Similarity adjacency graph construction method and device, computer device and storage medium

Abstract

The application discloses a similarity adjacency graph construction method and device, computer equipment and a storage medium. The method acquires a set of images to be processed, the set of images to be processed comprising a first image and N second images; calculates the image similarity between the first image and each second image, determines the first K second images with higher image similarity as third images; determines a target difference sequence value; determines a target segmentation turning point using double sliding windows; acquires M fourth images; constructs a similarity subgraph corresponding to the first image based on the first image and the M fourth images, iterates through the set of images to be processed, determines the similarity subgraphs of all first images, and constructs a similarity adjacency graph according to the similarity subgraphs of all first images. The method improves the integrity and accuracy of constructing the similarity subgraph corresponding to each first image, constructs a high-quality similarity adjacency graph, facilitates improving the effect and efficiency of face clustering, and reduces the difficulty of face clustering.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device, computer equipment and storage medium for constructing a similarity adjacency graph. Background Art

[0002] Graph neural network technology has developed rapidly in recent years, and many researchers have applied it to large-scale facial image clustering tasks. For example, using graph neural networks, all facial images of the same person can be clustered under the same profile ID based on facial features and related information, while facial images belonging to different individuals can be clustered under different profile IDs, achieving a one-person-one-profile approach. This approach primarily implements facial image clustering based on the similarity adjacency graph.

[0003] In the prior art, a K-nearest neighbor graph (K-NearestNeighborGraph) is typically constructed for all facial images based on feature similarity, and the entire adjacency graph is modified using some supervised information to achieve the facial image clustering task. The K-nearest neighbor graph is a type of similarity adjacency graph. However, due to the varying facial features of different individuals, the similarity threshold used when constructing the K-nearest neighbor graph can also vary significantly, resulting in a different number of facial images with high image similarity among the face images to be clustered. This makes it difficult to determine the K-nearest neighbor graphs for facial images of different individuals using a unified similarity threshold. Furthermore, using a single similarity threshold for facial image clustering does not produce good results. Furthermore, for supervised facial clustering tasks, clustering facial images of different individuals requires training different facial image clustering models, which takes a long time to train and significantly reduces the transfer performance of the facial image clustering models. Furthermore, the need to label training data increases the clustering cost, reducing the efficiency and increasing the difficulty of achieving facial image clustering. Therefore, how to construct a high-quality similarity adjacency graph to improve the subsequent face image clustering effect and efficiency, and reduce the clustering difficulty and clustering cost, is an urgent problem to be solved. Summary of the Invention

[0004] Embodiments of the present invention provide a similarity adjacency graph construction method, apparatus, computer equipment, and storage medium to solve the problem of how to construct a high-quality similarity adjacency graph.

[0005] A similarity adjacency graph construction method, comprising:

[0006] Acquire a set of images to be processed, the set of images to be processed comprising a first image and N second images, where N is greater than or equal to 2;

[0007] Calculating the image similarity between the first image and each of the second images, and determining the first K second images with higher image similarity as third images; wherein K ≥ 1;

[0008] determining a target differential sequence value corresponding to each of the K third images according to image similarities of the K third images;

[0009] Using a double sliding window to perform sliding detection on the target differential sequence value corresponding to each of the third images, and determine K target segmentation change points corresponding to the third images;

[0010] Cleaning the K third images according to target segmentation change points corresponding to the K third images to obtain M fourth images; where M ≥ 1;

[0011] Based on the first image and the M fourth images, a similarity subgraph corresponding to the first image is constructed, the set of images to be processed is traversed to determine the similarity subgraphs corresponding to all the first images, and a similarity adjacency graph is constructed based on the similarity subgraphs corresponding to all the first images.

[0012] Preferably, determining the target differential sequence value corresponding to each of the third images according to the image similarities of the K third images includes:

[0013] Normalizing the image similarities of the K third images to obtain normalized similarities corresponding to the K third images;

[0014] Performing descending processing on the normalized similarities corresponding to the K third images to obtain a target similarity sequence;

[0015] Perform differential processing on the target similarity sequence to obtain a target differential sequence value.

[0016] Preferably, the step of using a double sliding window to perform sliding detection on the target differential sequence value corresponding to each of the third images to determine K target segmentation change points corresponding to the third images includes:

[0017] Using a long sliding window and a short sliding window, perform sliding detection on the target differential sequence value corresponding to each of the third images to determine K original segmentation change points corresponding to the third images;

[0018] K target segmentation change points corresponding to the third image are determined according to the K original segmentation change points corresponding to the third image.

[0019] Preferably, the adopting of a long sliding window and a short sliding window to perform sliding detection on the target differential sequence value corresponding to each of the third images to determine the original segmentation change points corresponding to K of the third images includes:

[0020] Using a long sliding window and a short sliding window with a fixed intersection length, sliding detection is performed on the target difference sequence value corresponding to each of the third images in sequence according to the target sliding order, and a first numerical standard deviation within the long sliding window and a second numerical standard deviation within the short sliding window are obtained;

[0021] determining whether a segmentation change point evaluation condition is satisfied based on the first numerical standard deviation and the second numerical standard deviation;

[0022] If the segmentation change point evaluation condition is met, the midpoint of the short sliding window is determined as the original segmentation change point corresponding to the third image.

[0023] Preferably, judging whether a segmentation change point evaluation condition is satisfied based on the first numerical standard deviation and the second numerical standard deviation includes:

[0024] multiplying the first numerical standard deviation by a preset parameter to determine the target numerical standard deviation;

[0025] If the second numerical standard deviation is greater than the target numerical standard deviation, it is determined that the segmentation change point evaluation condition is met.

[0026] Preferably, determining K target segmentation change points corresponding to the third image based on the K original segmentation change points corresponding to the third image includes:

[0027] determining at least one continuous change point sequence from the target similarity sequences of the K third images according to the original segmentation change points corresponding to the K third images;

[0028] The continuous change point sequence with the largest continuous length is determined as the target change point sequence;

[0029] The starting point in the target change point sequence is determined as the target segmentation change points corresponding to the K third images.

