Picture similarity determination method and device, equipment, storage medium and program product
The radial basis function neural network improves image similarity determination accuracy by converting initial feature vectors into target vectors using a radial basis kernel function matrix, addressing the limitations of traditional methods.
Patent Information
- Application Number
- CN202510275995.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-10
- Publication Date
- 2025-06-24
AI Technical Summary
Existing image similarity determination methods, such as histogram, grayscale, and hash algorithms, suffer from low accuracy in determining image similarity.
Utilize a radial basis function (RBF) neural network to determine image similarity by converting initial feature vectors into target feature vectors using a radial basis kernel function matrix, and train the network with gradient descent to improve accuracy.
The RBF neural network approach effectively addresses the issues of ignoring spatial information and losing color information, enhancing the accuracy and efficiency of image similarity determination.
Smart Images
Figure CN120198690A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image processing technologies, and in particular, to a method, apparatus, device, storage medium, and program product for determining image similarity. Background Art
[0002] Determining image similarity plays an important role in the fields of computer vision and image processing. It can not only provide strong support for applications such as image recognition, classification, and retrieval, but also has wide application value in many practical scenarios, such as image retrieval, copyright protection, e-commerce, etc.
[0003] Currently, common methods for determining image similarity include histogram algorithms, grayscale image algorithms, hash algorithms, etc., but the accuracy of these methods for determining image similarity is relatively low. Summary of the Invention
[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, apparatus, device, storage medium, and program product for determining image similarity that can improve the accuracy of determining image similarity.
[0005] In a first aspect, this application provides a method for determining image similarity. The method includes:
[0006] Obtain initial feature vectors corresponding to at least two images to be processed respectively;
[0007] For each of the initial feature vectors, input the initial feature vector into a target radial basis neural network to determine a target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network;
[0008] Determine the similarity between two images to be processed in the group of images to be processed according to the target feature vectors corresponding to the group of images to be processed; the group of images to be processed includes any two images to be processed among the at least two images to be processed.
[0009] In one embodiment, the method further includes:
[0010] Determine the neuron centers of the hidden layer of the initial radial basis neural network based on a training sample set, and determine the width parameter of the radial basis kernel function of the initial radial basis neural network;
[0011] Input the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each of the neuron centers and the training sample set, and process the training sample set through the radial basis kernel function matrix to obtain intermediate feature vectors;
[0012] Update the parameters of the initial radial basis neural network based on the intermediate feature vector and the gradient descent algorithm to obtain the target radial basis neural network.
[0013] In one embodiment, constructing the radial basis kernel function matrix based on each neuron center and the training sample set includes:
[0014] For each training sample in the training sample set, determine the first distance between the training sample and each neuron center;
[0015] For each first distance, determine the radial basis kernel function value corresponding to the first distance according to the first distance and the radial basis kernel function of the initial radial basis neural network;
[0016] Construct the radial basis kernel function matrix based on each training sample, each neuron center, and each radial basis kernel function value.
[0017] In one embodiment, determining the neuron centers of the hidden layer of the initial radial basis neural network based on the training sample set includes:
[0018] Determine a preset number of initial neuron centers;
[0019] For each training sample in the training sample set, determine the second distance between the training sample and each initial neuron center;
[0020] Divide the training sample into the cluster where the initial neuron center corresponding to the minimum second distance is located;
[0021] Based on the training samples in each cluster, determine the intermediate neuron centers, and use the intermediate neuron centers as the new initial neuron centers, and re - execute the step of determining the intermediate neuron centers until the convergence condition is met, and determine the last determined intermediate neuron centers as the neuron centers of the hidden layer of the initial radial basis neural network.
[0022] In one embodiment, determining the similarity between two to - be - processed pictures in the to - be - processed picture group according to the target feature vector corresponding to the to - be - processed picture group includes:
[0023] Determine the Euclidean distance between the target feature vectors corresponding to two to - be - processed pictures in the to - be - processed picture group;
[0024] Determine the Euclidean distance as the similarity between two to - be - processed pictures in the to - be - processed picture group.
[0025] In one embodiment, respectively obtaining the initial feature vectors corresponding to at least two to - be - processed pictures includes:
[0026] Perform grayscale processing and normalization processing on each of the to-be-processed images to obtain the target images corresponding to each of the to-be-processed images;
[0027] Based on the Scale-Invariant Feature Transform (SIFT) matching algorithm and each of the target images, determine the initial feature vectors corresponding to each of the to-be-processed images.
[0028] In a second aspect, the present application also provides an apparatus for determining image similarity. The apparatus includes:
[0029] An acquisition module, configured to respectively acquire the initial feature vectors corresponding to at least two to-be-processed images;
[0030] A first determination module, configured to input, for each of the initial feature vectors, the initial feature vector into a target radial basis neural network, so as to determine the target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network;
[0031] A second determination module, configured to determine the similarity between two to-be-processed images in the to-be-processed image group according to the target feature vectors corresponding to the to-be-processed image group; the to-be-processed image group includes any two to-be-processed images among the at least two to-be-processed images.
[0032] In a third aspect, the present application also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of any of the above methods are implemented.
[0033] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.
[0034] In a fifth aspect, the present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.
