-D images 41 and 43.

[0050] Figure 19 4 is the segmentation result of the comparative method 6 (random Henry gas solubility optimization algorithm + fuzzy C-means clustering algorithm) of the embodiment of the present invention on the SUN RGB-D images 41 and 43.

[0051] Figure 20 4 is the segmentation result of the method of the present invention on the SUN RGB-D images 41 and 43 according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0053] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0054] RGB-D images play a crucial role in autonomous driving technology. By fusing color information and depth data, they provide vehicles with precise three-dimensional environmental perception. These images not only enable autonomous driving systems to effectively detect and avoid obstacles, but also help vehicles identify and track lane lines, detect free space, and recognize traffic signs and signals. RGB-D images also have significant application value in vehicle positioning and mapping, human-computer interaction, augmented reality displays, path planning, and vehicle control. They enable vehicles to understand their surroundings more comprehensively and accurately, significantly improving the safety and reliability of autonomous driving and providing strong support for the development of autonomous driving technology.

[0055] Example 1

[0056] like Figure 1As shown, this embodiment provides a weighted rough membership clustering RGB-D image segmentation method based on composite distance, including the following steps:

[0057] Extract image features of RGB-D images, including color features, point cloud features, and normal features;

[0058] Density-based class center selection uses points in high-density areas as initial class centers.

[0059] The clustering algorithm is optimized based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm.

[0060] Adaptive weighting involves adaptively weighting each feature dimension through local feature weighting method;

[0061] The composite distance measurement method includes calculating the similarity between samples and cluster centers by fusing Euclidean distance and cosine distance;

[0062] The RGB-D image is clustered based on the weighted rough membership clustering algorithm, and the clustering results are merged to generate a segmented image.

[0063] The specific technical solution consists of the following steps:

[0064] (1) Initialization phase;

[0065] For multi-target image segmentation tasks, a reasonable initial cluster center strategy can significantly improve segmentation accuracy and accelerate algorithm convergence. By using high-density points as initial cluster centers, the inherent convergence instability of random initialization methods can be effectively overcome, thereby improving image segmentation accuracy. First, K similar samples are formed into a dense point, called "density". Its calculation formula is as follows:

[0066]

[0067] Among them, d f (x i ,x j ) represents the sample point x i Its neighboring sample point x j First, by calculating the composite distance between the sample points x i The sum of the distances of the K neighboring samples is used to obtain the local density of the sample point. Then, the sample density is sorted and a density threshold ρ is set. t , select points with density greater than the threshold as high-density points. Finally, randomly select C points from these high-density points as initial cluster centers.

[0068] b in_c ={xi |ρ i >ρ t} (2)

[0069] where b in_c Indicates the initialization of cluster centers.

[0070] (2) Construction of feature-weighted clustering criterion function;

[0071] In multimodal data analysis, different modal features often show significant differences in importance, among which some key features have a more important discriminative effect on clustering results. However, the current image segmentation algorithms based on traditional FCM generally have the problem of insufficient consideration of the differences in the importance of multimodal features, and fail to fully reflect the dominant role of key features in the formation of optimal clusters. To address this problem, a local feature weighting scheme is designed to improve clustering accuracy, and adaptive weighting is performed on each feature dimension. When calculating similarity, different features are adjusted according to their actual importance and contribution. The adaptive weights can be dynamically adjusted through prior analysis of the data set or feedback from the training set, thereby optimizing the entire clustering process. Taking the above considerations into account, the improved feature weighted clustering criterion function is expressed as:

[0072]

[0073] In formula (3), n, M, and C represent the number of data samples, the spatial dimension of the feature, and the number of cluster centers, respectively. cm Represents the feature weight of the mth dimension in the cth class, and q is an exponential parameter used to adjust the weight W cm The strength of the effect on the feature, is x i to x c Based on the rough membership of the grid neighborhood binary relation R, d f Represents the sample point x i To cluster center b c The composite distance, x im represents the m-th dimension feature of the i-th sample, b cm Represents the m-th dimension feature of the c-th cluster center.

[0074] (3) Calculation of composite distance;

[0075] Compared with traditional RGB images, RGB-D images have color features In order to deeply explore the feature information of the depth image, the present invention extracts the feature of the depth information and obtains the point cloud feature of the image. and normal features Among them, the first two dimensions of point cloud features Represents the coordinate position of the pixel in the two-dimensional plane space, Corresponding to its depth value; normal feature This describes the orientation of the pixel in three-dimensional space. By combining these geometric features with the original RGB color features, a nine-dimensional feature vector space is constructed. Therefore, in the proposed representation method, each pixel of the RGB-D image can be represented by this nine-dimensional feature vector.

