Image segmentation method and system based on large model

Abstract

The invention is suitable for the technical field of image segmentation, and particularly relates to an image segmentation method and system based on a large model, and the method comprises the steps: obtaining an original image needing to be segmented, recognizing the composition pixels of the original image, configuring the number of each composition pixel through a preset numbering rule, and carrying out the numbering of each composition pixel through an edge detection algorithm, segmenting a plurality of instances from the original image; marking the edge pixel of each instance, and integrating the serial numbers of the edge pixels to generate an identifier; selecting endpoint blocks from the edge pixels, connecting the endpoint blocks to generate an undirected graph, reading the number of each endpoint block, and integrating the endpoint blocks to obtain a fingerprint value; and creating an instance comparison model consisting of a plurality of verification layers. According to the method, a plurality of possible target images can be synchronously sent to the user by generating the queue, so that the recognition accuracy is remarkably improved through a man-machine cooperation mode, manual error correction is performed, a large model blind area is complemented, and the accuracy is further improved.

Description

Technical Field

[0001] This invention relates to the field of image segmentation technology, and in particular to an image segmentation method and system based on a large model. Background Technology

[0002] Image segmentation refers to the process of dividing an image into several regions or objects with specific meanings. Its purpose is to extract regions in the image that have independent semantic or structural features, such as people, vehicles, buildings, or backgrounds.

[0003] However, in existing technologies, the lack of unique fingerprints for segmented images leads to a lack of binding relationships with the original images. This results in low efficiency and low accuracy when performing reverse lookups (i.e., finding the original image from the segmented image) in massive images. Furthermore, the inability to verify the source of the segmented image greatly affects its credibility.

[0004] Therefore, "how to trace back to the original image based on image segmentation and large model technology" is the technical problem that this invention needs to solve. Summary of the Invention

[0005] The purpose of this invention is to provide an image segmentation method and system based on a large model, in order to solve the problem mentioned in the background art of "how to trace back the original image based on image segmentation and large model technology".

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An image segmentation method based on a large model, the method comprising:

[0008] The original image to be segmented is obtained, the constituent pixels of the original image are identified, and the number of each constituent pixel is configured according to the preset numbering rules. Then, an edge detection algorithm is used to segment several instances from the original image.

[0009] Mark the edge pixels of each instance, combine the edge pixel numbers, and generate an identifier;

[0010] From the edge pixels, select endpoint blocks, connect them to generate an undirected graph, read the number of each endpoint block, and integrate them to obtain the fingerprint value;

[0011] Create an instance comparison model consisting of several verification layers, edit a multi-level verification strategy set and write it into the instance comparison model, receive query examples uploaded by users, input them into the instance comparison model, calculate the similarity between the query examples and each original image, output several target images, sort them in descending order of similarity, generate a queue, and send the queue to the user terminal.

[0012] Furthermore, the step of segmenting several instances from the original image using an edge detection algorithm includes:

[0013] Configure the category corresponding to each instance and the original image, integrate all categories, generate a source database, and connect the instance comparison model to the source database;

[0014] Construct a time window, calculate the usage frequency of each category, and adjust the sorting order of each category in the source database.

[0015] Furthermore, the method also includes:

[0016] Establish the correspondence between instances and original images;

[0017] From the endpoint blocks, select the marker points, and using the number of the marker points according to the correspondence, generate an index for each original image and write it into the source database.

[0018] Furthermore, the step of reading the number of each endpoint block and integrating it to obtain the fingerprint value includes:

[0019] Construct a hash function, perform chain hashing on the number of each endpoint block in sequence to obtain a verification symbol, and insert a label generated by the verification symbol into the original image;

[0020] Collect attribute data of the original image, wherein the attribute data includes at least: source and generation time, integrate the attribute data and fingerprint value, and generate a source watermark.

[0021] Furthermore, the method also includes:

[0022] Set up an edge platform and a cloud platform, synchronize the instance to the edge platform, and upload the original image to the cloud platform;

[0023] A middle layer is established to integrate a pre-edited linkage mechanism, linking the edge and the cloud, and the instance comparison model is migrated to the middle layer.