[0035] For the above method, apparatus, device, storage medium, and program product for determining image similarity, by respectively acquiring the initial feature vectors corresponding to at least two to-be-processed images, inputting, for each initial feature vector, the initial feature vector into a target radial basis neural network to determine the target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network, and then determining the similarity between two to-be-processed images in the to-be-processed image group according to the target feature vectors corresponding to the to-be-processed image group. Since the radial basis neural network has characteristics such as simple structure, concise training, and fast learning convergence speed, determining the similarity between to-be-processed images based on the radial basis neural network can overcome problems existing in traditional methods for determining image similarity, such as ignoring spatial information and losing image color information, and improve the accuracy of the method for determining image similarity. Brief Description of the Drawings
[0036] Figure 1 is the internal structure diagram of a computer device provided by an embodiment of the present application;
[0037] Figure 2 is the schematic flowchart of a method for determining picture similarity provided by an embodiment of the present application;
[0038] Figure 3 is the schematic flowchart of a method for obtaining a target radial basis neural network provided by an embodiment of the present application;
[0039] Figure 4 is the schematic flowchart of a method for constructing a radial basis kernel function matrix provided by an embodiment of the present application;
[0040] Figure 5 is the schematic flowchart of a method for determining neuron centers provided by an embodiment of the present application;
[0041] Figure 6 is the schematic flowchart of a method for determining similarity provided by an embodiment of the present application;
[0042] Figure 7 is the schematic flowchart of a method for obtaining an initial feature vector provided by an embodiment of the present application;
[0043] Figure 8 is the schematic flowchart of a method for calculating picture similarity based on a radial basis kernel function network provided by an embodiment of the present application;
[0044] Figure 9 is the structural block diagram of a device for determining picture similarity provided by an embodiment of the present application. Detailed Description of the Embodiments
[0045] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0046] Determining picture similarity has an important position in the fields of computer vision and image processing. It can not only provide strong support for applications such as image recognition, classification, and retrieval, but also has wide application value in many actual scenarios, such as image retrieval, copyright protection, e-commerce, etc.
[0047] Currently, common methods for determining picture similarity include histogram algorithms, grayscale algorithms, hash algorithms, etc., but the accuracy of these methods for determining picture similarity is relatively low.
[0048] The method for determining the similarity of pictures provided by the embodiments of the present application can be applied to, for example, Figure 1 the application environment shown. Figure 1 Figure 1 is an internal structure diagram of a computer device provided by the embodiments of the present application. The computer device can be a server, and its internal structure diagram can be as shown in Figure 1
[0049] Those skilled in the art can understand that Figure 1 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.
[0050] In one embodiment, as shown in Figure 2 Figure 2 is a schematic flowchart of a method for determining the similarity of pictures provided by the embodiments of the present application. The method can be applied to the computer device in Figure 1 and includes the following steps:
[0051] S201, respectively obtain the initial feature vectors corresponding to at least two pictures to be processed.
[0052] In the embodiments of the present application, the similarity between two pictures to be processed can be determined, or the similarity between each pair of the multiple pictures to be processed can be determined. Therefore, at least two pictures to be processed can be obtained.
[0053] Exemplarily, assuming there are two pictures to be processed, A and B, the similarity between A and B can be determined; if there are three pictures to be processed, A, B, and C, the similarities between A and B, A and C, and B and C can be determined.
[0054] In one embodiment, the pictures to be processed obtained can be preprocessed first, for example, the pictures to be processed are grayscaled and normalized to obtain the target image. Then, based on the feature extraction algorithm, the target image is feature-extracted to obtain the initial feature vector corresponding to the picture to be processed.
[0055] Optionally, the feature extraction algorithm may, for example, include a Scale-Invariant Feature Transform (SIFT) matching algorithm.
[0056] S202. For each initial feature vector, input the initial feature vector into the target radial basis neural network to determine the target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network.
[0057] In the embodiments of the present application, each initial feature vector may be used as the input of the target radial basis neural network, so as to obtain the target feature vector corresponding to each initial feature vector through the target radial basis neural network.
[0058] In one embodiment, the initial radial basis neural network may be trained based on a training sample set by using forward propagation and backpropagation to construct a radial basis kernel function matrix and determine the target radial basis neural network.
[0059] Optionally, during the process of training the initial radial basis neural network, the neuron centers may be determined first based on the training sample set and the K-means algorithm, and the width parameter of the initial radial basis neural network may be determined according to the neuron centers and the training sample set. Then, the initial radial basis neural network is trained based on the training sample set by using forward propagation and backpropagation to obtain the target radial basis neural network.
[0060] In the embodiments of the present application, during the forward propagation process, each training sample in the training sample set may be used as a data point first, and then the Euclidean distances between each data point and each neuron center are calculated respectively. The Euclidean distances are substituted into the radial basis kernel function to obtain the radial basis kernel function values corresponding to the Euclidean distances. Then, a similarity matrix between each data point and each neuron center, that is, the radial basis kernel function matrix, is constructed based on the radial basis kernel function values.
[0061] S203. Determine the similarity between two to-be-processed pictures in the to-be-processed picture group according to the target feature vectors corresponding to the to-be-processed picture group.
[0062] Wherein, the to-be-processed picture group includes any two to-be-processed pictures among at least two to-be-processed pictures.
[0063] In one embodiment, several to-be-processed picture groups may be determined according to at least two to-be-processed pictures obtained, and each to-be-processed picture group includes any two to-be-processed pictures among the at least two to-be-processed pictures obtained.
[0064] Exemplarily, assuming there are two pictures A and B to be processed, a picture group to be processed can be determined, and this picture group to be processed includes A and B; if there are three pictures A, B, and C to be processed, three picture groups to be processed can be determined, and the three picture groups to be processed include A and B, A and C, and B and C respectively.
[0065] Optionally, the Euclidean distance between the target feature vectors corresponding to two pictures to be processed in the picture group to be processed can be calculated, and then the calculated Euclidean distance can be used as the similarity between the two pictures to be processed in the picture group to be processed.