[0076] In traditional clustering methods, Euclidean distance is usually used to calculate the similarity between sample points. However, for the feature analysis of RGB-D images, a single measurement method is difficult to characterize the essential differences between multidimensional features. To this end, the present invention proposes a composite distance measurement method to achieve the difference evaluation of different feature spaces. Since the normal exists in the form of a vector, the use of cosine distance can more effectively characterize its spatial distribution relationship. In the clustering process, the similarity measure between the sample and the class center uses a composite distance. For RGB features and point cloud features, the Euclidean distance is used to calculate their similarity, while for normal features, the cosine distance is used to measure their similarity.

[0077] Composite distance d f The calculation method is as follows:

[0078] d f =d r '+d e ' (4)

[0079] in

[0080]

[0081] Where x im represents the m-th dimension feature of the i-th sample, b cm represents the m-th dimension feature of the c-th cluster center, d r Represents the Euclidean distance from the sample point to the cluster center, d e Indicates the cosine distance from the sample point to the cluster center. In order to eliminate the dimension difference between different distance metrics, the linear normalization method is used to normalize the distance value. Specifically, for each distance d r and d e , the normalized value d r ' and d e The calculation formula is:

[0082]

[0083] where d rmax and d rmin is the maximum and minimum value in the Euclidean distance, d emax and d eminare the maximum and minimum values in the cosine distance.

[0084] (4) Clustering optimization;

[0085] Optimization of rough membership U;

[0086] For the clustering criterion function in formula (3), its constraints are as follows:

[0087]

[0088] According to the constraints, the Lagrangian function can be obtained:

[0089]

[0090] Find F with respect to u ci The partial derivative of and set it to 0 is:

[0091]

[0092] but

[0093]

[0094] Because of satisfaction Constraints can be obtained

[0095]

[0096] but

[0097]

[0098] At λ→0, if is the minimum value in the range of 1≤l≤C, then the above formula approaches 1, otherwise it is 0. Finally, we get u ci The update rules are as follows:

[0099]

[0100] but

[0101]

[0102] Among them, R it Represents pixel x i With pixel x t The binary relationship between them is defined as formula (17), R it =1 means pixel x t Located at pixel x i In the I×I neighborhood of , the opposite is not satisfied.

[0103]

[0104] Optimization of clustering weight W;

[0105] Fixed cluster centers C and u ci , find F with respect to W cm The partial derivative of and set it to 0 is:

[0106]

[0107] Also available:

[0108]

[0109] By constraints We can get:

[0110]

[0111] but

[0112]

[0113] Will Substituting into formula (19) we can get:

[0114]

[0115] The principle of weighting multimodal data in clustering is to assign greater weights to feature variables with smaller intra-cluster distances and smaller weights to feature variables with larger intra-cluster distances in related clusters. This principle is called the "cluster weighting principle," and it can prevent the generation of unbalanced clusters. When the parameter q = 0, formula (23) is equivalent to the unweighted rough membership clustering algorithm. When 0 < q < 1, the greater the sum of intra-cluster distances, the greater the weight obtained, which contradicts the "cluster weighting principle." Therefore, the parameter q is taken within the range of q < 0 and q > 1.

[0116] Optimization of cluster center b;

[0117] Fixed feature weight W cm and u ci , find F with respect to b cm The partial derivative of and set it to 0 is:

[0118]

[0119] Available

[0120]

[0121] Table 1 shows the pseudo code of the proposed composite distance weighted rough membership clustering RGB-D image segmentation algorithm.

[0122] Table 1

[0123]

[0124] The calculation process of the present invention is described in detail below in conjunction with Table 1 and the following examples, but the present invention is not limited to these examples.

[0125] First, feature extraction is performed on the depth image to extract the image's point cloud features and normal features. The point cloud represents the position information of the image pixels in two-dimensional space, and the normal features represent the orientation of the pixels in three-dimensional space. The point cloud and normal features are combined with the RGB color features to form a 9-dimensional combined feature set.

[0126] (1) Initialization phase;

[0127] Rows 1 to 3 of Table 1 represent the initialization phase. Row 3 introduces a density-based initialization method that analyzes the distribution of data points and prioritizes areas with higher density as cluster centers. First, similar samples are clustered together, and the density is reflected by calculating the local density of the sample points. For RGB-D multimodal data, Euclidean distance is used for color and point cloud features, and cosine distance is used for normal features. Then, the sample density is sorted, and a density threshold is set. Points with a density greater than the threshold are selected as high-density points. Finally, C points are randomly selected from these high-density points as initial cluster centers.