[0024] Furthermore, the steps of creating an instance comparison model consisting of several verification layers, editing a multi-level verification strategy set, and writing it into the instance comparison model include:

[0025] From the similarity scores, a threshold is selected, and a correspondence between the threshold and the verification layer is established.

[0026] The original image with a similarity greater than a threshold is defined as the target image.

[0027] Furthermore, the method also includes:

[0028] A preset number of target images are selected from the front of the queue to generate query results;

[0029] The query results are integrated with selection buttons, feedback links are enabled, and the query results are sent to the user's terminal.

[0030] Furthermore, the system includes:

[0031] The segmentation module is used to acquire the original image to be segmented, identify the constituent pixels of the original image, configure the number of each constituent pixel according to the preset numbering rules, and use the edge detection algorithm to segment several instances from the original image.

[0032] The integration module is used to mark the edge pixels of each instance, integrate the edge pixel numbers, generate an identifier, select endpoint blocks from the edge pixels, connect them to generate an undirected graph, read the number of each endpoint block, and integrate them to obtain a fingerprint value.

[0033] The delivery module is used to create an instance comparison model consisting of several verification layers, edit a multi-level verification strategy set and write it into the instance comparison model, receive query examples uploaded by users, input them into the instance comparison model, calculate the similarity between the query example and each original image, output several target images, sort them in descending order of similarity, generate a queue, and deliver the queue to the user terminal.

[0034] Furthermore, the segmentation module includes:

[0035] The access unit is used to configure the category corresponding to each instance and the original image, integrate all categories, generate a source database, and access the instance comparison model into the source database;

[0036] The adjustment unit is used to construct a time window, calculate the usage frequency of each category, and adjust the sorting order of each category in the source database.

[0037] Furthermore, the integration module includes:

[0038] An insertion unit is used to construct a hash function, sequentially perform chain hashing on the number of each endpoint block to obtain a verification symbol, and insert a label generated by the verification symbol into the original image;

[0039] The generation unit is used to collect attribute data of the original image, wherein the attribute data includes at least: source and generation time, and integrates the attribute data and fingerprint value to generate a source watermark.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] By utilizing edge detection algorithms, instance boundaries can be accurately located, improving segmentation accuracy. Identifying instances facilitates comparison and provides a data foundation for tracing the origin of the original image. Determining fingerprint values ​​enables the construction of lightweight instance fingerprints, significantly improving comparison efficiency. Constructing a multi-level verification strategy set allows for the initial use of fuzzy verification to quickly eliminate a large number of irrelevant images, gradually transitioning to precise verification. This ensures both tracing efficiency and accuracy. Generating queues allows for the simultaneous sending of multiple potential target images to the user, significantly improving recognition accuracy through human-machine collaboration. Human error correction fills in blind spots in the large model, further enhancing accuracy. Attached Figure Description

[0042] Figure 1 A flowchart illustrating the image segmentation method based on a large model provided in an embodiment of the present invention;

[0043] Figure 2 This is a first sub-flowchart of the image segmentation method based on a large model provided in an embodiment of the present invention;

[0044] Figure 3 This is a second sub-flow flowchart of the image segmentation method based on a large model provided in an embodiment of the present invention;

[0045] Figure 4 This is a third sub-flow flowchart of the image segmentation method based on a large model provided in an embodiment of the present invention;

[0046] Figure 5 This is a block diagram of the composition of an image segmentation system based on a large model provided in an embodiment of the present invention;

[0047] Figure 6 A block diagram illustrating the composition of the segmentation module in a large-model-based image segmentation system provided in an embodiment of the present invention;

[0048] Figure 7 A block diagram of the integrated module in the large-model-based image segmentation system provided in the embodiments of the present invention;

[0049] Figure 8 This is a block diagram of the distribution module in the large-model-based image segmentation system provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention 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 merely illustrative and not intended to limit the invention.

[0051] In Example 1, Figure 1The implementation flow of the image segmentation method based on a large model provided in this embodiment of the invention is illustrated below, and is described in detail below:

[0052] S100: Obtain the original image to be segmented, identify the constituent pixels of the original image, configure the number of each constituent pixel according to the preset numbering rules, and use the edge detection algorithm to segment several instances from the original image.