[0066] In the embodiments of the present application, by respectively obtaining the initial feature vectors corresponding to at least two pictures to be processed, for each initial feature vector, the initial feature vector is input into the target radial basis neural network to determine the target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network, and then according to the target feature vectors corresponding to the picture group to be processed, the similarity between two pictures to be processed in the picture group to be processed is determined. Since the radial basis neural network has characteristics such as simple structure, concise training, and fast learning convergence speed, determining the similarity between pictures to be processed based on the radial basis neural network can not only overcome the problems of the histogram algorithm ignoring spatial information and being sensitive to image rotation and scaling, but also does not lose the picture color information like the grayscale image algorithm, resulting in ineffective comparison for color-sensitive cases, and can also effectively overcome the possible false positives and false negatives in algorithms such as the similar hashing algorithm, thereby improving the accuracy and efficiency of the picture similarity determination method.
[0067] Refer to Figure 3 , Figure 3 is a schematic flowchart of a method for obtaining a target radial basis neural network provided by the embodiments of the present application. On the basis of the above embodiments, the method further includes the following steps:
[0068] S301, determining the neuron centers of the hidden layer of the initial radial basis neural network based on the training sample set, and determining the width parameter of the radial basis kernel function of the initial radial basis neural network.
[0069] In one embodiment, in the process of training the initial radial basis neural network based on the training sample set, the neuron centers can be first determined based on the training sample set and the K-means algorithm. Then, according to the determined neuron centers and the training sample set, the width parameter of the radial basis kernel function is determined.
[0070] Optionally, determining the neuron centers based on the training sample set and the K-means algorithm can be achieved through the following steps: Regarding each training sample in the training sample set as a data point, then randomly initialize K initial neuron centers, and assign each data point to the cluster where the nearest initial neuron center is located. After completing the assignment of each data point, the cluster center can be re-determined according to the mean value of all data points within each cluster, that is, the intermediate neuron centers are obtained. Then, regarding each intermediate neuron center as a new initial neuron center, repeat the above steps for determining the intermediate neuron centers until the neuron centers no longer change, and determine the finally unchanged neuron centers as the neuron centers of the hidden layer of the initial radial basis neural network.
[0071] In one embodiment, determining the width parameter of the radial basis kernel function according to the determined neuron centers and the training sample set can be achieved in the following way: After determining the neuron centers based on the K-means algorithm, the average distance from the training samples within each cluster to the neuron centers can be calculated, and then a preset multiple of the average distance is determined as the width parameter of the radial basis kernel function.
[0072] Exemplarily, assuming that the average distance from the training samples in the first cluster to the neuron center is L and the preset multiple is 1.5 times, then 1.5L can be determined as the width parameter of the radial basis kernel function.
[0073] Optionally, the determination of the width parameter can be implemented based on the following code:
[0074] def calculate_sigma(X, centers, kmeans_labels):
[0075] sigma = []
[0076] for i in range(len(centers)):
[0077] distances = [euclidean_distance(X[j], centers[i]) for j in range(len(X)) if kmeans_labels[j] == i]
[0078] sigma.append(np.mean(distances))
[0079] return np.array(sigma)
[0080] S302. Input the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each neuron center and the training sample set, and process the training sample set through the radial basis kernel function matrix to obtain an intermediate feature vector.
[0081] In the embodiments of the present application, the initial radial basis neural network can be trained based on the training sample set using forward propagation and backpropagation. During the forward propagation process, a radial basis kernel function matrix can be constructed first based on the input training sample set, and then the intermediate feature vector corresponding to the training sample set can be determined based on the constructed radial basis kernel function matrix.
[0082] Optionally, during the forward propagation process, each training sample in the training sample set can be used as a data point first, and then the Euclidean distance between each data point and each neuron center can be calculated respectively. Substitute each Euclidean distance into the radial basis kernel function to obtain the radial basis kernel function value corresponding to each Euclidean distance, and then construct a similarity matrix between each data point and each neuron center based on each radial basis kernel function value, that is, the radial basis kernel function matrix.
[0083] S303. Update the parameters of the initial radial basis neural network based on the intermediate feature vector and the gradient descent algorithm to obtain the target radial basis neural network.
[0084] In one embodiment, the initial radial basis neural network includes an input layer, a hidden layer, and an output layer. During the above forward propagation process, the output of the hidden layer and the output of the output layer (the output of the output layer is the intermediate feature vector) can be determined based on the input training sample set. Then, a suitable loss function is defined, the gradient of the loss function with respect to the weights of the initial radial basis neural network is calculated, and the weights of the initial radial basis neural network are updated along the opposite direction of the gradient to obtain the target radial basis neural network.
[0085] Exemplarily, the following code can be used to calculate the output layer weights based on the gradient descent method:
[0086] import numpy as np
[0087] # Assume that the hidden layer output phi and the target output target are already available, the learning rate is learning_rate, and the weights are initialized as w
[0088] learning_rate = 0.01
[0089] num_epochs = 100
[0090] n_samples = len(phi)
[0091] input_size = len(phi[0])
[0092] output_size = len(target[0])
[0093] w = np.random.rand(input_size, output_size)
[0094] for epoch in range(num_epochs):
[0095] output = np.dot(phi, w)
[0096] error = target - output
[0097] gradient = -2 * np.dot(phi.T, error) / n_samples
[0098] w = w - learning_rate * gradient
[0099] In the embodiments of the present application, the neuron centers of the hidden layer of the initial radial basis neural network are determined based on the training sample set, and the width parameter of the radial basis kernel function of the initial radial basis neural network is determined. Then, the training sample set is input into the initial radial basis neural network to construct a radial basis kernel function matrix based on each neuron center and the training sample set, and the training sample set is processed through the radial basis kernel function matrix to obtain an intermediate feature vector. Finally, the parameters of the initial radial basis neural network are updated based on the intermediate feature vector and the gradient descent algorithm to obtain the target radial basis neural network, so that the input data can be transformed into a new feature space through the constructed radial basis kernel function matrix, where the similarity and difference between data points are more obvious, and the non-linear structure of the data can be better captured, thereby improving the accuracy of subsequent similarity calculation.