[0030] Preferably, cleaning the K third images according to the target segmentation change points corresponding to the K third images to obtain M fourth images includes:

[0031] Determining a target position of the target segmentation change point in the target similarity sequences of the K third images;

[0032] The target position and all the third images before the target position are determined as M fourth images.

[0033] A similarity adjacency graph construction device, comprising:

[0034] An image set to be processed acquisition module is configured to acquire an image set to be processed, the image set to be processed comprising a first image and N second images; wherein N is greater than or equal to 2.

[0035] A third image determination module is configured to calculate image similarity between the first image and each of the second images, and determine the first K second images with higher image similarity as third images; wherein K is greater than or equal to 1.

[0036] A target difference sequence value determination module is configured to determine a target difference sequence value corresponding to each of the third images according to image similarity of the K third images.

[0037] A target segmentation change point determination module is configured to perform sliding detection on the target difference sequence value corresponding to each of the third images by using a double sliding window, and determine target segmentation change points corresponding to the K third images.

[0038] A fourth image acquisition module is configured to perform cleaning on the K third images according to the target segmentation change points corresponding to the K third images, and acquire M fourth images; wherein M is greater than or equal to 1.

[0039] A similarity adjacency graph construction module is configured to construct a similarity subgraph corresponding to the first image based on the first image and the M fourth images, traverse the image set to be processed, determine similarity subgraphs corresponding to all the first images, and construct a similarity adjacency graph according to the similarity subgraphs corresponding to all the first images.

[0040] A computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the similarity adjacency graph construction method when executing the computer program.

[0041] A computer readable storage medium stores a computer program, and the computer program is executable on a processor to implement the similarity adjacency graph construction method.

[0042] The above-mentioned similarity adjacency graph construction method, device, computer equipment and storage medium obtain a third image based on the first image and the second image, determine the target differential sequence value corresponding to each third image based on the image similarity of the K third images with the first image, use a double sliding window to perform sliding detection on the target differential sequence value corresponding to each third image, and determine the target segmentation change points corresponding to the K third images. The segmentation change points obtained by this method are more convenient, faster and more accurate, and can determine the target segmentation change points of the third images corresponding to different first images, and have strong transferability; according to the target segmentation change points corresponding to the K third images, the target segmentation change points of the K third images are determined. The image is cleaned to obtain M fourth images, and based on the first image and the M fourth images, a similarity subgraph corresponding to the first image is constructed to realize unsupervised construction of similarity subgraphs of all first images in the image set to be processed. This method is simple and convenient, and does not require the construction of multiple supervised facial feature models. The top M fourth images with high similarity to the first image can be determined, which saves training costs and has strong transferability. This method constructs a high-quality similarity adjacency graph by improving the completeness and accuracy of the similarity subgraph corresponding to each first image, which is convenient for improving the effect of subsequent face clustering, improving the efficiency of face clustering, and reducing the difficulty of face clustering. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 1 is a schematic diagram of an application environment of a similarity adjacency graph construction method according to an embodiment of the present invention;

[0045] Figure 2 is a flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0046] Figure 3 is another flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0047] Figure 4 is another flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0048] Figure 5 is another flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0049] Figure 6 is another flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0050] Figure 7 is another flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0051] Figure 8 is another flow chart of a method for constructing a similarity adjacency graph in one embodiment of the present invention;

[0052] Figure 9 is a schematic diagram of a similarity adjacency graph construction device according to an embodiment of the present invention;

[0053] Figure 10 is a schematic diagram of a computer device according to an embodiment of the present invention;

[0054] Figure 11 is the image similarity between two different first images and K third images in one embodiment of the present invention;

[0055] Figure 12 is a target differential sequence value of two different first images and K third images in one embodiment of the present invention;

[0056] Figure 13 is a schematic diagram of sliding detection of a double sliding window corresponding to a first image in one embodiment of the present invention;

[0057] Figure 14 1 is a diagram illustrating a corresponding relationship between a target similarity sequence and a target difference sequence value between a first image and a third image in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] The similarity adjacency graph construction method provided by the embodiment of the present invention can be applied as follows: Figure 1 Specifically, the similarity adjacency graph construction method is applied in a similarity adjacency graph construction system, which includes Figure 1The client and server shown in the figure communicate with each other via a network to achieve the purpose of constructing a high-quality similarity adjacency graph. The client, also known as the user end, refers to a program corresponding to the server that provides local services to clients. The client can be installed on, but is not limited to, various personal computers and laptops. The server can be implemented as an independent server or a server cluster consisting of multiple servers. In this embodiment, the client is used to collect facial images of different people, and the server is used to store and process the facial images of different people collected by the client to construct similarity adjacency graphs corresponding to the facial images of different people.

[0060] In one embodiment, if Figure 2 As shown, a similarity adjacency graph construction method is provided, which is applied in Figure 1 The server in the example is used as an example, and the steps are as follows:

[0061] S201: Acquire a set of images to be processed, where the set of images to be processed includes a first image and N second images, where N is greater than or equal to 2.

[0062] S202: Calculate the image similarity between the first image and each second image, and determine the first K second images with higher image similarity as the third image; wherein K ≥ 1;

[0063] S203: Determine a target differential sequence value corresponding to each third image based on the image similarities of the K third images;

[0064] S204: Using a double sliding window, perform sliding detection on the target differential sequence value corresponding to each third image to determine the target segmentation change points corresponding to K third images;

[0065] S205: Cleaning the K third images according to the target segmentation change points corresponding to the K third images to obtain M fourth images; where M ≥ 1;

[0066] S206: Based on the first image and the M fourth images, construct a similarity subgraph corresponding to the first image, traverse the image set to be processed, determine the similarity subgraphs corresponding to all first images, and construct a similarity adjacency graph based on the similarity subgraphs corresponding to all first images.