[0128] (2) Iteration phase;

[0129] The 4th to 10th rows of Table 1 are the iteration phase, which is performed by the criterion function To find the partial derivatives of each parameter, we can get the analytical optimization formula of each parameter. f By combining Euclidean distance and cosine distance, we can fully explore the potential information of each feature dimension and effectively integrate the features of each modality to improve the accuracy and comprehensiveness of data analysis. The updating process of the rough membership is based on the neighborhood relationship between the pixel and the surrounding pixels, the distance between the pixel and the cluster center, and the weight distribution of the features of each dimension. Under the optimization framework of multi-factor fusion, the local spatial structure and global statistical characteristics of the data can be captured simultaneously, thereby accurately reflecting the degree of membership of each pixel to the class center. The 8th line updates the weight of the feature variable. In this method, features with smaller intra-cluster distances are given larger weights, while features with larger intra-cluster distances are given smaller weights. Based on this principle, it can ensure that more attention is given to closer features, avoid serious imbalance between clusters during the clustering process, and improve the accuracy and rationality of the clustering results. The 9th line represents the update of the cluster center, which is based on the current data point classification and the rough membership matrix. For each cluster C, the new cluster center b cmA weighted average is calculated based on all data points in the rough membership matrix, where the weights are the rough memberships of the data points to the cluster center. If some points in a cluster have a high degree of membership to the cluster, this indicates that the spatial distance between these points and the cluster center is large. When the cluster center is updated, the cluster center will be adjusted toward the location of these points.

[0130] (3) Regional merger strategy;

[0131] Row 13 of Table 1 illustrates the region merging method. During region merging, each labeled region is treated as a node. By individually checking each node's candidacy, merging eligibility, and verifying the merge consistency, we determine which nodes can be merged. After the merge operation, the information about the merged nodes and their edges is updated to reflect the new merged structure. This iterative process continues until there are no more candidate nodes to merge in the system, ensuring the comprehensiveness and effectiveness of region merging. This approach results in more accurate region segmentation results.

[0132] In RGB-D image segmentation based on the weighted rough membership clustering algorithm, the region merging step optimizes clustering results through rigorous eligibility screening and consistency verification. First, a neighborhood search is performed to identify candidate regions that are spatially adjacent and have feature similarity above a preset threshold. These candidate region pairs are then checked for merging eligibility, including feature similarity, area ratio, and boundary consistency, to ensure that the merged regions maintain consistency in features and structure. Finally, a merge consistency verification step is performed to further assess whether the merged region features and boundaries remain consistent. This process is iterated until no more candidate regions can be merged, resulting in a more accurate and robust segmentation image.

[0133] To verify the beneficial effects of the present invention, the inventors conducted experiments on images in the NYUD2 and SUN RGB-D datasets using the weighted rough membership clustering RGB-D image segmentation method based on composite distance according to an embodiment of the present invention. The various experimental conditions are as follows.

[0134] 1. Experimental conditions;

[0135] The experimental test environment is a HP computer with Windows 11 (64) as the operating system, which is configured with an AMD Ryzen77735H processor 3.20GHz and 16GB of memory. The experimental operations are performed on the MATLAB2022a platform.

[0136] 2. Test image;

[0137] NYUD2 dataset, SUN RGB-D dataset.

[0138] 3. Experimental content;

[0139] (1) The method of the embodiment of the present invention (hereinafter referred to as the present method) was evaluated with the hard C-means clustering algorithm (HCM), the fast robust fuzzy C-means clustering algorithm (FRFCM), the multi-layer clustering algorithm (MLC), the joint color-space-direction clustering and region merging algorithm (JCSD-RM), the chaotic kbest gravitational search algorithm (cKGSA-FC), and the random Henry gas solubility optimization algorithm (RHGSO-FC) under the same number of clusters. The visual clustering effect is Figure 5-Figure 20 As shown in Table 2 and Table 4, respectively, the evaluation index analysis results of randomly sampled images on the NYUD2 and SUN RGB-D datasets are presented, and Table 3 statistics the average performance of this method on the NYUD2 dataset. In terms of experimental parameter settings, the configurations for different datasets are as follows: in the NYUD2 dataset, the number of clusters is set to 13, the neighborhood size is 5×5, and the weight parameter q=2 is used to balance the distance measurement of the feature space; in the SUN RGB-D dataset, the number of clusters is set to 7, and the same 5×5 neighborhood size and weight parameter q=2 are used to ensure consistency comparison of the algorithm in different scenarios. In addition, in order to ensure the stability and convergence efficiency of the algorithm, the convergence threshold ε=0.0001 is uniformly set, and the maximum number of iterations iter max = 50. The evaluation indicators include boundary displacement error (BDE), probability Rand Index (PRI), pixel accuracy (Pixel Accuracy) and normalized mutual information (NMI). For the evaluation indicators, the lower the BDE value and the higher the PRI, ACC and NMI values, the better the segmentation effect of the method.