[0100] Refer to Figure 4 , Figure 4 is a schematic flowchart of a method for constructing a radial basis kernel function matrix provided by an embodiment of the present application. This embodiment relates to a possible implementation manner of how to construct a radial basis kernel function matrix based on each neuron center and the training sample set. On the basis of the above embodiment, the above S302 includes the following steps:
[0101] S401, for each training sample in the training sample set, determine the first distance between the training sample and each neuron center.
[0102] In the embodiments of the present application, for each training sample in the training sample set, the Euclidean distance between the training sample and each neuron center can be calculated, and then the calculated Euclidean distances are determined as the first distances between the training sample and each neuron center.
[0103] S402. For each of the first distances, according to the first distance and the radial basis kernel function of the initial radial basis neural network, determine the radial basis kernel function value corresponding to the first distance.
[0104] In one embodiment, the radial basis kernel function can be expressed as: . Wherein, is the input data point, is the neuron center, is the width parameter, is the Euclidean distance. Each of the first distances can be substituted into the radial basis kernel function to obtain the radial basis kernel function value corresponding to each first distance.
[0105] S403. Based on each training sample, each neuron center, and each radial basis kernel function value, construct a radial basis kernel function matrix.
[0106] Optionally, after determining the radial basis kernel function values corresponding to each of the first distances, a similarity matrix between each data point and each neuron center, that is, a radial basis kernel function matrix, can be constructed based on each radial basis kernel function value.
[0107] In the embodiments of the present application, for each training sample in the training sample set, the first distance between the training sample and each neuron center is determined. For each of the first distances, according to the first distance and the radial basis kernel function of the initial radial basis neural network, the radial basis kernel function value corresponding to the first distance is determined. Based on each training sample, each neuron center, and each radial basis kernel function value, a radial basis kernel function matrix is constructed, so that the input data can be transformed into a new feature space, making the similarity and difference between data points more obvious, and further improving the accuracy of subsequent similarity calculation.
[0108] Refer to Figure 5 , Figure 5 is a schematic flowchart of a method for determining neuron centers provided by an embodiment of the present application. This embodiment relates to a possible implementation manner of how to determine the neuron centers of the hidden layer of the initial radial basis neural network based on a training sample set. On the basis of the above embodiment, the above S301 includes the following steps:
[0109] S501. Determine a preset number of initial neuron centers.
[0110] In one embodiment, each training sample in the training sample set can be used as a data point, and then K initial neuron centers are randomly initialized. Wherein, K is the preset number.
[0111] S502. For each training sample in the training sample set, determine the second distance between the training sample and each initial neuron center.
[0112] In the embodiments of the present application, for each data point, the Euclidean distance between the data point and each initial neuron center can be calculated respectively, and then each Euclidean distance is determined as the second distance between the training sample and each initial neuron center.
[0113] S503. Divide the training sample into the cluster where the initial neuron center corresponding to the minimum second distance is located.
[0114] Exemplarily, for each data point, the corresponding second distances of the data point can be sorted, and then the minimum second distance is determined as the target distance, and the data point is divided into the cluster where the initial neuron center corresponding to the target distance is located. Among them, one neuron center corresponds to one cluster.
[0115] S504. Based on the training samples in each cluster, determine the intermediate neuron center, and use the intermediate neuron center as the new initial neuron center, and re - execute the step of determining the intermediate neuron center until the convergence condition is met, and determine the intermediate neuron center determined last time as the neuron center of the hidden layer of the initial radial basis neural network.
[0116] Optionally, after the allocation of each data point is completed, the cluster center can be re - determined according to the mean value of all data points in each cluster, that is, the intermediate neuron center is obtained. Then, each intermediate neuron center is used as the new initial neuron center, and the above - mentioned step of determining the intermediate neuron center is repeated until the neuron center no longer changes, and the finally unchanged neuron centers are determined as the neuron centers of the hidden layer of the initial radial basis neural network.
[0117] Exemplarily, the determination of the neuron center can be implemented based on the following code:
[0118] import numpy as np
[0119] # Calculate the Euclidean distance
[0120] def euclidean_distance(x1, x2):
[0121] return np.sqrt(np.sum((x1 - x2) ** 2))
[0122] # K - Means algorithm
[0123] def kmeans_centers(X, K, max_iterations=100):
[0124] n, m = X.shape
[0125] centers = X[np.random.choice(n, K, replace=False)]
[0126] for _ in range(max_iterations):
[0127] clusters = [[] for _ in range(K)]
[0128] for x in X:
[0129] distances = [euclidean_distance(x, center) for center in centers]
[0130] cluster_index = np.argmin(distances)
[0131] clusters[cluster_index].append(x)
[0132] new_centers = np.array([np.mean(cluster, axis=0) if cluster else centers[i] for i, cluster in enumerate(clusters)])
[0133] if np.allclose(centers, new_centers):
[0134] break
[0135] centers = new_centers
[0136] return centers
[0137] In the embodiments of the present application, a preset number of initial neuron centers are determined. For each training sample in the training sample set, the second distance between the training sample and each initial neuron center is determined, and the training sample is assigned to the cluster where the initial neuron center corresponding to the minimum second distance is located. Based on the training samples within each cluster, intermediate neuron centers are determined, and the intermediate neuron centers are used as new initial neuron centers. The step of determining the intermediate neuron centers is repeated until the convergence condition is met. The intermediate neuron centers determined last time are determined as the neuron centers of the hidden layer of the initial radial basis neural network, so as to realize determining the neuron centers of the hidden layer of the initial radial basis neural network based on the K-means algorithm, improve the accuracy of determining the neuron centers, and thus be able to better distinguish different feature regions of the input data, and improve the generalization ability and performance of the network.
