Image searching method and device, electronic equipment and storage medium

Abstract

The invention provides an image search method, which comprises the following steps: in a multi-round image search process, acquiring first input data of a current round, and acquiring second input data of a historical round, an interval time between the historical round and the current round being less than a preset interval time; based on the second input data, extracting auxiliary search semantics; and based on the first input data and the auxiliary search semantics, performing search processing in the image database to obtain a search result corresponding to the first input data, and returning the search result. The image search method and device solve the problems that according to an existing image search method, each round of search is independently carried out based on the input data of the current round, an accurate result can not be obtained in each time of search, the search result accuracy is not high, a user needs to continuously search the input data, and consequently the working efficiency is low.

Description

Technical Field

[0001] This invention relates to the field of image management technology, and in particular to an image search method, apparatus, electronic device, and storage medium. Background Technology

[0002] In image search technology, due to variations in user input intent and errors in the system's understanding of that intent, accurate results are not always obtained. Users may need to perform multiple searches for inaccurate results. However, existing database rules require each search to be performed independently based on the input data of that round, necessitating continuous searching of the input data by the user, resulting in low efficiency in image search under multiple scenarios. Therefore, there is an urgent need for an image search method that can accurately search for images, addressing the problems of existing methods where each search is performed independently based on the input data of the current round, resulting in low search result accuracy and requiring users to continuously search the input data, leading to low work efficiency. Summary of the Invention

[0003] This invention provides an image search method to address the problems of existing image search methods, which perform independent searches based on the input data of each round, resulting in inaccurate search results and low efficiency due to the need for users to continuously search for input data. This invention obtains the first input data of the current round and the second input data of previous rounds during a multi-round image search process. Based on the second input data, auxiliary search semantics are extracted. Then, based on the first input data and the auxiliary search semantics, a search is performed in the image database to obtain the search result corresponding to the first input data, and the search result is returned. This solves the problems of existing image search methods, which perform independent searches based on the input data of each round, resulting in inaccurate search results and low efficiency due to the need for users to continuously search for input data.

[0004] In a first aspect, embodiments of the present invention provide an image search method, the method comprising the following steps: During the multi-round image search process, the first input data of the current round is obtained, and the second input data of the historical rounds is obtained, wherein the interval between the historical rounds and the current round is less than a preset interval. Based on the second input data, auxiliary search semantics are extracted; Based on the first input data and the auxiliary search semantics, a search process is performed in the image database to obtain the search results corresponding to the first input data, and the search results are returned.

[0005] Optionally, the second input data is multimodal input data, and the step of extracting auxiliary search semantics based on the second input data includes: Based on a multimodal large model, semantic extraction processing is performed on the second input data to obtain the auxiliary search semantics corresponding to the second input data.

[0006] Optionally, the step of performing search processing in the image database based on the first input data and the auxiliary search semantics to obtain the search results corresponding to the first input data includes: Based on the multimodal large model, semantic extraction processing is performed on the first input data to obtain the main search semantics of the first input data; Based on the primary search semantics and the auxiliary search semantics, a search process is performed in the image database to obtain the search results corresponding to the first input data.

[0007] Optionally, the step of performing search processing in the image database based on the primary search semantics and the auxiliary search semantics to obtain the search results corresponding to the first input data includes: Obtain the first weight value corresponding to the primary search semantic and the second weight value corresponding to the auxiliary search semantic; Based on the first weight value and the second weight value, the main search semantics and the auxiliary search semantics are weighted and added together to obtain the fused semantics; The fusion semantics are searched in the image database to obtain the search results corresponding to the first input data.

[0008] Optionally, obtaining the first weight value corresponding to the main search semantic includes: The first initial dynamic weight value is dynamically calculated based on the time interval between the current round and the historical round, and the time interval is negatively correlated with the first initial dynamic weight value. Based on the semantic relevance between the primary search semantic and the auxiliary search semantic, the first initial dynamic weight value is adjusted to obtain the first weight value corresponding to the primary search semantic.