[0067] The "unprocessed image set" refers to a set of images that require processing, specifically a facial image set that stores facial images of different individuals. It is understood that the unprocessed image set may store facial images of at least one individual, or at least one facial image of the same individual. A "first image" refers to a facial image currently being processed in the unprocessed image set. A "second image" refers to all facial images in the unprocessed image set other than the "first image." It is understood that the "second image" is determined relative to the "first image."

[0068] As an example, in step S201, the server obtains facial images of different people from the client to form a set of images to be processed. In this example, the set of images to be processed includes a first image and N second images. The first image refers to the facial image that currently needs to be processed, and the second image is all N facial images in the set of images to be processed except the first image, where N≥2. It can be understood that the set of images to be processed is a collection of facial images of at least one person, and may include multiple facial images of the same person. That is, the set of images to be processed includes multiple facial images of multiple people, and these facial images may include multiple facial images of the same person and facial images of multiple different people. The set of images to be processed can be stored in the system database, so that the server can call it for processing at any time.

[0069] The third image refers to the second image having a higher image similarity with the first image. It can be understood that the third image is a facial image obtained by filtering the second image based on the image similarity with the first image.

[0070] As an example, in step S202, after obtaining the first image and N second images, the server calculates the image similarity between the first image and each second image, and determines the first K second images with higher image similarity as the third image. Wherein, K≥1. In this example, after determining the first image to be processed, the server uses N facial images other than the first image in the image set to be processed as the second image, determines the image similarity between each second image and the first image, obtains N image similarities, and sorts the obtained N image similarities in descending order to form a descending sequence corresponding to the N image similarities, selects the first K image similarities with higher image similarities in the descending sequence corresponding to the N image similarities, and uses the second images corresponding to the K image similarities as the third image. Wherein, K≥1. As Figure 11 As shown in the figure, in the image set to be processed, the image similarities of K third images corresponding to two different first images are calculated, where K is 256. In this example, the image similarity between the first image and each second image is determined, and the second images corresponding to the first K images with higher image similarity are selected as the third images, so that the target segmentation change points determined subsequently are more accurate.

[0071] The target differential sequence value refers to a differential sequence value obtained according to the image similarity of the third image. It is understandable that the target differential sequence value can be a first-order difference or a second-order difference.

[0072] As an example, in step S203, the server determines a target differential sequence value corresponding to each third image based on the image similarities of the K third images. In this example, after determining the K third images, the server associates each third image with the image similarity corresponding to each third image, and arranges the image similarities corresponding to the K third images in descending order to obtain a descending sequence corresponding to the image similarities corresponding to the K third images. Then, the server performs a first-order difference or a second-order difference on the image similarities of two adjacent third images in the image similarities to obtain a target differential sequence value corresponding to each third image. It is understandable that, in the descending sequence corresponding to the image similarities of the first image and each third image, the image similarity corresponding to the preceding third image is subtracted from the image similarity corresponding to the adjacent following third image to obtain a target differential sequence value corresponding to each third image. The image similarities corresponding to the K third images are traversed to obtain a target differential sequence value corresponding to the descending sequence formed by the image similarities of the K third images, with each third image corresponding to one target differential sequence value. Understandably, the target differential sequence values ​​of the third images correspond one-to-one to the descending sequence of image similarities, i.e., each degree of image similarity between the third image and the first image corresponds to a target differential sequence value. In this example, the target differential sequence value corresponding to each third image is determined based on the image similarities of the K third images, facilitating the determination of the target segmentation change point based on the target differential sequence value.

[0073] The dual sliding windows include a long sliding window and a short sliding window. The target segmentation change point refers to the similarity threshold point for image screening of the third image. Before this similarity threshold point, the third image has a higher image similarity with the first image, while after this similarity threshold point, the third image has a lower image similarity with the first image.

[0074] As an example, in step S204, the server uses a double sliding window to perform sliding detection on the target differential sequence value corresponding to each third image, and determines the target segmentation change points corresponding to the K third images. In this example, after determining the target differential sequence value corresponding to each third image, the server uses a double sliding window to perform sliding detection in a back-to-front order, traversing each target differential sequence value, and determining the target segmentation change points corresponding to the K third images based on the traversal results. It can be understood that, if Figure 12As shown, since the target differential sequence values ​​corresponding to the front third image have greater volatility, while the target differential sequence values ​​corresponding to the back third image have less volatility, using a dual sliding window to perform sliding detection on the target differential sequence values ​​in a back-to-front order makes it easier to quickly find the target segmentation change points. In this example, using a dual sliding window to determine the target segmentation change points corresponding to K third images is a faster, more convenient, and more accurate method for determining target segmentation change points. Furthermore, this method can determine the target segmentation change points of third images corresponding to different first images, and has strong transferability.

[0075] The fourth image is an image obtained by cleaning the third image according to the target segmentation change point. It is understandable that the fourth image can be the third image that meets the condition of the target segmentation change point.

[0076] As an example, in step S205, the server cleans the K third images based on the target segmentation change points corresponding to the K third images to obtain M fourth images. In this example, after obtaining the target segmentation change points corresponding to the third images, the server determines, from the sequence corresponding to the descending sequence of image similarities formed by the target difference sequence values ​​corresponding to each third image, the third image corresponding to the target segmentation change point and the third image corresponding to the target difference sequence value before the target segmentation change point as the fourth image. In this example, the third images are cleaned based on the target segmentation change points to obtain a fourth image with a higher image similarity to the first image, enabling a more accurate similarity subgraph to be constructed based on the fourth image.

[0077] The similarity subgraph refers to the adjacency graph between the first image in the image set to be processed and the fourth image corresponding to the first image. This similarity subgraph can be a K-nearest neighbor graph with the first image as the core. The similarity adjacency graph is the adjacency graph obtained by fusing the similarity subgraphs corresponding to all first images in the image set to be processed. The similarity subgraph corresponding to each first image is a local subgraph of the similarity adjacency graph.