[0138] Referring to Figure 6 , Figure 6 is a schematic flowchart of a similarity determination method provided by an embodiment of the present application. This embodiment relates to a possible implementation manner of how to determine the similarity between two to-be-processed pictures in a to-be-processed picture group according to the target feature vectors corresponding to the to-be-processed picture group. On the basis of the above embodiment, the above S203 includes the following steps:
[0139] S601, determine the Euclidean distance between the target feature vectors corresponding to two to-be-processed pictures in the to-be-processed picture group.
[0140] In one embodiment, a plurality of to-be-processed picture groups can be determined according to at least two to-be-processed pictures obtained, and each to-be-processed picture group includes any two to-be-processed pictures among the at least two to-be-processed pictures obtained.
[0141] Exemplarily, assuming there are two to-be-processed pictures A and B, then a to-be-processed picture group can be determined, and this to-be-processed picture group includes A and B. Then, calculate the Euclidean distance between the target feature vector corresponding to A and the target feature vector corresponding to B; if there are three to-be-processed pictures A, B, and C, then three to-be-processed picture groups can be determined. The three to-be-processed picture groups respectively include A and B, A and C, and B and C. That is to say, it is necessary to calculate the Euclidean distance between the target feature vector corresponding to A and the target feature vector corresponding to B, the Euclidean distance between the target feature vector corresponding to A and the target feature vector corresponding to C, and the Euclidean distance between the target feature vector corresponding to B and the target feature vector corresponding to C respectively.
[0142] S602, determine the Euclidean distance as the similarity between two to-be-processed pictures in the to-be-processed picture group.
[0143] Optionally, the calculated Euclidean distance can be directly used as the similarity between two to-be-processed pictures in the to-be-processed picture group.
[0144] In the embodiments of the present application, the Euclidean distance between the target feature vectors corresponding to two to-be-processed images in the to-be-processed image group is determined, and the Euclidean distance is determined as the similarity between the two to-be-processed images in the to-be-processed image group. Since the target feature vectors in the embodiments of the present application are determined based on a radial basis neural network, the similarity between images calculated based on the target feature vectors can overcome problems existing in traditional image similarity determination methods, such as ignoring spatial information and losing image color information, and improve the accuracy of the image similarity determination method.
[0145] Refer to Figure 7 , Figure 7 is a schematic flowchart of an initial feature vector acquisition method provided by an embodiment of the present application. This embodiment relates to a possible implementation manner of how to respectively acquire the initial feature vectors corresponding to at least two to-be-processed images. On the basis of the above embodiment, the above S201 includes the following steps:
[0146] S701, perform grayscale processing and normalization processing on each to-be-processed image to obtain the target image corresponding to each to-be-processed image.
[0147] Optionally, grayscale processing can be performed based on each to-be-processed image to process the image into a grayscale image to reduce the complexity of later calculations. And normalization processing is performed on each to-be-processed image to scale the to-be-processed image to a specified size, and the size of the scaled to-be-processed image is determined by the information amount and complexity of the to-be-processed image.
[0148] Exemplarily, the preprocessing of the to-be-processed image can be implemented based on the following code:
[0149] / / Read the image
[0150] Mat img = imread("test.png", IMREAD_COLOR);
[0151] if (img.empty())
[0152] {
[0153] cerr << "Failed to load the image. Please check if the file path is correct." << endl;
[0154] return -1;
[0155] }
[0156] / / Define the target matrix
[0157] Mat gray;
[0158] cvtColor(img, gray, COLOR_BGR2GRAY); / / Convert to grayscale image
[0159] / / Define the target matrix
[0160] Mat imgNormalized;
[0161] / / Set the normalization parameters
[0162] double alpha = 1; / / Normalized norm value
[0163] double beta = 0; / / Not used during norm normalization
[0164] int normType = NORM_L2; / / L2 norm normalization
[0165] / / Apply normalization
[0166] normalize(gray, imgNormalized, alpha, beta, normType);
[0167] / / Display and output the original image, grayscale image, and normalized image
[0168] namedWindow("Source Image", WINDOW_NORMAL);
[0169] imshow("Source Image", img);
[0170] namedWindow("Grayscale Image", WINDOW_NORMAL);
[0171] imshow("Grayscale Image", gray);
[0172] namedWindow("Normalized Image", WINDOW_NORMAL);
[0173] imshow("Normalized Image", imgNormalized);
[0174] S702. Based on the scale-invariant feature transform matching algorithm and each target image, determine the initial feature vectors corresponding to each picture to be processed.
[0175] In one embodiment, the Scale-Invariant Feature Transform (SIFT) matching algorithm can be used to extract features from each target image to obtain the initial feature vectors corresponding to each image to be processed.
[0176] Exemplarily, the acquisition of the initial feature vectors can be implemented based on the following code:
[0177] import cv2
[0178] # Read the image
[0179] image = cv2.imread('your_image.jpg', cv2.IMREAD_GRAYSCALE)
[0180] # Initialize the SIFT detector (note: in some OpenCV versions, an additional module may need to be installed)
[0181] sift = cv2.SIFT_create()
[0182] # Detect keypoints and compute descriptors
[0183] keypoints, descriptors = sift.detectAndCompute(image, None)
[0184] # If you need to convert the descriptors to a single feature vector (usually not recommended as spatial information will be lost)
[0185] # You can flatten the descriptor matrix
[0186] if descriptors is not None:
[0187] feature_vector = descriptors.flatten()
[0188] else:
[0189] feature_vector = None # If no keypoints are detected, there are no descriptors
[0190] # Print the shape of the feature vector (if it exists)
[0191] if feature_vector is not None:
[0192] print(f"Feature vector shape: {feature_vector.shape}")
[0193] else:
[0194] print("No keypoints detected, no feature vector available.")