[0009] Optionally, obtaining the second weight value corresponding to the auxiliary search semantic includes: The second initial dynamic weight value is dynamically calculated based on the source round sorting of the auxiliary search semantics. The later the source round sorting is, the larger the second initial dynamic weight value is. Based on the semantic confidence calculation of the auxiliary search semantics, the second initial dynamic weight value is adjusted based on the confidence value to obtain the second weight value corresponding to the auxiliary search semantics.

[0010] Optionally, the fused semantics are searched in the image database to obtain the search results corresponding to the first input data, including: Calculate the semantic similarity between the fused semantics and the image semantics corresponding to the images in the image database; The K images with the highest semantic similarity are selected as the search results corresponding to the first input data.

[0011] Secondly, embodiments of the present invention provide an image search device, the image search device comprising: The acquisition module is used to acquire the first input data of the current round and the second input data of the historical round during the multi-round image search process, wherein the interval between the historical round and the current round is less than a preset interval. The extraction module is used to extract auxiliary search semantics based on the second input data; The search module is used to perform search processing in the image database based on the first input data and the auxiliary search semantics, obtain the search results corresponding to the first input data, and return the search results.

[0012] Thirdly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the image search method provided in embodiments of the present invention.

[0013] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the image search method provided in the embodiments of the present invention.

[0014] In this embodiment of the invention, during a multi-round image search, first input data for the current round is acquired, and second input data from previous rounds is acquired, with the interval between the previous and current rounds being less than a preset interval. Based on the second input data, auxiliary search semantics are extracted. Based on the first input data and the auxiliary search semantics, a search is performed in the image database to obtain the search result corresponding to the first input data, and the search result is returned. This invention solves the problem of existing image search methods where each search is performed independently based on the input data of the current round, resulting in inaccurate results and low search result accuracy. Users also need to continuously search for input data, leading to low work efficiency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of an image search method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an image search device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, Figure 1 This is a flowchart of an image search method provided by an embodiment of the present invention. The image search method includes the following steps: 101. During the multi-round image search process, obtain the first input data of the current round and the second input data of the previous rounds.

[0019] In this embodiment of the invention, the image search method described above can be applied to an image search platform, which can be built on a server-based or distributed platform. The image search platform includes a data interface (for sensors or users to upload data), a knowledge database, and a knowledge database construction program. The data interface can be used to obtain the first input data of the current round and the second input data of previous rounds during multi-round image search processes. The knowledge database construction program can be used to construct the knowledge database, which is specifically designed to provide additional relational information for the identified data entities, thereby improving the depth of the data recognition system's understanding of the content.

[0020] The aforementioned multi-round image search can be understood as the system continuously receiving user input data to perform image searches during the image search process.

[0021] The first input data mentioned above can be understood as the input data for the current image search process. The first input data can be a text description of the image provided by the user in the current round. The current round mentioned above can be understood as the round in which the image search is currently being performed.

[0022] The second input data mentioned above is the input data for historical rounds of image search. This second input data can be the image text description provided by the user in previous rounds. The historical rounds can be understood as all previously completed image search rounds during the multi-round image search process.

[0023] It should be noted that due to differences in user input intent and errors in the system's understanding of that intent, not every image search will yield accurate results. Users may need to perform multiple searches for inaccurate results.

[0024] 102. Based on the second input data, extract the auxiliary search semantics.

[0025] In this embodiment of the invention, the above-mentioned auxiliary search semantics can be understood as the semantic features of auxiliary search. Keywords or category information related to image search can be extracted by analyzing the second input data of historical rounds.

[0026] Specifically, features such as frequently occurring keywords and related categories can be extracted from the second input data of historical rounds, and these features can be used as auxiliary search semantics.

[0027] 103. Based on the first input data and auxiliary search semantics, perform search processing in the image database to obtain the search results corresponding to the first input data, and return the search results.