[0078] As an example, in step S206, after the server obtains the fourth image corresponding to the first image, it constructs a high-quality similarity subgraph corresponding to the first image based on the first image and M fourth images. In the similarity subgraph, the M fourth images are respectively connected to the first image, and the similarity subgraph also includes the image similarity between each fourth image and the first image. After the similarity subgraph corresponding to the currently processed first image is constructed, the image set to be processed is traversed to determine the high-quality similarity subgraphs corresponding to all the first images in the image set to be processed, and the high-quality similarity subgraphs corresponding to all the first images are merged into a similarity adjacency graph corresponding to all the first images in the image set to be processed. Wherein, M≥1, that is, there is at least one fourth image. It can be understood that the similarity adjacency graph is a matrix formed by splicing and combining the similarity subgraphs corresponding to all the first images, and the similarity adjacency graph corresponding to each first image is a row or a column of the matrix. The similarity subgraphs corresponding to all the first images in the image set to be processed constitute the corresponding similarity adjacency graph. In this example, an unsupervised construction of a similarity subgraph of all first images in the image set to be processed is implemented. The similarity subgraph includes as many facial images of the same person in the image set to be processed as possible. This method is simple and convenient, and does not require the construction of multiple supervised facial feature models to determine the top M fourth images with high similarity to the first image, saving training costs and having strong portability. This method improves the completeness and accuracy of the similarity subgraph corresponding to each first image, thereby obtaining a high-quality similarity adjacency graph through fusion, which facilitates improving the effect of face clustering, improving the efficiency of face clustering, and reducing the difficulty of face clustering.

[0079] The similarity adjacency graph construction method provided in this embodiment obtains a third image based on the first image and the second image, determines the target differential sequence value corresponding to each third image based on the image similarity between the K third images and the first image, uses a double sliding window to perform sliding detection on the target differential sequence value corresponding to each third image, and determines the target segmentation change points corresponding to the K third images. The segmentation change points obtained by this method are more convenient, faster and more accurate, and can determine the target segmentation change points of the third images corresponding to different first images, and have strong portability; according to the target segmentation change points corresponding to the K third images, the K third images are cleared. Wash, obtain M fourth images, and construct a similarity subgraph corresponding to the first image based on the first image and the M fourth images, so as to realize unsupervised construction of similarity subgraphs of all first images in the image set to be processed. This method is simple and convenient, and does not require the construction of multiple supervised facial feature models. The top M fourth images with high similarity to the first image can be determined, which saves training costs and has strong transferability. This method constructs a high-quality similarity adjacency graph by improving the completeness and accuracy of the similarity subgraph corresponding to each first image, which is convenient for improving the effect of subsequent face clustering, improving the efficiency of face clustering, and reducing the difficulty of face clustering.

[0080] In one embodiment, if Figure 3 As shown, step S203, i.e., determining the target differential sequence value corresponding to each third image according to the image similarities of the K third images, includes:

[0081] S301: performing normalization processing on the image similarities of the K third images to obtain normalized similarities corresponding to the K third images;

[0082] S302: performing descending processing on the normalized similarities corresponding to the K third images to obtain a target similarity sequence;

[0083] S303: Perform differential processing on the target similarity sequence to obtain a target differential sequence value.

[0084] As an example, in step S301, after obtaining the image similarities of the K third images, the server normalizes the image similarities of the K third images to obtain the normalized similarities corresponding to the K third images. In this example, normalizing the image similarities of the K third images unifies the data calculation dimension or simplifies the data calculation range to ensure the feasibility of descending and differencing processing based on the normalized similarities, thereby helping to improve processing efficiency.

[0085] As an example, in step S302, the server performs descending processing on the normalized similarities corresponding to the K third images to obtain a target similarity sequence corresponding to the K third images. In this example, the normalized similarities corresponding to the K third images are performed in descending order to determine the target similarity sequence, which provides a guarantee for subsequent differential processing of data with different normalized similarities.

[0086] As an example, in step S303, after obtaining the target similarity sequence in descending order, the server performs differential processing on the two adjacent normalized similarities in the target similarity sequence to obtain the target differential sequence value. In this example, the server performs first-order differential processing or second-order differential processing on the previous normalized similarity and the adjacent next normalized similarity in the target similarity sequence, traverses the normalized similarity sequence corresponding to the third image, and obtains the target differential sequence value corresponding to the third image. Figure 12 As shown, Figure 11 The target difference sequence values ​​of the K third images corresponding to the two different first images are obtained. K is 256. In this example, the target difference sequence values ​​corresponding to the target similarity sequence of the third image are obtained to facilitate the subsequent determination of the target segmentation change points corresponding to the first image based on the target difference sequence values. This method of obtaining the target difference sequence values ​​is convenient, fast, and easy to implement.

[0087] The similarity adjacency graph construction method provided in this embodiment performs descending processing on the normalized similarities corresponding to K third images to determine the normalized similarities, determines a target similarity sequence based on the normalized similarities, performs differential processing on the target similarity sequence to obtain a target differential sequence value, and facilitates subsequent determination of the target segmentation change point corresponding to the first image based on the target differential sequence value. This method of obtaining the target differential sequence value is convenient, fast, and easy to implement.

[0088] In one embodiment, if Figure 4 As shown, step S204, i.e., using a double sliding window to perform sliding detection on the target differential sequence value corresponding to each third image, to determine the target segmentation change points corresponding to K third images, includes:

[0089] S401: Using a long sliding window and a short sliding window, perform sliding detection on the target differential sequence value corresponding to each third image to determine the original segmentation change points corresponding to K third images;

[0090] S402: Determine target segmentation change points corresponding to the K third images according to the original segmentation change points corresponding to the K third images.

[0091] The long sliding window and the short sliding window together constitute a double sliding window. The original segmentation change point refers to the segmentation change point determined after the double sliding window performs traversal sliding detection on the target differential sequence value corresponding to the third image.