[0053] The original image to be segmented is determined, and the original image is analyzed at the pixel level to identify all the pixels contained in the original image, i.e., the constituent pixels. According to the preset numbering rules, the number of each constituent pixel is determined, where the numbering rules can be sequential numbering or coordinate numbering. In this application, sequential numbering is used as an example. For example, according to the sequential numbering, the constituent pixel in the first row and first column is numbered 1, the number in the first row and second column is numbered 2, and so on.

[0054] Edge detection algorithms (such as Canny) are used to identify and extract contours and structural boundaries in the original image. Boundaries are located by detecting pixel value gradient changes and gray-scale abrupt changes. Furthermore, continuous edge lines and closed regions in the original image are used as target regions. Using the target regions, several instances with clear boundaries (such as people and objects) are segmented from the original image. Each instance corresponds to a portion of pixels in the original image.

[0055] S200: Mark the edge pixels of each instance, integrate the edge pixel numbers, generate an identifier, select endpoint blocks from the edge pixels, connect them to generate an undirected graph, read the number of each endpoint block, and integrate them to obtain the fingerprint value.

[0056] Identify all edge pixels that make up each instance, i.e., edge pixels, which are usually located in the transition area between the instance and the background or adjacent instances; determine the number of each edge pixel and integrate them to generate an identifier; specifically, the numbers of all edge pixels can be integrated in the order from top to bottom and from left to right. There may be a few identical identifiers, but this does not affect instance tracing and can be used for subsequent image retrieval, instance comparison and cross-image recognition operations.

[0057] From the edge pixels, select the endpoint blocks that have structural significance. The endpoint blocks are the top, bottom, left, right and turning points of the edge pixels. The identification of endpoint blocks is based on the connectivity, curvature change and pixel degree (number of adjacent pixels) of the edge pixels.

[0058] Using endpoint blocks as nodes, an undirected graph is constructed based on their spatial adjacency relationships in the instance using the nearest neighbor algorithm. It should be noted that adjacent endpoint blocks are directly connected in the undirected graph. The number corresponding to each endpoint block is read out, and the numbers of each endpoint block are integrated in the same order as above (from top to bottom, from left to right) to generate a fingerprint value. The fingerprint value can be simply understood as a simplified representation of the identifier, which facilitates the rapid comparison of instances.

[0059] S300: Creates an instance comparison model consisting of several verification layers, edits a multi-level verification strategy set and writes it into the instance comparison model, receives query examples uploaded by users, inputs them into the instance comparison model, calculates the similarity between the query example and each original image, outputs several target images, sorts them in descending order of similarity, generates a queue, and sends the queue to the user terminal.

[0060] By using a hierarchical and progressively convergent approach, an instance comparison model consisting of multiple verification layers is constructed. The instance comparison model can compare and match query instances with the original images in the database. Each verification layer corresponds to a verification method, and the set of verification methods is a multi-level verification strategy set.

[0061] Specifically, the instance comparison model consists of three verification layers (layer 1, layer 2, and layer 3). The specific number should be determined based on user needs, the complexity of the original image, etc. Specifically, in layer 1, after receiving the query instance uploaded by the user, which is the instance to be queried, it can be simply understood as a part of the original image. In layer 1, the undirected graph of the instance is identified and the topological structure is verified to filter out candidate original images. In layer 2, the fingerprint value of the instance is determined and compared with the filtered candidate original images. In layer 3, the identifier is compared with the comparison result again.

[0062] The query example is input into the instance comparison model, and the similarity of each verification layer is calculated sequentially. For example, in the first layer, assuming the undirected graph of the query example is a circle, the parts containing circular instances are found in all the original images (let's call it set A). Using the cosine similarity calculation formula, the similarity between each endpoint block in set A and the endpoint block of the query example is calculated sequentially. The parts with similarity greater than the threshold are filtered out (let's call it set B), and the similarity between the edge pixels of set B and the edge pixels of the instance is calculated. This similarity is the similarity between the query example and the original image. The parts with similarity greater than the threshold are filtered out again to obtain the target image. The target images are sorted in descending order of similarity to generate a queue. The target image is the original image that contains similar or identical features to the instance. The queue is a set of target images. The queue is sent to the user terminal, and the user selects the target image containing the instance from the queue.