[0140] Table 2

[0141]

[0142]

[0143] Table 3

[0144]

[0145] Experimental results show that this method exhibits good segmentation performance on the NYUD2 indoor scene dataset. Figure 5-12As shown in the figure, compared with existing methods, our method demonstrates significant improvements in the following aspects: First, it demonstrates higher boundary accuracy in wall segmentation tasks, effectively reducing the mis-segmentation rate in shadow areas; second, it maintains good segmentation accuracy for small objects in the scene. These advantages fully demonstrate that our method has greater robustness and higher segmentation accuracy when dealing with complex indoor scenes.

[0146] As can be seen from Tables 2 and 3, the segmentation performance of the method of the present invention on NYUD2 is significantly better than that of the existing methods. As can be seen from the data in Table 3, the method of the present invention has achieved the optimal values in the three indicators of BDE, PRI, and ACC, which shows that the method has higher segmentation accuracy and more accurate boundary recognition ability. The experimental results show that the method can effectively reduce the false segmentation rate and the missed segmentation rate, and show excellent stability and accuracy when processing complex scenes and images with rich details. Although the JCSD-RM method performs best in the NMI indicator, it shows that the method has good performance in similarity evaluation. However, the gap between the NMI value of this method and the JCSD-RM method is small, and it performs well in indicators such as BDE, PRI and ACC, with more outstanding comprehensive performance, and finally achieves the best segmentation effect.

[0147] Table 4

[0148]

[0149] Figure 13-Figure 20 A visual comparison of the image segmentation results of different algorithms on the SUN RGB-D dataset is presented. It can be seen that the proposed method exhibits the best visual segmentation performance, accurately identifying stools and planar objects while maintaining clear boundary demarcation and rich detail features. At the same time, this method significantly reduces segmentation noise and effectively suppresses missegmentation and the generation of redundant regions, further verifying its robustness and accuracy in complex scenes. Through precise boundary detection and region segmentation, the proposed method excels in processing complex backgrounds and details, avoiding common artifacts and unnecessary region divisions, and ultimately achieving clearer and more accurate regions in the segmented image. Table 4 shows the comparison results of all the compared algorithms on the three metrics of PRI, ACC, and NMI for the above images. The proposed method achieves the best performance, fully verifying its superiority in the multi-dimensional evaluation system. Specifically, the proposed method can effectively distinguish different objects, comprehensively and accurately extract key information from the image, and effectively improve overall segmentation accuracy. Although the BDE metric does not reach optimality, a comprehensive evaluation of the visual segmentation effect and overall metrics shows that the proposed method performs better than the other compared algorithms.

[0150] Figure 2This paper presents an analysis of the impact of feature weighting strategy on the clustering and segmentation quality of randomly sampled images from the NYUD2 dataset. This example uses the NMI indicator for quantitative evaluation. Figure 2 As can be seen from the results, the NMI values of the weighted algorithm are generally higher than those of the unweighted algorithm. For the NYUD2_367 image, the weighted algorithm's NMI value is nearly 0.7, a significant improvement over the unweighted algorithm's NMI value of 0.63. For the NYUD2_39 image, the weighted algorithm's NMI value is approximately 0.55, also higher than the unweighted algorithm's NMI value of 0.48. For the segmentation results of the remaining images, the weighted algorithm's NMI values all show some improvement over the unweighted algorithm's. Figure 3 This is an analysis of the impact of the composite distance metric proposed in this paper on the image clustering and segmentation quality of the NYUD2 dataset. Experimental results show that in multiple test images (such as NYUD2_396, NYUD2_420, etc.), the clustering effect of the composite distance metric is significantly better than the clustering result of the traditional Euclidean distance. Specifically, for Figure 3 In the comparison results, the maximum NMI using the composite distance metric reached 0.68, higher than the maximum NMI of 0.63 using the Euclidean distance. For eight randomly selected images from the NYUD2 dataset, the NMI using the composite distance metric was higher than the NMI using the traditional Euclidean distance method. Experimental data demonstrates that the composite distance metric can more effectively measure the similarity between RGB-D multimodal data, thereby improving clustering results. In summary, the method of the present invention combines the advantages of feature weighting and composite distance to effectively improve clustering performance.