[0092] As an example, in step S401, after obtaining the target differential sequence value corresponding to the third image, the server uses a long sliding window and a short sliding window to simultaneously perform sliding detection on the target differential sequence value corresponding to each third image. Each time it slides, a judgment is made to determine whether the target differential sequence value corresponding to the current window has an original segmentation change point. After traversing the target differential sequence values ​​corresponding to the K third images, the original segmentation change points corresponding to the K third images are determined. In this example, a long sliding window and a short sliding window are used as a dual sliding window to traverse the target differential sequence values ​​corresponding to the third image, detect the segmentation change points corresponding to the K third images, and use the detected segmentation change points as the original segmentation change points. This example uses dual sliding windows for simultaneous sliding detection, which has a stronger judgment ability than a single sliding window. Using dual sliding windows for sliding detection, the original segmentation change points obtained are more accurate. Understandably, there is at least one original segmentation change point.

[0093] As an example, in step S402, the server determines K target segmentation change points corresponding to the third images based on the original segmentation change points corresponding to the K third images. In this example, the server selects the original segmentation change point that best matches the actual situation based on the original segmentation change points as the target segmentation change point. This method is convenient and fast, and more realistic.

[0094] The similarity adjacency graph construction method provided in this embodiment uses a long sliding window and a short sliding window to perform sliding detection on the target differential sequence values ​​corresponding to each third image, determining the original segmentation change points corresponding to the K third images. Using dual sliding windows for simultaneous sliding detection offers stronger judgment capabilities than a single sliding window, and using dual sliding windows for sliding detection yields more accurate original segmentation change points. Based on the original segmentation change points corresponding to the K third images, the target segmentation change points corresponding to the K third images are determined. This method is convenient, fast, and more practical.

[0095] In one embodiment, if Figure 5 As shown, step S401, i.e., using a long sliding window and a short sliding window, performs sliding detection on the target differential sequence value corresponding to each third image to determine the original segmentation change points corresponding to K third images, includes:

[0096] S501: Using a long sliding window and a short sliding window with a fixed intersection length, perform sliding detection on the target difference sequence values ​​corresponding to each third image in sequence according to the target sliding order, and obtain a first numerical standard deviation in the long sliding window and a second numerical standard deviation in the short sliding window;

[0097] S502: Determine whether a segmentation change point evaluation condition is met based on the first numerical standard deviation and the second numerical standard deviation;

[0098] S503: If the segmentation change point evaluation condition is met, the midpoint of the short sliding window is determined as the original segmentation change point corresponding to the third image.

[0099] The first numerical standard deviation refers to the standard deviation of the target differential sequence values ​​within the long sliding window, determined each time the long sliding window slides. The second numerical standard deviation refers to the standard deviation of the target differential sequence values ​​within the short sliding window, determined each time the short sliding window slides. The target sliding order refers to the sliding order in which the long and short sliding windows slide from back to front, and the target differential sequence values ​​are detected. The target differential sequence values ​​are detected in a sliding order in which the long and short sliding windows slide from back to front.

[0100] As an example, in step S501, the server uses a long sliding window and a short sliding window with a fixed intersection length, and performs sliding detection on the target differential sequence value corresponding to each third image in turn according to the target sliding order, and obtains the first numerical standard deviation in the long sliding window and the second numerical standard deviation in the short sliding window. In this example, the intersection length of the long sliding window and the short sliding window is fixed and they slide together according to the target sliding order for detection. It can be understood that the target sliding order is a sliding order that slides from back to front to detect the target differential sequence value. Figure 13As shown, a sliding detection diagram of a double sliding window corresponding to a first image in a set of images to be processed is shown, wherein the length of the long sliding window is T, the length of the short sliding window is t, the intersection length between the long sliding window and the short sliding window is fixed, and the lengths are all At. The double sliding window slides from back to front according to the target difference sequence value corresponding to the third image. Each time the double sliding window slides, the first numerical standard deviation in the long sliding window and the second numerical standard deviation in the long sliding window are calculated. In this example, the long sliding window and the short sliding window with a fixed intersection length are used to obtain the first numerical standard deviation and the second numerical standard deviation, which facilitates subsequent segmentation point condition evaluation and determination of the target segmentation point.

[0101] The segmentation point evaluation condition is used to determine whether the target difference sequence value at the current position can be used as the original segmentation point according to the first numerical standard deviation and the second numerical standard deviation when the double sliding window slides each time.

[0102] As an example, in step S502, the server determines whether the segmentation point evaluation condition is met according to the first numerical standard deviation and the second numerical standard deviation. For example, the server can determine whether the current first numerical standard deviation and the second numerical standard deviation of the double sliding window meet the segmentation point evaluation condition by determining whether a specific numerical relationship or size relationship between the first numerical standard deviation and the second numerical standard deviation is met. In this example, the first numerical standard deviation and the second numerical standard deviation can be used to accurately determine whether the segmentation point evaluation condition is met.

[0103] As an example, in step S503, the server determines the midpoint of the short sliding window as the original segmentation point corresponding to the third image when it is determined that the segmentation point evaluation condition is met. In this example, when the server determines that the first numerical standard deviation and the second numerical standard deviation meet the segmentation point evaluation condition, the midpoint of the short sliding window in the double sliding window is determined as the original segmentation point corresponding to the third image. That is, the position of the third image corresponding to the target difference sequence value in the middle of the current short sliding window is determined as the original segmentation point, and all original segmentation points are determined by traversing all target difference sequence values. This way of obtaining the original segmentation point is simple and the obtained original segmentation point is accurate.

[0104] The similarity adjacency graph construction method provided in this embodiment utilizes a long sliding window and a short sliding window with a fixed intersection length. The method performs sliding detection on the target differential sequence values ​​corresponding to each third image, sequentially following the target sliding order. The method obtains the first numerical standard deviation within the long sliding window and the second numerical standard deviation within the short sliding window. Based on the first and second numerical standard deviations, it is possible to accurately determine whether the segmentation change point evaluation criteria are met. If the segmentation change point evaluation criteria are met, the midpoint of the short sliding window is determined as the original segmentation change point corresponding to the third image. This method for obtaining the original segmentation change point is relatively simple and accurate.