[0028] In this embodiment of the invention, the image database described above is a database used for storing, retrieving, and managing image data.

[0029] The above search process can be understood as the process of searching in the image database based on the first input data and auxiliary search semantics during the image search process.

[0030] The search results mentioned above are obtained by image retrieval in the first image database using the first input data and auxiliary search semantics.

[0031] It should be noted that the first input data and auxiliary search semantics can be used to generate new input data, and the image database can be searched based on the new input data to obtain the search results. This invention can accurately find the search results corresponding to the user's input data, solving the problem that existing image search methods perform independent searches based on the input data of the current round, and do not always obtain accurate results, resulting in low search result accuracy and low work efficiency due to the need for users to continuously search for input data.

[0032] In this embodiment of the invention, during a multi-round image search, first input data for the current round is acquired, and second input data from previous rounds is acquired, with the interval between the previous and current rounds being less than a preset interval. Based on the second input data, auxiliary search semantics are extracted. Based on the first input data and the auxiliary search semantics, a search is performed in the image database to obtain the search result corresponding to the first input data, and the search result is returned. This invention solves the problem of existing image search methods where each search is performed independently based on the input data of the current round, resulting in inaccurate results and low search result accuracy. Users also need to continuously search for input data, leading to low work efficiency.

[0033] It is understood that in the specific implementation of this application, data such as image data, input data, knowledge data, and user data are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required. Furthermore, the collection, use, and processing of related data, as well as the training, deployment, and invocation of algorithm models, must comply with relevant laws, regulations, and standards.

[0034] Optionally, the second input data is multimodal input data. In the step of extracting auxiliary search semantics based on the second input data, semantic extraction processing can be performed on the second input data based on a multimodal large model to obtain the auxiliary search semantics corresponding to the second input data.

[0035] In this embodiment of the invention, the above-mentioned multimodal input data can be understood as input data of multiple information modalities, such as the length and complexity of the input data.

[0036] The aforementioned multimodal large models can be multimodal large models built based on deep learning or machine learning, such as CLIP, LLM, etc. These multimodal large models are capable of recognizing the semantic features of the input data.

[0037] The second input data mentioned above is the input data for image search in historical rounds. The second input data can be the image text description provided by the user in historical rounds.

[0038] The semantic extraction process described above can be understood as the process of extracting semantics from the second input data using a multimodal large model.

[0039] The aforementioned auxiliary search semantics can be semantic features of auxiliary search.

[0040] Optionally, in the step of performing search processing in the image database based on the first input data and auxiliary search semantics to obtain the search results corresponding to the first input data, semantic extraction processing can be performed on the first input data based on a multimodal large model to obtain the main search semantics of the first input data; and search processing can be performed in the image database based on the main search semantics and auxiliary search semantics to obtain the search results corresponding to the first input data.

[0041] In this embodiment of the invention, the aforementioned multimodal large model can be a multimodal large model built based on deep learning or machine learning, such as CLIP, LLM, etc. The aforementioned multimodal large model is capable of recognizing the semantic features of the input data.

[0042] The first input data mentioned above can be the input data for the current image search process, and it can be the image text description provided by the user in the current round. The current round can be understood as the round in which the image search is being performed.

[0043] The semantic extraction process described above can be understood as the process of extracting semantics from the first input data using a multimodal large model.

[0044] The aforementioned main search semantics can be understood as the main information or concepts extracted from the first input data.

[0045] The aforementioned auxiliary search semantics can be achieved by analyzing the second input data from historical rounds to extract keywords or category information related to image search.

[0046] The aforementioned image database is used for storing, retrieving, and managing image data.

[0047] The above search process can be understood as the process of searching the image database with the main search semantics and auxiliary search semantics during the image search process.

[0048] It should be noted that the main search semantics and auxiliary search semantics can be used to generate new input data, and the search results can be obtained by performing search processing in the image database based on the new input data.