[0151] Figure 4 The CPU time consumption of different algorithms in image segmentation tasks was compared. Among them, the cKGSA-FC and RHGSO-FC methods adopted the SAS framework segmentation technology based on five-seed graph aggregation. This solution not only has complex algorithm design, but also requires the adjustment of multiple parameters, resulting in a significant increase in computational complexity. Experimental data show that compared with the CKGSA-FC and RHGSO-FC methods, the method of the present invention has significant advantages in computational efficiency. Compared with the JCSD-RM method, its computation time is slightly lower, which is due to the lower algorithm complexity. Although the HCM, FRFCM and MLC methods have shorter computational times, their segmentation performance is poor. Therefore, the method of the present invention can achieve good segmentation effects while ensuring higher computational efficiency.

[0152] This embodiment further provides a weighted rough membership clustering RGB-D image segmentation system based on composite distance, including:

[0153] Initialization module, used to improve the traditional random initial class center selection method and select points in high-density areas as initial class centers;

[0154] A feature extraction module is used to extract image features of the RGB-D image, wherein the image features include color features, point cloud features, and normal features;

[0155] An optimization module is used to optimize the clustering algorithm based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm;

[0156] Among them, adaptive weighting includes adaptively weighting each feature dimension through local feature weighting method;

[0157] Among them, the composite distance measurement method includes calculating the similarity between the sample and the cluster center by integrating the composite distance measurement method of Euclidean distance and cosine distance;

[0158] The image segmentation module is used to cluster RGB-D images based on the weighted rough membership clustering algorithm, and merge the clustering results into regions to generate segmented images.

[0159] This embodiment further provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0160] This embodiment further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0161] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A weighted rough membership clustering RGB-D image segmentation method based on composite distance, comprising the following steps: Extracting image features of the RGB-D image, wherein the image features include color features, point cloud features, and normal features; The density-based method selects the class center and uses the high-density area as the initial class center; The clustering algorithm is optimized based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm. Adaptive weighting involves adaptively weighting each feature dimension through local feature weighting method; The composite distance measurement method includes calculating the similarity between samples and cluster centers by fusing Euclidean distance and cosine distance; The RGB-D image is clustered based on the weighted rough membership clustering algorithm, and the clustering results are merged to generate a segmented image.

2. The method according to claim 1, characterized in that When selecting the initial class center, the neighborhood density of each point is calculated and the high-density point is selected as the initial class center.

3. The method according to claim 1, characterized in that In adaptive weighting, different weights are assigned to color features, point cloud features, and normal features according to the importance of each feature in clustering, and the weights are dynamically adjusted according to the intra-cluster distance during the iterative process.

4. The method according to claim 1, wherein In the composite distance measurement method, Euclidean distance is used to calculate the similarity of color features and point cloud features, cosine distance is used to calculate the similarity of normal features, and different distance metrics are normalized by a linear normalization method.

5. The method according to claim 1, wherein The criterion function expression of the weighted rough membership clustering algorithm is: In the formula, n, M and C represent the number of data samples, the spatial dimension of the feature and the number of cluster centers respectively, and W cm Represents the feature weight of the mth dimension in the cth class, and q is an exponential parameter used to adjust the weight W cm The strength of the effect on the feature, For sample x i To cluster center b c Based on the rough membership of the grid neighborhood binary relation R, d f Represents the sample point x im To cluster center b cm The composite distance, x im represents the m-th dimension feature of the i-th sample, b cm Represents the m-th dimension feature of the c-th cluster center.

6. The method according to claim 1, characterized in that This method updates the rough membership through Lagrangian function and constraint conditions, and comprehensively considers the local structure and global characteristics of the data.

7. The method according to claim 1, characterized in that The region merging includes: treating each label region as a node, checking the candidate eligibility and merging eligibility of the node and verifying the merging consistency, and merging the nodes that meet the conditions until there are no more candidate nodes that can be merged.

8. A weighted rough membership clustering RGB-D image segmentation system based on composite distance, characterized by: include: Initialization module, used to improve the traditional random initial class center selection method and select points in high-density areas as initial class centers; A feature extraction module is used to extract image features of the RGB-D image, wherein the image features include color features, point cloud features, and normal features; An optimization module is used to optimize the clustering algorithm based on adaptive weighting and composite distance measurement methods to obtain a weighted rough membership clustering algorithm; Among them, adaptive weighting includes adaptively weighting each feature dimension through local feature weighting method; Among them, the composite distance measurement method includes calculating the similarity between samples and cluster centers by fusing the composite distance of Euclidean distance and cosine distance; The image segmentation module is used to cluster RGB-D images based on the weighted rough membership clustering algorithm, and merge the clustering results into regions to generate segmented images.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images