[0105] In one embodiment, if Figure 6 As shown, step S502, i.e., judging whether the segmentation change point evaluation condition is met based on the first numerical standard deviation and the second numerical standard deviation, includes:

[0106] S601: multiplying the first numerical standard deviation by a preset parameter to determine the target numerical standard deviation;

[0107] S602: If the second numerical standard deviation is greater than the target numerical standard deviation, it is determined that the segmentation change point evaluation condition is met.

[0108] The preset parameters are used to correct the first numerical standard deviation.

[0109] As an example, in step S601, the server determines the product of the first numerical standard deviation and the preset parameter as the target numerical standard deviation. For example, the first numerical standard deviation is δ T , the preset parameter is μ, then the target value standard deviation is μδ T In this example, the target numerical standard deviation is determined to facilitate the subsequent determination of whether the segmentation change point evaluation conditions are met based on the target numerical standard deviation.

[0110] As an example, in step S602, when the server determines that the second numerical standard deviation is greater than the target numerical standard deviation, it determines that the segmentation change point evaluation condition is met. For example, the second numerical standard deviation is δ t , if δ t >μδ T , then the segmentation change point evaluation condition is determined to be satisfied. In this example, whether the segmentation change point evaluation condition is satisfied is determined based on whether the second numerical standard deviation is greater than the target numerical standard deviation. Each time the dual sliding window slides, whether the segmentation change point evaluation condition is satisfied is determined once. This method is more convenient and the judgment result is more accurate.

[0111] The similarity adjacency graph construction method provided in this embodiment determines the product of the first numerical standard deviation and the preset parameter as the target numerical standard deviation. If the second numerical standard deviation is greater than the target numerical standard deviation, it is determined that the segmentation change point evaluation condition is met. Whether the segmentation change point evaluation condition is met is determined based on whether the second numerical standard deviation is greater than the target numerical standard deviation. Each time the double sliding window slides, it is determined whether the segmentation change point evaluation condition is met. This method of determining whether the segmentation change point evaluation condition is met is relatively convenient, and the judgment result is relatively accurate.

[0112] In one embodiment, if Figure 7 As shown, step S402, i.e., determining target segmentation change points corresponding to the K third images based on the original segmentation change points corresponding to the K third images, includes:

[0113] S701: Determine at least one continuous change point sequence from target similarity sequences of the K third images according to original segmentation change points corresponding to the K third images;

[0114] S702: Determine the continuous change point sequence with the largest continuous length as the target change point sequence;

[0115] S703: Determine the starting point in the target change point sequence as the target segmentation change point corresponding to the K third images.

[0116] The continuous change point sequence refers to a change point sequence corresponding to the continuous original segmentation change points after determining the original segmentation change points and mapping all the original segmentation change points to the target similarity sequence corresponding to the third image.

[0117] As an example, in step S701, the server determines at least one continuous change point sequence from the target similarity sequence of the K third images based on the original segmentation change points corresponding to the K third images. For example, if the K third images correspond to L original segmentation change points, the L original segmentation change points are mapped to the target similarity sequence corresponding to the K third images, and continuous original segmentation change points are selected as the target change point sequence. For example, among the L original segmentation change points, if the image similarities corresponding to the 2nd to the 10th original segmentation change points are also a continuous sequence, then the 2nd to the 10th original segmentation change points can be used as a continuous change point sequence. In this example, at least one continuous change point sequence is determined from the target similarity sequence of the K third images, and this method is more intuitive and convenient.

[0118] The target change point sequence is a continuous change point sequence with the highest degree of continuity among at least one continuous change point sequence.

[0119] As an example, in step S702, the server determines the continuous variable point sequence with the maximum continuous length as the target variable point sequence. For example, among the L original segmentation variable points, the 2nd to 10th original segmentation variable points can be a continuous variable point sequence, and the 9th to 30th original segmentation variable points can also be a continuous variable point sequence, but since the 9th to 30th original segmentation variable points have a higher degree of continuity, the 9th to 30th original segmentation variable points are determined as the target variable point sequence.

[0120] As an example, in step S703, after determining the target variable point sequence, the server determines the starting point in the target variable point sequence as the target segmentation variable point corresponding to the K third images. For example, the 9th to 30th original segmentation variable points are the target variable point sequence, and the 9th original segmentation variable point is determined as the target segmentation variable point corresponding to the K third images. This method is simple and efficient, and can better reflect the third image with a higher image similarity to the first image in the last target similarity sequence.

[0121] The similarity adjacency graph construction method provided by the embodiment determines at least one continuous variable point sequence from the target similarity sequence of the K third images according to the original segmentation variable points corresponding to the K third images, determines the continuous variable point sequence with the maximum continuous length as the target variable point sequence, and determines the starting point in the target variable point sequence as the target segmentation variable point corresponding to the K third images. This method of determining the target segmentation variable point is simple and efficient, and can better intuitively reflect the third image with a higher image similarity to the first image in the last target similarity sequence.

[0122] In an embodiment, as shown in FIG. 2, Figure 8 Step S205, that is, cleaning the K third images according to the target segmentation variable points corresponding to the K third images to obtain M fourth images, includes:

[0123] S801: determining a target position of the target segmentation variable point in the target similarity sequence of the K third images;

[0124] S802: determining the target position and all third images before the target position as the M fourth images.

[0125] The target position refers to the third image adjacent to the first third image with a low image similarity to the first image.

[0126] As an example, in step S801, the server determines the target position of the target segmentation variable point in the target similarity sequence of the K third images. As shown in FIG. 2, Figure 14As shown, this is a corresponding relationship diagram between the target similarity sequence and the target difference sequence value of the first image and the third image. The diagram reflects the one-to-one mapping relationship between the target similarity sequence and the target difference sequence value. The position corresponding to the vertical line is the target position. The method for determining the target position is: after determining the starting point in the target change point sequence as the target segmentation change point, the specific position of the target segmentation change point is directly determined in the target similarity sequence of the image. In this example, the server determines the image similarity corresponding to the target segmentation change point in the image similarity column, and uses the third image corresponding to the image similarity as the target position.