[0049] Optionally, in the step of performing search processing in the image database based on the primary search semantics and the auxiliary search semantics to obtain the search results corresponding to the first input data, a first weight value corresponding to the primary search semantics and a second weight value corresponding to the auxiliary search semantics can be obtained; based on the first weight value and the second weight value, the primary search semantics and the auxiliary search semantics are weighted and added together to obtain the fused semantics; the fused semantics are then used to perform search processing in the image database to obtain the search results corresponding to the first input data.

[0050] In this embodiment of the invention, the first weight may be the weight value of the main search semantics of the first input data.

[0051] The aforementioned second weight value can be the weight value of the auxiliary search semantics of the second input data.

[0052] The weighted sum described above can be understood as the process of adding the weight values ​​of the primary search semantics to the weight values ​​of the auxiliary search semantics. This weighted sum is a method of adding the weight values ​​of different search keywords to obtain a merged search keyword.

[0053] The aforementioned fusion semantics can be understood as adding the weight values ​​of the main search semantics and the auxiliary search semantics to obtain a more accurate semantic understanding.

[0054] The aforementioned image database is used for storing, retrieving, and managing image data.

[0055] The above search process can be understood as the process of searching the image database by incorporating semantics during the image search process.

[0056] The search results described above are obtained by performing semantic fusion search processing on the image database to obtain the search results for the first input data.

[0057] It's important to note that you can assign a weight value to the primary search semantic and another to the secondary search semantic. A higher weight value indicates greater importance in the search results. For example, if the primary search semantic has a weight of 0.8 and the secondary search semantic has a weight of 0.2, then images related to the primary search semantic will be more important than those related to the secondary search semantic in the search results.

[0058] Furthermore, the weight values ​​of the primary search semantics and the auxiliary search semantics can be added together to obtain a fused search keyword.

[0059] In one possible implementation, for example, when the primary search semantic is "apple" and the secondary search semantics are "fruit" and "health," the weight value of "apple" is 0.8, the weight value of "fruit" is 0.4, and the weight value of "health" is 0.3. The fused semantic for "types of apples" is obtained by weighted summing of the weight values ​​of "apple" and "fruit"; or, the fused semantic for "nutritional value of apples" is obtained by weighted summing of the weight values ​​of "apple" and "health," etc.

[0060] Optionally, in the step of obtaining the first weight value corresponding to the main search semantic, the first initial dynamic weight value can be dynamically calculated based on the time interval between the current round and the historical round, and the time interval is negatively correlated with the first initial dynamic weight value; based on the semantic relevance between the main search semantic and the auxiliary search semantic, the first initial dynamic weight value is adjusted to obtain the first weight value corresponding to the main search semantic.

[0061] In this embodiment of the invention, the aforementioned time interval can be understood as the time interval between the current round and the historical round, and the time interval is used to describe event delay, duration, etc.

[0062] The above dynamic calculation can be understood as a calculation process that actively adjusts the weight value based on the time change status of the time interval between the current round and the historical rounds.

[0063] The aforementioned initial dynamic weight value can be understood as a dynamic weight value dynamically calculated based on the time interval between the current round and previous rounds. This time interval is negatively correlated with the initial dynamic weight value. This negative correlation can be understood as the two variables changing in opposite directions; when one variable increases, the other decreases. For example, the shorter the time interval, the larger the initial dynamic weight value; the longer the time interval, the smaller the initial dynamic weight value.

[0064] The semantic relevance between the primary search semantics and the auxiliary search semantics can be understood as the degree of semantic similarity between the primary search semantics and the auxiliary search semantics.

[0065] The first weight value mentioned above is the weight value of the main search semantics.

[0066] In one possible embodiment, the first initial dynamic weight value is calculated using exponential smoothing. The exponential smoothing formula can be used to calculate the first initial dynamic weight value: W t = (1 - α ) * W {t-1} + α * X t in, W t It is the weight value at the current time point. W {t-1} It is the weight value at the previous time point. X t This is the data value at the current time. α It is a smoothing coefficient (0 < α < 1). The initial dynamic weight value can be dynamically adjusted based on the relationship between historical data and current data.