[0195] In the embodiments of the present application, grayscale processing and normalization processing are performed on each picture to be processed to obtain the target image corresponding to each picture to be processed. Based on the scale-invariant feature transform matching algorithm and each target image, the initial feature vector corresponding to each picture to be processed is determined. Preprocessing the pictures to be processed improves the quality of the input data and reduces the interference of irrelevant information on the similarity calculation. And feature extraction of the pictures to be processed can better adapt to the subsequent process of determining the picture similarity based on the radial basis neural network, improving the efficiency and accuracy of the calculation.
[0196] Refer to Figure 8 , Figure 8 which is a schematic flow chart of a method for calculating picture similarity based on a radial basis kernel function network provided by an embodiment of the present application. The method includes the following steps:
[0197] S801, determining the neuron centers of the hidden layer of the initial radial basis neural network based on the training sample set and the K-means algorithm, and determining the width parameter of the radial basis kernel function of the initial radial basis neural network.
[0198] S802, inputting the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each neuron center and the training sample set, and updating the parameters of the initial radial basis neural network by the gradient descent algorithm to obtain the target radial basis neural network.
[0199] S803, performing grayscale processing and normalization processing on each picture to be processed to obtain the target image corresponding to each picture to be processed.
[0200] S804, determining the initial feature vector corresponding to each picture to be processed based on the scale-invariant feature transform matching algorithm and each target image.
[0201] S805, for each initial feature vector, inputting the initial feature vector into the target radial basis neural network to determine the target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network.
[0202] S806, determine the Euclidean distance between the target feature vectors corresponding to two to-be-processed images in the to-be-processed image group, so as to obtain the similarity between the two to-be-processed images in the to-be-processed image group.
[0203] It should be understood that although each step in the flowcharts involved in the above-described embodiments is sequentially shown according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limitation, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.
[0204] Based on the same inventive concept, an embodiment of the present application further provides an image similarity determination device for implementing the above-mentioned image similarity determination method. The implementation solutions provided by this device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the following image similarity determination device can refer to the limitations on the image similarity determination method in the above text, and will not be repeated here.
[0205] In one embodiment, as Figure 9 shown, Figure 9 FIG. is a structural block diagram of an image similarity determination device provided by an embodiment of the present application. The device 900 includes:
[0206] An acquisition module 901, configured to respectively acquire initial feature vectors corresponding to at least two to-be-processed images.
[0207] A first determination module 902, configured to input each initial feature vector into a target radial basis neural network for each initial feature vector, so as to determine a target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network.
[0208] A second determination module 903, configured to determine the similarity between two to-be-processed images in the to-be-processed image group according to the target feature vectors corresponding to the to-be-processed image group; the to-be-processed image group includes any two to-be-processed images among at least two to-be-processed images.
[0209] In one of the embodiments, the device 900 further includes:
[0210] A third determination module, configured to determine neuron centers of a hidden layer of an initial radial basis neural network based on a training sample set, and determine a width parameter of a radial basis kernel function of the initial radial basis neural network.
[0211] A construction module, configured to input the training sample set into the initial radial basis neural network, construct a radial basis kernel function matrix based on each neuron center and the training sample set, and process the training sample set through the radial basis kernel function matrix to obtain an intermediate feature vector.
[0212] An update module, configured to update parameters of the initial radial basis neural network based on the intermediate feature vector and a gradient descent algorithm to obtain a target radial basis neural network.
[0213] In one embodiment, the construction module includes:
[0214] A first determination unit, configured to determine a first distance between each training sample in the training sample set and each neuron center.
[0215] A second determination unit, configured to determine a radial basis kernel function value corresponding to the first distance according to the first distance and the radial basis kernel function of the initial radial basis neural network for each first distance.
[0216] A construction unit, configured to construct a radial basis kernel function matrix based on each training sample, each neuron center, and each radial basis kernel function value.
[0217] In one embodiment, the third determination module includes:
[0218] A third determination unit, configured to determine a preset number of initial neuron centers.
[0219] A fourth determination unit, configured to determine a second distance between each training sample in the training sample set and each initial neuron center.
[0220] A partitioning unit, configured to partition the training samples into the cluster where the initial neuron center corresponding to the minimum second distance is located.
[0221] A fifth determination unit, configured to determine intermediate neuron centers based on the training samples within each cluster, use the intermediate neuron centers as new initial neuron centers, and re-execute the step of determining the intermediate neuron centers until a convergence condition is met, and determine the intermediate neuron centers determined last time as the neuron centers of the hidden layer of the initial radial basis neural network.
[0222] In one embodiment, the second determination module 903 includes:
[0223] A sixth determination unit, configured to determine an Euclidean distance between target feature vectors corresponding to two to-be-processed pictures in a to-be-processed picture group.
[0224] A seventh determination unit, configured to determine the Euclidean distance as the similarity between two to-be-processed pictures in a to-be-processed picture group.
[0225] In one embodiment, the obtaining module 901 includes:
[0226] A picture processing unit, configured to perform grayscale processing and normalization processing on each to-be-processed picture to obtain a target image corresponding to each to-be-processed picture.
[0227] An eighth determination unit, configured to determine an initial feature vector corresponding to each to-be-processed picture based on the scale-invariant feature transform matching algorithm and each target image.