[0127] As an example, in step S802, the server determines the target position and all third images before the target position as M fourth images, where M≥1. Figure 14 The figure shows the target position determination result in an example. The position corresponding to the vertical line is the target position. All third images after the target position are discarded, and the third images corresponding to the target position and before the target position are used as the fourth image. In this example, all third images after the target position are discarded, and the remaining third images are all facial images with a high similarity to the first image. The remaining third images are used as the fourth image, which facilitates the subsequent construction of a high-quality similarity subgraph of the first image with a high similarity based on the fourth image and the first image.

[0128] The similarity adjacency graph construction method provided in this embodiment, after determining the target position, discards all third images after the target position and obtains M fourth images, so that the similarity subgraph of the first image subsequently constructed based on the fourth image and the first image has high quality, and constructs a high-quality similarity adjacency graph based on all the similarity subgraphs, so that the subsequent face clustering accuracy is high, which facilitates improving the effect of the face clustering task and reducing the difficulty of the face clustering task.

[0129] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0130] In one embodiment, a similarity adjacency graph construction device is provided, which corresponds to the similarity adjacency graph construction method in the above embodiment. Figure 9 As shown, the similarity adjacency graph construction device includes a to-be-processed image set acquisition module 901, a third image determination module 902, a target differential sequence value determination module 903, a target segmentation change point determination module 904, a fourth image acquisition module 905, and a similarity adjacency graph construction module 906. The functional modules are described in detail as follows:

[0131] The image set to be processed acquisition module 901 is used to acquire the image set to be processed, where the image set to be processed includes a first image and N second images, where N ≥ 2;

[0132] The third image determination module 902 is configured to calculate the image similarity between the first image and each second image, and determine the first K second images with the highest image similarity as the third image; wherein K ≥ 1;

[0133] a target differential sequence value determination module 903, configured to determine a target differential sequence value corresponding to each third image based on image similarities of the K third images;

[0134] A target segmentation change point determination module 904 is configured to perform sliding detection on the target difference sequence values ​​corresponding to each third image using a double sliding window to determine target segmentation change points corresponding to K third images;

[0135] The fourth image acquisition module 905 is configured to clean the K third images according to target segmentation change points corresponding to the K third images to obtain M fourth images, where M≥1.

[0136] The similarity adjacency graph construction module 906 constructs a similarity subgraph corresponding to the first image based on the first image and the M fourth images, traverses the set of images to be processed, determines the similarity subgraphs corresponding to all the first images, and constructs a similarity adjacency graph based on the similarity subgraphs corresponding to all the first images.

[0137] In one embodiment, the target differential sequence value determination module 903 includes:

[0138] A normalization submodule, configured to perform normalization processing on the image similarities of the K third images to obtain normalized similarities corresponding to the K third images;

[0139] A first acquisition submodule is used to perform descending processing on the normalized similarities corresponding to the K third images to obtain a target similarity sequence;

[0140] The second acquisition submodule is used to perform differential processing on the target similarity sequence to obtain a target differential sequence value.

[0141] In one embodiment, the target segmentation change point determination module 904 includes:

[0142] A first determination submodule is configured to perform sliding detection on the target differential sequence value corresponding to each third image using a long sliding window and a short sliding window to determine the original segmentation change points corresponding to the K third images;

[0143] The second determining submodule is configured to determine target segmentation change points corresponding to the K third images based on the original segmentation change points corresponding to the K third images.

[0144] In one embodiment, the first determining submodule includes:

[0145] a numerical standard deviation obtaining unit, configured to perform sliding detection on the target differential sequence values ​​corresponding to each third image using a long sliding window and a short sliding window with fixed intersection lengths, in sequence according to the target sliding order, to obtain a first numerical standard deviation within the long sliding window and a second numerical standard deviation within the short sliding window;

[0146] an evaluation condition judgment unit, configured to judge whether a segmentation change point evaluation condition is satisfied based on the first numerical standard deviation and the second numerical standard deviation;

[0147] The first determining unit is configured to determine the midpoint of the short sliding window as the original segmentation change point corresponding to the third image if the segmentation change point evaluation condition is met.

[0148] In one embodiment, the evaluation condition determination unit includes:

[0149] A first determining subunit is configured to determine the product of the first numerical standard deviation and a preset parameter as a target numerical standard deviation;

[0150] The second determining subunit is configured to determine that a segmentation change point evaluation condition is satisfied if the second numerical standard deviation is greater than the target numerical standard deviation.

[0151] In one embodiment, the second determining submodule includes:

[0152] A second determining unit is configured to determine at least one continuous change point sequence from the target similarity sequences of the K third images according to the original segmentation change points corresponding to the K third images;

[0153] a third determining unit, configured to determine a continuous change point sequence with the largest continuous length as a target change point sequence;

[0154] The fourth determining unit is configured to determine the starting point in the target change point sequence as the target segmentation change point corresponding to the K third images.

[0155] In one embodiment, the fourth image acquisition module 905 includes:

[0156] A target position determination submodule, for determining the target position of the target segmentation change point in the target similarity sequence of the K third images;

[0157] The fourth image acquisition submodule is configured to determine the target position and all third images before the target position as M fourth images.

[0158] For the specific definition of the similarity adjacency graph construction device, please refer to the definition of the method above and will not be repeated here. The various modules in the above-mentioned similarity adjacency graph construction device can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0159] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 10 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data used or generated during the execution of the similarity adjacency graph construction method. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method is implemented.

[0160] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method in the above embodiment is implemented, for example Figure 2 S201-S206 shown, or Figures 3 to 8 Alternatively, when the processor executes the computer program, the functions of each module / unit in the embodiment of the similarity adjacency graph construction device are realized, for example Figure 9 To avoid repetition, the functions of the to-be-processed image set acquisition module 901, the third image determination module 902, the target differential sequence value determination module 903, the target segmentation change point determination module 904, the fourth image acquisition module 905 and the similarity adjacency graph construction module 906 are not repeated here.