[0067] Optionally, in the step of obtaining the second weight value corresponding to the auxiliary search semantics, the second initial dynamic weight value can be dynamically calculated based on the source round sorting of the auxiliary search semantics. The later the source round sorting is, the larger the second initial dynamic weight value is. Based on the semantic confidence of the auxiliary search semantics, the second initial dynamic weight value is adjusted based on the confidence to obtain the second weight value corresponding to the auxiliary search semantics.

[0068] In this embodiment of the invention, the source round sorting of the above-mentioned auxiliary search semantics can be obtained by sorting the source rounds of the auxiliary search semantics.

[0069] The above dynamic calculation can be understood as a calculation process that actively adjusts the weight values ​​based on the source round ranking of the auxiliary search semantics.

[0070] The aforementioned second initial dynamic weight value is dynamically calculated based on the source round ranking of the auxiliary search semantics. It can be understood that the later the source round ranking in the calculation, the larger the second initial dynamic weight value.

[0071] The semantic confidence of the above-mentioned auxiliary search semantics can be understood as the degree of semantic confidence of the auxiliary search semantics in the search results. The higher the semantic confidence, the higher the degree of semantic confidence of the auxiliary search semantics in the search results.

[0072] The second weight value mentioned above is a weight value that assists in the semantics of the search.

[0073] It should be noted that the second initial dynamic weight value can be dynamically calculated based on the source round ranking of the auxiliary search semantics. The later the source round ranking, the larger the second initial dynamic weight value. The second initial dynamic weight value is adjusted according to the semantic confidence of the auxiliary search semantics to obtain the second weight value corresponding to the auxiliary search semantics.

[0074] Optionally, in the step of performing search processing on the image database using fused semantics to obtain the search results corresponding to the first input data, the semantic similarity between the fused semantics and the image semantics corresponding to the images in the image database can be calculated; the K images with the highest semantic similarity can be determined as the search results corresponding to the first input data.

[0075] In this embodiment of the invention, the above-mentioned fused semantics can be obtained by adding the weight values ​​of the main search semantics and the weight values ​​of the auxiliary search semantics.

[0076] The aforementioned image database is used for storing, retrieving, and managing image data.

[0077] The semantic similarity mentioned above can be understood as the degree of semantic similarity between the fused semantics and the corresponding image semantics of the images in the image database. The semantic similarity between the fused semantics and the corresponding image semantics of the images in the image database can be calculated using similarity calculation methods such as cosine similarity and Euclidean distance. The greater the similarity, the greater the semantic similarity; the smaller the similarity, the smaller the semantic similarity.

[0078] The K above is a quantifier, which can be 10, 15, 20, etc.

[0079] Specifically, the semantic similarity between the fused semantics and the corresponding image semantics in the image database can be calculated using cosine similarity. The images can be sorted from high to low based on their semantic similarity, and the K images with the highest semantic similarity can be determined as the search results corresponding to the first input data.

[0080] like Figure 2As shown, an embodiment of the present invention provides an image search device, which includes: The acquisition module 201 is used to acquire the first input data of the current round and the second input data of the historical round during the multi-round image search process, wherein the interval between the historical round and the current round is less than a preset interval. Extraction module 202 is used to extract auxiliary search semantics based on the second input data; The search module 203 is used to perform search processing in the image database based on the first input data and the auxiliary search semantics, obtain the search results corresponding to the first input data, and return the search results.

[0081] Optionally, the extraction module 202 is further configured to perform semantic extraction processing on the second input data based on a multimodal large model to obtain the auxiliary search semantics corresponding to the second input data.

[0082] Optionally, the search module 203 is further configured to perform semantic extraction processing on the first input data based on the multimodal large model to obtain the main search semantics of the first input data; and perform search processing in the image database based on the main search semantics and the auxiliary search semantics to obtain the search results corresponding to the first input data.