[0228] Each module in the above picture similarity determination device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor in the computer device in the form of hardware or be independent of the processor, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0229] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the following steps are implemented:
[0230] Obtain initial feature vectors corresponding to at least two to-be-processed pictures respectively;
[0231] For each initial feature vector, input the initial feature vector into a target radial basis neural network to determine a target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network;
[0232] Determine the similarity between two to-be-processed pictures in the to-be-processed picture group according to the target feature vectors corresponding to the to-be-processed picture group; the to-be-processed picture group includes any two to-be-processed pictures among at least two to-be-processed pictures.
[0233] In one embodiment, when the processor executes the computer program, the following steps are further implemented:
[0234] Determine the neuron centers of the hidden layer of the initial radial basis neural network based on a training sample set, and determine the width parameter of the radial basis kernel function of the initial radial basis neural network;
[0235] Input the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each neuron center and the training sample set, and process the training sample set through the radial basis kernel function matrix to obtain intermediate feature vectors;
[0236] Update the parameters of the initial radial basis neural network based on the intermediate feature vectors and the gradient descent algorithm to obtain the target radial basis neural network.
[0237] In one embodiment, when the processor executes the computer program, the following steps are further implemented:
[0238] For each training sample in the training sample set, determine the first distance between the training sample and each neuron center;
[0239] For each first distance, determine the radial basis kernel function value corresponding to the first distance according to the first distance and the radial basis kernel function of the initial radial basis neural network;
[0240] Based on each training sample, each neuron center, and each radial basis kernel function value, construct a radial basis kernel function matrix.
[0241] In one embodiment, when the processor executes the computer program, the following steps are further implemented:
[0242] Determine a preset number of initial neuron centers;
[0243] For each training sample in the training sample set, determine the second distance between the training sample and each initial neuron center;
[0244] Divide the training sample into the cluster where the initial neuron center corresponding to the minimum second distance is located;
[0245] Based on the training samples within each cluster, determine the intermediate neuron centers, and use the intermediate neuron centers as the new initial neuron centers, and re-execute the step of determining the intermediate neuron centers until the convergence condition is met, and determine the intermediate neuron centers determined last time as the neuron centers of the hidden layer of the initial radial basis neural network.
[0246] In one embodiment, when the processor executes the computer program, the following steps are further implemented:
[0247] Determine the Euclidean distance between the target feature vectors corresponding to two to-be-processed pictures in the to-be-processed picture group;
[0248] Determine the Euclidean distance as the similarity between two to-be-processed pictures in the to-be-processed picture group.
[0249] In one embodiment, when the processor executes the computer program, the following steps are further implemented:
[0250] Perform grayscale processing and normalization processing on each to-be-processed picture to obtain the target image corresponding to each to-be-processed picture;
[0251] Based on the scale-invariant feature transform matching algorithm and each target image, determine the initial feature vector corresponding to each to-be-processed picture.
[0252] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:
[0253] Obtain initial feature vectors corresponding to at least two pictures to be processed respectively;
[0254] For each initial feature vector, input the initial feature vector into a target radial basis neural network to determine a target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network;
[0255] Determine the similarity between two pictures to be processed in the picture group to be processed according to the target feature vectors corresponding to the picture group to be processed; the picture group to be processed includes any two pictures among at least two pictures to be processed.
[0256] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0257] Determine the neuron centers of the hidden layer of the initial radial basis neural network based on a training sample set, and determine the width parameter of the radial basis kernel function of the initial radial basis neural network;
[0258] Input the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each neuron center and the training sample set, and process the training sample set through the radial basis kernel function matrix to obtain intermediate feature vectors;
[0259] Update the parameters of the initial radial basis neural network based on the intermediate feature vectors and the gradient descent algorithm to obtain a target radial basis neural network.
[0260] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0261] For each training sample in the training sample set, determine the first distance between the training sample and each neuron center;
[0262] For each first distance, determine the radial basis kernel function value corresponding to the first distance according to the first distance and the radial basis kernel function of the initial radial basis neural network;
[0263] Construct a radial basis kernel function matrix based on each training sample, each neuron center and each radial basis kernel function value.
[0264] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0265] Determine a preset number of initial neuron centers;
[0266] For each training sample in the training sample set, determine the second distance between the training sample and each initial neuron center;
[0267] Divide the training sample into the cluster where the initial neuron center corresponding to the minimum second distance is located;
[0268] Based on the training samples within each cluster, determine the intermediate neuron centers, and use the intermediate neuron centers as the new initial neuron centers, and re - execute the step of determining the intermediate neuron centers until the convergence condition is met, and determine the intermediate neuron centers determined in the last time as the neuron centers of the hidden layer of the initial radial basis neural network.
[0269] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0270] Determine the Euclidean distance between the target feature vectors corresponding to two to - be - processed pictures in the to - be - processed picture group;
[0271] Determine the Euclidean distance as the similarity between two to - be - processed pictures in the to - be - processed picture group.
[0272] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0273] Perform grayscale processing and normalization processing on each to - be - processed picture to obtain the target image corresponding to each to - be - processed picture;
[0274] Based on the scale - invariant feature transform matching algorithm and each target image, determine the initial feature vector corresponding to each to - be - processed picture.
[0275] In one embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0276] Respectively obtain the initial feature vectors corresponding to at least two to - be - processed pictures;
[0277] For each initial feature vector, input the initial feature vector into the target radial basis neural network to determine the target feature vector corresponding to the initial feature vector based on the radial basis kernel function matrix of the target radial basis neural network;
[0278] According to the target feature vectors corresponding to the to - be - processed picture group, determine the similarity between two to - be - processed pictures in the to - be - processed picture group; the to - be - processed picture group includes any two to - be - processed pictures among at least two to - be - processed pictures.