[0161] 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 method in the above embodiment is implemented, for example Figure 2 S201-S206 shown, or Figures 3 to 8 Alternatively, when the computer program is executed by a processor, the functions of the modules / units in the embodiment of the similarity adjacency graph construction device are realized, for example, Figure 9The functions of the illustrated to-be-processed image set obtaining module 901, the third image determining module 902, the target difference sequence value determining module 903, the target segmentation turning point determining module 904, the fourth image obtaining module 905, and the similarity adjacency graph constructing module 906 are not repeated here to avoid redundancy. The computer-readable storage medium can be non-volatile or volatile.

[0162] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiments can be included. Any reference to memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0163] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is exemplified. In actual applications, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the similarity adjacency graph constructing apparatus is divided into different functional units or modules to complete all or part of the above-mentioned functions.

[0164] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, and not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A similarity adjacency graph construction method, characterized in that: include: Acquire a set of images to be processed, the set of images to be processed comprising a first image and N second images, where N is greater than or equal to 2; Calculating the image similarity between the first image and each of the second images, and determining the first K second images with higher image similarity as third images; wherein K ≥ 1; determining a target differential sequence value corresponding to each of the K third images according to image similarities of the K third images; Using a double sliding window to perform sliding detection on the target differential sequence value corresponding to each of the third images, and determine K target segmentation change points corresponding to the third images; Cleaning the K third images according to target segmentation change points corresponding to the K third images to obtain M fourth images; where M ≥ 1; Based on the first image and the M fourth images, a similarity subgraph corresponding to the first image is constructed, the set of images to be processed is traversed to determine the similarity subgraphs corresponding to all the first images, and a similarity adjacency graph is constructed based on the similarity subgraphs corresponding to all the first images.

2. The similarity adjacency graph construction method according to claim 1, wherein: Determining a target differential sequence value corresponding to each of the third images based on the image similarities of the K third images includes: Normalizing the image similarities of the K third images to obtain normalized similarities corresponding to the K third images; Performing descending processing on the normalized similarities corresponding to the K third images to obtain a target similarity sequence; Perform differential processing on the target similarity sequence to obtain a target differential sequence value.

3. The similarity adjacency graph construction method according to claim 2, wherein: The method of using a double sliding window to perform sliding detection on the target differential sequence value corresponding to each of the third images to determine K target segmentation change points corresponding to the third images includes: Using a long sliding window and a short sliding window, perform sliding detection on the target differential sequence value corresponding to each of the third images to determine K original segmentation change points corresponding to the third images; K target segmentation change points corresponding to the third image are determined according to the K original segmentation change points corresponding to the third image.

4. The similarity adjacency graph construction method according to claim 3, wherein: The method of using a long sliding window and a short sliding window to perform sliding detection on the target differential sequence value corresponding to each of the third images to determine K original segmentation change points corresponding to the third images includes: Using a long sliding window and a short sliding window with a fixed intersection length, sliding detection is performed on the target difference sequence value corresponding to each of the third images in sequence according to the target sliding order, and a first numerical standard deviation within the long sliding window and a second numerical standard deviation within the short sliding window are obtained; determining whether a segmentation change point evaluation condition is satisfied based on the first numerical standard deviation and the second numerical standard deviation; If the segmentation change point evaluation condition is met, the midpoint of the short sliding window is determined as the original segmentation change point corresponding to the third image.

5. The similarity adjacency graph construction method according to claim 4, wherein: The determining, based on the first numerical standard deviation and the second numerical standard deviation, whether a segmentation change point evaluation condition is satisfied includes: multiplying the first numerical standard deviation by a preset parameter to determine the target numerical standard deviation; If the second numerical standard deviation is greater than the target numerical standard deviation, it is determined that the segmentation change point evaluation condition is met.

6. The similarity adjacency graph construction method according to claim 3, wherein: The determining, based on the K original segmentation change points corresponding to the third images, the K target segmentation change points corresponding to the third images includes: determining at least one continuous change point sequence from the target similarity sequences of the K third images according to the original segmentation change points corresponding to the K third images; The continuous change point sequence with the largest continuous length is determined as the target change point sequence; The starting point in the target change point sequence is determined as the target segmentation change points corresponding to the K third images.

7. The method for constructing a similarity adjacency graph according to claim 2, wherein: Cleaning the K third images according to the target segmentation change points corresponding to the K third images to obtain M fourth images includes: Determining a target position of the target segmentation change point in the target similarity sequences of the K third images; The target position and all the third images before the target position are determined as M fourth images.

8. A similarity adjacency graph construction device, characterized in that: include: A module for acquiring an image set to be processed, configured to acquire an image set to be processed, wherein the image set to be processed includes a first image and N second images; wherein N ≥ 2; A third image determination module is configured to calculate the image similarity between the first image and each of the second images, and determine the first K second images with the highest image similarity as third images; wherein K ≥ 1; a target differential sequence value determination module, configured to determine a target differential sequence value corresponding to each of the K third images based on image similarities of the K third images; a target segmentation change point determination module, configured to perform sliding detection on the target differential sequence values ​​corresponding to each of the third images using a double sliding window, and determine target segmentation change points corresponding to K of the third images; a fourth image acquisition module, configured to clean the K third images according to target segmentation change points corresponding to the K third images, and acquire M fourth images; wherein M ≥ 1; A similarity adjacency graph construction module constructs a similarity subgraph corresponding to the first image based on the first image and the M fourth images, traverses the set of images to be processed, determines the similarity subgraphs corresponding to all the first images, and constructs a similarity adjacency graph based on the similarity subgraphs corresponding to all the first images.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the similarity adjacency graph construction method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the similarity adjacency graph construction method according to any one of claims 1 to 7 is implemented.

Images