[0083] Optionally, the search module 203 is further configured to obtain a first weight value corresponding to the main search semantic and a second weight value corresponding to the auxiliary search semantic; based on the first weight value and the second weight value, the main search semantic and the auxiliary search semantic are weighted and added together to obtain a fused semantic; the fused semantic is used for search processing in the image database to obtain the search results corresponding to the first input data.

[0084] Optionally, the search module 203 is further configured to dynamically calculate a first initial dynamic weight value based on the time interval between the current round and the historical round, wherein the time interval is negatively correlated with the first initial dynamic weight value; and adjust the first initial dynamic weight value based on the semantic relevance between the main search semantic and the auxiliary search semantic to obtain a first weight value corresponding to the main search semantic.

[0085] Optionally, the search module 203 is further configured to dynamically calculate a second initial dynamic weight value based on the source round sorting of the auxiliary search semantics, wherein the later the source round sorting is, the larger the second initial dynamic weight value is; calculate the semantic confidence of the auxiliary search semantics, and adjust the second initial dynamic weight value based on the confidence to obtain the second weight value corresponding to the auxiliary search semantics.

[0086] Optionally, the search module 203 is further configured to calculate the semantic similarity between the fused semantics and the image semantics corresponding to the images in the image database; and to determine the K images with the highest semantic similarity as the search results corresponding to the first input data.

[0087] like Figure 3 As shown, this embodiment of the invention also provides an electronic device, including a processor, which can execute any of the above-described image search methods.

[0088] Specifically, it includes a processor 301 and a memory 302, as well as a computer program stored in the memory 302 and capable of running on the processor 301 to execute the image search method, wherein: Processor 301 executes the calculator program for the image search method stored in memory 302, performing the following steps: During the multi-round image search process, the first input data of the current round is obtained, and the second input data of the historical rounds is obtained, wherein the interval between the historical rounds and the current round is less than a preset interval. Based on the second input data, auxiliary search semantics are extracted; Based on the first input data and the auxiliary search semantics, a search process is performed in the image database to obtain the search results corresponding to the first input data, and the search results are returned.

[0089] Optionally, the second input data is multimodal input data, and the step of extracting auxiliary search semantics based on the second input data executed by the processor 301 includes: Based on a multimodal large model, semantic extraction processing is performed on the second input data to obtain the auxiliary search semantics corresponding to the second input data.

[0090] Optionally, the process executed by processor 301, which involves performing a search in the image database based on the first input data and the auxiliary search semantics to obtain the search results corresponding to the first input data, includes: Based on the multimodal large model, semantic extraction processing is performed on the first input data to obtain the main search semantics of the first input data; Based on the primary search semantics and the auxiliary search semantics, a search process is performed in the image database to obtain the search results corresponding to the first input data.

[0091] Optionally, the process executed by processor 301, which involves performing a search in the image database based on the primary search semantics and the secondary search semantics to obtain the search results corresponding to the first input data, includes: Obtain the first weight value corresponding to the primary search semantic and the second weight value corresponding to the auxiliary search semantic; Based on the first weight value and the second weight value, the main search semantics and the auxiliary search semantics are weighted and added together to obtain the fused semantics; The fusion semantics are searched in the image database to obtain the search results corresponding to the first input data.

[0092] Optionally, the step of obtaining the first weight value corresponding to the main search semantic, performed by processor 301, includes: The first initial dynamic weight value is dynamically calculated based on the time interval between the current round and the historical round, and the time interval is negatively correlated with the first initial dynamic weight value. Based on the semantic relevance between the primary search semantic and the auxiliary search semantic, the first initial dynamic weight value is adjusted to obtain the first weight value corresponding to the primary search semantic.