[0063] In Example 2, Figure 2 The implementation flow of the image segmentation method based on a large model provided by an embodiment of the present invention is illustrated. The following details the steps of segmenting several instances from the original image using an edge detection algorithm:

[0064] S101: Configure the category corresponding to each instance and the original image, integrate all categories, generate the source database, and connect the instance comparison model to the source database.

[0065] Semantic recognition and classification are performed on each instance and the original image to determine the corresponding category, such as "person", "car", "building", "animal", "electronic product" or a combination thereof. The specific classification process can be automatically completed by a pre-trained large model, or it can be combined with manual review or existing image classification standards for annotation. According to the category, all instances and original images are integrated to generate a source database. The constructed instance comparison model is connected to the source database. When a user uploads a query example, the instance comparison model can automatically extract the instance features (such as edge structure, endpoint number and category) in the query example and match and compare them with existing instances or original images in the source database.

[0066] S102: Construct a time window, calculate the usage frequency of each category, and adjust the order of each category in the source database.

[0067] Two time points are selected to construct a time window. The frequency of instances or original images in each category is calculated, and the sorting position of the categories in the source database is adjusted according to the frequency from high to low. The advantage of this method is that it can prioritize the comparison of instances and original images with higher usage frequency, thereby improving the source tracing efficiency of query cases.

[0068] In Embodiment 3, unlike Embodiment 1, the method further includes:

[0069] Establish the correspondence between instances and original images;

[0070] From the endpoint blocks, select the marker points, and using the number of the marker points according to the correspondence, generate an index for each original image and write it into the source database.

[0071] The source of each instance is determined and a correspondence with the original image is established. Representative "identifier points", such as key turning points or contour feature mutation points, are selected from the endpoint blocks. The numbers corresponding to the identifier points are integrated to generate indexes and written into the source database. By constructing indexes, the efficiency of calling the original data can be greatly improved.

[0072] In Example 4, Figure 3 The implementation flow of the image segmentation method based on a large model provided by an embodiment of the present invention is illustrated below. The steps of reading the number of each endpoint block and integrating them to obtain the fingerprint value are described in detail below:

[0073] S201: Construct a hash function, perform chain hashing on the number of each endpoint block in sequence to obtain a verification symbol, and insert a label generated by the verification symbol into the original image.

[0074] From existing technologies, a hash function (such as SHA-256) is selected, and the number of each endpoint block is hashed sequentially to obtain a verifier. For example, following the order from top to bottom and from left to right, the first endpoint block is found, and its number is hashed to obtain the first hash value. The first hash value is combined with the number of the second endpoint block, and the hashing continues to obtain the second hash value, and so on. This hashing method is called chain hashing. The final Nth hash value is defined as the verifier, where each original image corresponds to a verifier. Verifiers are stored using labels. By constructing verifiers, it is possible to determine whether the original image has been tampered with during storage, thereby improving the accuracy of tracing the source of query cases.

[0075] S202: Collect attribute data of the original image, wherein the attribute data includes at least: source and generation time, integrate the attribute data and fingerprint value, and generate a source watermark.

[0076] Using attribute data and fingerprint values, watermark information is generated, defined as a traceability watermark, and written into the instance.

[0077] In Example 5, unlike Example 1, the method further includes:

[0078] Set up an edge platform and a cloud platform, synchronize the instance to the edge platform, and upload the original image to the cloud platform;

[0079] A middle layer is established to integrate a pre-edited linkage mechanism, linking the edge and the cloud, and the instance comparison model is migrated to the middle layer.

[0080] Edge computing nodes are built on the edge side, and a central data platform is built in the cloud. Instances are synchronized to the edge side, and original images are uploaded to the cloud. By integrating the edge side and the cloud, a distributed storage architecture is constructed. The advantage of this approach is that it can perform two-way authenticity verification using instances and original images. An intermediate layer is used to link the edge layer and the cloud. The intermediate layer can be built by edge nodes or by components of the central data platform. A linkage mechanism is embedded in the intermediate layer. The linkage mechanism means that after the query example is input into the instance comparison model, the cloud automatically calls the target image, and the edge side synchronously outputs the instance and attribute data of all target images.