[0279] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0280] Determine the neuron centers of the hidden layer of the initial radial basis neural network based on the training sample set, and determine the width parameter of the radial basis kernel function of the initial radial basis neural network;
[0281] Input the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each neuron center and the training sample set, and process the training sample set through the radial basis kernel function matrix to obtain an intermediate feature vector;
[0282] Update the parameters of the initial radial basis neural network based on the intermediate feature vector and the gradient descent algorithm to obtain the target radial basis neural network.
[0283] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0284] For each training sample in the training sample set, determine the first distance between the training sample and each neuron center;
[0285] For each first distance, determine the radial basis kernel function value corresponding to the first distance according to the first distance and the radial basis kernel function of the initial radial basis neural network;
[0286] Construct a radial basis kernel function matrix based on each training sample, each neuron center, and each radial basis kernel function value.
[0287] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0288] Determine a preset number of initial neuron centers;
[0289] For each training sample in the training sample set, determine the second distance between the training sample and each initial neuron center;
[0290] Divide the training sample into the cluster where the initial neuron center corresponding to the minimum second distance is located;
[0291] Based on the training samples within each cluster, determine the intermediate neuron center, and use the intermediate neuron center as the new initial neuron center, and re-execute the step of determining the intermediate neuron center until the convergence condition is met, and determine the intermediate neuron center determined last time as the neuron center of the hidden layer of the initial radial basis neural network.
[0292] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0293] Determine the Euclidean distance between the target feature vectors corresponding to two to-be-processed pictures in the to-be-processed picture group;
[0294] Determine the Euclidean distance as the similarity between the two to-be-processed pictures in the to-be-processed picture group.
[0295] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:
[0296] Perform grayscale processing and normalization processing on each picture to be processed to obtain a target image corresponding to each picture to be processed;
[0297] Based on the scale-invariant feature transform matching algorithm and each target image, determine an initial feature vector corresponding to each picture to be processed.
[0298] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.
[0299] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered as the scope recorded in this specification.
[0300] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.
Claims
1. A method for determining image similarity, characterized in that: The method comprises: Obtaining initial feature vectors corresponding to at least two images to be processed respectively; For each of the initial feature vectors, input the initial feature vector into a target radial basis neural network to determine a target feature vector corresponding to the initial feature vector based on a radial basis kernel function matrix of the target radial basis neural network; The similarity between two pictures to be processed in the picture group to be processed is determined according to the target feature vector corresponding to the picture group to be processed; the picture group to be processed includes any two pictures to be processed among the at least two pictures to be processed.
2. The method according to claim 1, characterized in that The method further comprises: Determine the neuron center of the hidden layer of the initial radial basis neural network based on the training sample set, and determine the width parameter of the radial basis kernel function of the initial radial basis neural network; Inputting the training sample set into the initial radial basis neural network to construct a radial basis kernel function matrix based on each of the neuron centers and the training sample set, and processing the training sample set by the radial basis kernel function matrix to obtain an intermediate feature vector; The parameters of the initial radial basis neural network are updated based on the intermediate feature vector and a gradient descent algorithm to obtain the target radial basis neural network.
3. The method according to claim 2, characterized in that The step of constructing a radial basis kernel function matrix based on each of the neuron centers and the training sample set includes: For each training sample in the training sample set, determining a first distance between the training sample and each neuron center; For each of the first distances, determining a radial basis kernel function value corresponding to the first distance according to the first distance and a radial basis kernel function of the initial radial basis neural network; The radial basis kernel function matrix is constructed based on each of the training samples, each of the neuron centers and each of the radial basis kernel function values.
4. The method according to claim 2, characterized in that: The step of determining the neuron center of the hidden layer of the initial radial basis neural network based on the training sample set includes: Determine a preset number of initial neuron centers; For each training sample in the training sample set, determining a second distance between the training sample and each of the initial neuron centers; Dividing the training samples into clusters where the initial neuron centers corresponding to the minimum second distance are located; Based on the training samples in each of the clusters, the intermediate neuron center is determined, and the intermediate neuron center is used as the new initial neuron center. The step of determining the intermediate neuron center is re-executed until the convergence condition is met, and the intermediate neuron center determined for the last time is determined as the neuron center of the hidden layer of the initial radial basis neural network.
5. The method according to any one of claims 1 to 4, characterized in that: The determining, according to the target feature vector corresponding to the to-be-processed picture group, the similarity between two to-be-processed pictures in the to-be-processed picture group comprises: Determine the Euclidean distance between target feature vectors corresponding to two pictures to be processed in the picture group to be processed; The Euclidean distance is determined as the similarity between two to-be-processed pictures in the to-be-processed picture group.
6. The method according to any one of claims 1 to 4, characterized in that: The step of respectively obtaining initial feature vectors corresponding to at least two images to be processed includes: Performing grayscale processing and normalization processing on each of the images to be processed to obtain a target image corresponding to each of the images to be processed; Based on the scale-invariant feature transformation matching algorithm and each of the target images, an initial feature vector corresponding to each of the images to be processed is determined.
7. A device for determining image similarity, characterized in that: The device comprises: An acquisition module, used to respectively acquire initial feature vectors corresponding to at least two images to be processed; A first determination module is used for inputting the initial feature vector into a target radial basis neural network for each of the initial feature vectors, so as to determine a target feature vector corresponding to the initial feature vector based on a radial basis kernel function matrix of the target radial basis neural network; The second determination module is used to determine the similarity between two pictures to be processed in the picture group to be processed according to the target feature vector corresponding to the picture group to be processed; the picture group to be processed includes any two pictures to be processed among the at least two pictures to be processed.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. 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 6 are implemented.
10. A computer program product, comprising a computer program, 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 6 are implemented.