[0093] Optionally, the step of obtaining the second weight value corresponding to the auxiliary search semantics executed by processor 301 includes: The second initial dynamic weight value is dynamically calculated based on the source round sorting of the auxiliary search semantics. The later the source round sorting is, the larger the second initial dynamic weight value is. Based on the semantic confidence calculation of the auxiliary search semantics, the second initial dynamic weight value is adjusted based on the confidence value to obtain the second weight value corresponding to the auxiliary search semantics.

[0094] Optionally, the fusion semantics executed by processor 301 performs search processing in the image database to obtain search results corresponding to the first input data, including: Calculate the semantic similarity between the fused semantics and the image semantics corresponding to the images in the image database; The K images with the highest semantic similarity are selected as the search results corresponding to the first input data.

[0095] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the image search method provided in this invention and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0096] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0097] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. An image search method, characterized in that, The method includes the following steps: During the multi-round image search process, the first input data of the current round is obtained, and the second input data of the historical rounds is obtained, wherein the interval between the historical rounds and the current round is less than a preset interval. Based on the second input data, auxiliary search semantics are extracted; Based on the first input data and the auxiliary search semantics, a search process is performed in the image database to obtain the search results corresponding to the first input data, and the search results are returned.

2. The image search method as described in claim 1, characterized in that, The second input data is multimodal input data. The step of extracting auxiliary search semantics based on the second input data includes: Based on a multimodal large model, semantic extraction processing is performed on the second input data to obtain the auxiliary search semantics corresponding to the second input data.

3. The image search method as described in claim 2, characterized in that, The step of performing search processing in the image database based on the first input data and the auxiliary search semantics to obtain the search results corresponding to the first input data includes: Based on the multimodal large model, semantic extraction processing is performed on the first input data to obtain the main search semantics of the first input data; Based on the primary search semantics and the auxiliary search semantics, a search process is performed in the image database to obtain the search results corresponding to the first input data.

4. The image search method as described in claim 3, characterized in that, The search process, based on the primary search semantics and the auxiliary search semantics, performs search processing in the image database to obtain the search results corresponding to the first input data, including: Obtain the first weight value corresponding to the primary search semantic and the second weight value corresponding to the auxiliary search semantic; Based on the first weight value and the second weight value, the main search semantics and the auxiliary search semantics are weighted and added together to obtain the fused semantics; The fusion semantics are searched in the image database to obtain the search results corresponding to the first input data.

5. The image search method as described in claim 4, characterized in that, Obtaining the first weight value corresponding to the main search semantic includes: The first initial dynamic weight value is dynamically calculated based on the time interval between the current round and the historical round, and the time interval is negatively correlated with the first initial dynamic weight value. Based on the semantic relevance between the primary search semantic and the auxiliary search semantic, the first initial dynamic weight value is adjusted to obtain the first weight value corresponding to the primary search semantic.

6. The image search method as described in claim 5, characterized in that, The step of obtaining the second weight value corresponding to the auxiliary search semantic includes: The second initial dynamic weight value is dynamically calculated based on the source round sorting of the auxiliary search semantics. The later the source round sorting is, the larger the second initial dynamic weight value is. Based on the semantic confidence calculation of the auxiliary search semantics, the second initial dynamic weight value is adjusted based on the confidence value to obtain the second weight value corresponding to the auxiliary search semantics.

7. The image search method according to any one of claims 4 to 6, characterized in that, The fusion semantics are searched in the image database to obtain the search results corresponding to the first input data, including: Calculate the semantic similarity between the fused semantics and the image semantics corresponding to the images in the image database; The K images with the highest semantic similarity are selected as the search results corresponding to the first input data.

8. An image search device, characterized in that, The image search device includes: The acquisition module is used to acquire the first input data of the current round and the second input data of the historical round during the multi-round image search process, wherein the interval between the historical round and the current round is less than a preset interval. The extraction module is used to extract auxiliary search semantics based on the second input data; The search module is used to perform search processing in the image database based on the first input data and the auxiliary search semantics, obtain the search results corresponding to the first input data, and return the search results.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the image search method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the image search method as described in any one of claims 1 to 7.

Images