[0081] In Example 6, Figure 4 The implementation flow of the image segmentation method based on a large model provided by an embodiment of the present invention is illustrated. The following details the steps of creating an instance comparison model consisting of several verification layers, editing a multi-level verification strategy set, and writing it into the instance comparison model:

[0082] S301: Select a threshold from the similarity and establish a correspondence between the threshold and the verification layer.

[0083] As described in S300, the similarity threshold in the second layer is not the same as the similarity threshold in the third layer, and each verification layer corresponds to a different threshold; thus, a correspondence between the threshold and the verification layer is established.

[0084] S302: Define the original image with a similarity greater than the threshold as the target image.

[0085] In Example 7, unlike Example 1, the method further includes:

[0086] A preset number of target images are selected from the front of the queue to generate query results;

[0087] The query results are integrated with selection buttons, feedback links are enabled, and the query results are sent to the user's terminal.

[0088] From the front of the queue, a preset number of target images are selected, which are the query results. These results are then sent to the user's terminal, which is the device of the user who uploaded the query example. The query results include a selection button, allowing the user to select the image containing the query example from multiple target images. A feedback link is also included in the query results, allowing the user to submit error feedback, suggestions for correction, or further explanation of the query intent.

[0089] Figure 5 This diagram illustrates the structural block diagram of a large-model-based image segmentation system provided in an embodiment of the present invention. The large-model-based image segmentation system 1 includes:

[0090] The segmentation module 11 is used to acquire the original image to be segmented, identify the constituent pixels of the original image, configure the number of each constituent pixel according to the preset numbering rules, and segment several instances from the original image using an edge detection algorithm.

[0091] Integration module 12 is used to mark the edge points of each instance and define them as target pixels, integrate the numbers of the target pixels, generate identifiers, select endpoint blocks from the target pixels, connect them to generate an undirected graph, read the number of each endpoint block, and integrate to obtain fingerprint values.

[0092] The distribution module 13 is used to create an instance comparison model consisting of several verification layers, edit a multi-level verification strategy set and write it into the instance comparison model, receive query examples uploaded by users, input them into the instance comparison model, calculate the similarity between the query examples and each original image, output several target images, sort them in descending order of similarity, generate a queue, and distribute the queue to the user terminal.

[0093] Figure 6 This diagram illustrates the structural block diagram of a large-model-based image segmentation system provided in an embodiment of the present invention. The segmentation module 11 includes:

[0094] Access unit 111 is used to configure the category corresponding to each instance and the original image, integrate all categories, generate a source database, and access the instance comparison model into the source database;

[0095] Adjustment unit 112 is used to construct a time window, calculate the usage frequency of each category, and adjust the sorting order of each category in the source database.

[0096] Figure 7 This diagram illustrates the structural block diagram of a large-model-based image segmentation system provided in an embodiment of the present invention. The integration module 12 includes:

[0097] Insertion unit 121 is used to construct a hash function, perform chain hashing on the number of each endpoint block in sequence to obtain a verification symbol, and insert a label generated by the verification symbol into the original image;

[0098] The generation unit 122 is used to collect attribute data of the original image, wherein the attribute data includes at least: source and generation time, and integrates the attribute data and fingerprint value to generate a source watermark.

[0099] Figure 8 This diagram illustrates the structural block diagram of the image segmentation system based on a large model provided in an embodiment of the present invention. The sending module 13 includes:

[0100] Establishment unit 131 is used to select a threshold from the similarity and establish a correspondence between the threshold and the verification layer;

[0101] Definition unit 132 is used to define the original image with a similarity greater than a threshold as the target image.

[0102] The segmentation module 11 is mainly used to complete step S100, the integration module 12 is mainly used to complete step S200, and the distribution module 13 is mainly used to complete step S300.

[0103] Access unit 111 is mainly used to complete step S101, and adjustment unit 112 is mainly used to complete step S102;

[0104] The insertion unit 121 is mainly used to complete step S201, and the generation unit 122 is mainly used to complete step S202.

[0105] The establishment unit 131 is mainly used to complete step S301, and the definition unit 132 is mainly used to complete step S302.

[0106] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An image segmentation method based on a large model, characterized in that, The method includes: The original image to be segmented is obtained, the constituent pixels of the original image are identified, and the number of each constituent pixel is configured according to the preset numbering rules. Then, an edge detection algorithm is used to segment several instances from the original image. Mark the edge pixels of each instance, combine the edge pixel numbers, and generate an identifier; From the edge pixels, select endpoint blocks, connect them to generate an undirected graph, read the number of each endpoint block, and integrate them to obtain the fingerprint value; Create an instance comparison model consisting of several verification layers, edit a multi-level verification strategy set and write it into the instance comparison model, receive query examples uploaded by users, input them into the instance comparison model, calculate the similarity between the query examples and each original image, output several target images, sort them in descending order of similarity, generate a queue, and send the queue to the user terminal.

2. The image segmentation method based on a large model according to claim 1, characterized in that, The step of segmenting several instances from the original image using an edge detection algorithm includes: Configure the category corresponding to each instance and the original image, integrate all categories, generate a source database, and connect the instance comparison model to the source database; Construct a time window, calculate the usage frequency of each category, and adjust the sorting order of each category in the source database.

3. The image segmentation method based on a large model according to claim 2, characterized in that, The method further includes: Establish the correspondence between instances and original images; From the endpoint blocks, select the marker points, and using the number of the marker points according to the correspondence, generate an index for each original image and write it into the source database.

4. The image segmentation method based on a large model according to claim 3, characterized in that, The step of reading the number of each endpoint block and integrating them to obtain the fingerprint value includes: Construct a hash function, perform chain hashing on the number of each endpoint block in sequence to obtain a verification symbol, and insert a label generated by the verification symbol into the original image; Collect attribute data of the original image, wherein the attribute data includes at least: source and generation time, integrate the attribute data and fingerprint value, and generate a source watermark.

5. The image segmentation method based on a large model according to claim 1, characterized in that, The method further includes: Set up an edge platform and a cloud platform, synchronize the instance to the edge platform, and upload the original image to the cloud platform; A middle layer is established to integrate a pre-edited linkage mechanism, linking the edge and the cloud, and the instance comparison model is migrated to the middle layer.

6. The image segmentation method based on a large model according to claim 5, characterized in that, The steps of creating an instance comparison model consisting of several verification layers, editing a multi-level verification strategy set, and writing it into the instance comparison model include: From the similarity scores, a threshold is selected, and a correspondence between the threshold and the verification layer is established. The original image with a similarity greater than a threshold is defined as the target image.

7. The image segmentation method based on a large model according to claim 1, characterized in that, The method further includes: A preset number of target images are selected from the front of the queue to generate query results; The query results are integrated with selection buttons, feedback links are enabled, and the query results are sent to the user's terminal.

8. An image segmentation system based on a large model, characterized in that, The system includes: The segmentation module is used to acquire the original image to be segmented, identify the constituent pixels of the original image, configure the number of each constituent pixel according to the preset numbering rules, and use the edge detection algorithm to segment several instances from the original image. The integration module is used to mark the edge pixels of each instance, integrate the edge pixel numbers, generate an identifier, select endpoint blocks from the edge pixels, connect them to generate an undirected graph, read the number of each endpoint block, and integrate them to obtain a fingerprint value. The delivery module is used to create an instance comparison model consisting of several verification layers, edit a multi-level verification strategy set and write it into the instance comparison model, receive query examples uploaded by users, input them into the instance comparison model, calculate the similarity between the query example and each original image, output several target images, sort them in descending order of similarity, generate a queue, and deliver the queue to the user terminal.

9. The image segmentation system based on a large model according to claim 8, characterized in that, The segmentation module includes: The access unit is used to configure the category corresponding to each instance and the original image, integrate all categories, generate a source database, and access the instance comparison model into the source database; The adjustment unit is used to construct a time window, calculate the usage frequency of each category, and adjust the sorting order of each category in the source database.

10. The image segmentation system based on a large model according to claim 9, characterized in that, The integration module includes: An insertion unit is used to construct a hash function, perform chain hashing on the number of each endpoint block in sequence to obtain a verification symbol, and insert a label generated by the verification symbol into the original image; The generation unit is used to collect attribute data of the original image, wherein the attribute data includes at least: source and generation time, and integrates the attribute data and fingerprint value to generate a source watermark.

Images