Image retrieval method and device, electronic device and readable storage medium

Abstract

This application provides an image retrieval method, apparatus, electronic device, and readable storage medium, relating to the field of image retrieval technology. The image retrieval method provided by this application includes: responding to a user's operation of selecting multiple target algorithm systems to retrieve an image on a first interface, sending the image to be retrieved to the multiple target algorithm systems; receiving retrieval results sent by the multiple target algorithm systems, and determining the fusion similarity of reference images in the retrieval results; based on the fusion similarity, determining multiple first target images from the reference images corresponding to the multiple target algorithm systems; and displaying a second interface, the second interface including the multiple first target images, and the identification information and fusion similarity of the target algorithm system corresponding to each first target image. The image retrieval method provided by this application can uniformly schedule multiple algorithm systems to retrieve the image to be retrieved, thereby improving the accuracy of image retrieval.

Description

Technical Field

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

[0002] Image retrieval is a search technology that retrieves similar images by inputting a target image. It is widely used in many fields, providing users with the ability to search for relevant graphic and image data. For example, in the AI ​​(Artificial Intelligence) security field, image retrieval functions for faces, bodies, motor vehicles, and non-motor vehicles are commonly used. Each algorithm vendor possesses independent computing, storage, and network resources, deploying its own algorithm systems to provide services. However, a single algorithm vendor cannot truly cover all industries and all scenarios with its algorithms. In practical applications, to improve the accuracy of image retrieval, it is often necessary to break down image retrieval requirements into multiple algorithm systems from multiple vendors.

[0003] Because each algorithm system uses different internal algorithms and rules for image retrieval, the search results from each system are not compatible and vary in accuracy. Currently, the common practice is for users to input the same criteria on different algorithm systems, retrieve corresponding images, and then manually compare them. However, when using multiple algorithm systems for image retrieval, users need to constantly switch between platforms provided by various vendors, significantly impacting the user experience and making it impossible to directly compare the merits of different vendors' algorithms.

[0004] Therefore, how to use multiple algorithm systems for image retrieval to improve the accuracy of image retrieval has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides an image retrieval method, apparatus, electronic device, and readable storage medium, which enables unified scheduling of multiple algorithm systems for image retrieval, thereby improving the accuracy of image retrieval.

[0006] The first aspect of this application provides an image retrieval method, comprising: responding to a user's operation of selecting multiple target algorithm systems to retrieve an image on a first interface, sending the image to be retrieved to the multiple target algorithm systems, so that the multiple target algorithm systems can retrieve the image using their respective retrieval algorithms; receiving retrieval results sent by the multiple target algorithm systems, and determining the fusion similarity of reference images in the retrieval results; wherein, the reference images are images retrieved by the target algorithm systems, and the fusion similarity is determined based on a benchmark similarity; the benchmark similarity is the similarity between the reference image and the image to be retrieved calculated by the benchmark algorithm system; determining multiple first target images from the reference images corresponding to the multiple target algorithm systems based on the fusion similarity; displaying a second interface, the second interface including multiple first target images, and the identification information and fusion similarity of the target algorithm system corresponding to each first target image; wherein, one first target image corresponds to one or more target algorithm systems.

[0007] The image retrieval method provided in this application, when a user selects multiple target algorithm systems to retrieve an image on a first interface, sends the image to be retrieved to multiple target algorithm systems for retrieval. Then, it determines the fusion similarity of reference images in the retrieval results of multiple target algorithm systems, and selects the first target image from the retrieval results based on the fusion similarity, and displays it accordingly. In this way, when a user wants to use multiple algorithm systems for image retrieval, they do not need to switch between algorithm systems; they can simply select the desired target algorithm system, and multiple algorithm systems can be uniformly scheduled to retrieve the image, thereby improving the accuracy of image retrieval.

[0008] In conjunction with the first implementation method of the first aspect, if a first target image corresponds to multiple target algorithm systems, the second interface displays the identification information of the multiple target algorithm systems corresponding to each first target image.

[0009] In conjunction with the second implementation method of the first aspect, the second interface also includes: a display control for each target algorithm system; the method further includes: after detecting the user's trigger operation on the display control of the target algorithm system, displaying a third interface, the third interface including multiple second target images and reference similarities corresponding to the second target images, the second target images being determined based on the reference images and reference similarities in the search results corresponding to the target algorithm system.

[0010] Combining the third implementation method of the first aspect, the steps for determining the fusion similarity of reference images in the search results include: converting the reference similarity of reference images in the search results into benchmark similarity, where the reference similarity is the similarity between the reference image and the image to be searched calculated by the corresponding target algorithm system; merging search results with the same image identifier, where the merged search results include multiple benchmark similarities, each benchmark similarity corresponding to a target algorithm system; and determining the fusion similarity based on the similarity value rules and multiple benchmark similarities.

[0011] Combining the fourth implementation method of the first aspect, the steps of converting the reference similarity of reference images in the search results into benchmark similarity include: obtaining a benchmark algorithm system; determining the similarity conversion rules between the target algorithm system and the benchmark algorithm system; the similarity conversion rules include the correspondence between similarity intervals and conversion algorithms; the conversion algorithm is used to convert the reference similarity of the reference images in the search results corresponding to the target algorithm system into the benchmark similarity corresponding to the benchmark algorithm system; determining the conversion algorithm corresponding to the search results based on the similarity interval in the search results of the target algorithm system; and converting the reference similarity of the reference images in the search results into benchmark similarity based on the conversion algorithm corresponding to the search results.

[0012] Combining the fifth implementation method of the first aspect, the identification information of the target algorithm system is a color identifier, and the background color of the second interface that integrates similarity is the same as the color of the identification information of the benchmark algorithm system.

[0013] A second aspect of this application provides an image retrieval device, comprising: a sending module for sending the image to be retrieved to a plurality of target algorithm systems, so that the plurality of target algorithm systems can retrieve the image to be retrieved using their respective retrieval algorithms; a determining module for receiving retrieval results sent by the plurality of target algorithm systems and determining the fusion similarity of reference images in the retrieval results; wherein the reference images are images retrieved by the target algorithm systems, and the fusion similarity is determined based on a benchmark similarity; the benchmark similarity is a similarity based on a benchmark algorithm system; the determining module is further configured to determine a plurality of first target images from the reference images corresponding to the plurality of target algorithm systems based on the fusion similarity; and a display module for displaying a second interface, the second interface including the plurality of first target images, and the identification information of the target algorithm system corresponding to each first target image and the fusion similarity; wherein one first target image corresponds to one or more target algorithm systems.

[0014] In conjunction with the first implementation method of the second aspect, if a first target image corresponds to multiple target algorithm systems, the second interface displays the identification information of the multiple target algorithm systems corresponding to each first target image;

[0015] The second interface also includes: display controls for each target algorithm system; the display module is also used to display a third interface, which includes multiple second target images and reference similarities corresponding to the second target images. The second target images are determined based on the reference images and reference similarities in the search results corresponding to the target algorithm system.

[0016] The module is specifically used to convert the reference similarity of reference images in the search results into benchmark similarity, where the reference similarity is the similarity between the reference image and the image to be searched calculated by the corresponding target algorithm system; merge search results with the same image identifier, the merged search results include multiple benchmark similarities, each benchmark similarity corresponds to a target algorithm system; and determine the fusion similarity based on the similarity value rules and multiple benchmark similarities.

[0017] The module is specifically used to: acquire the benchmark algorithm system; determine the similarity conversion rules between the target algorithm system and the benchmark algorithm system; the similarity conversion rules include the correspondence between similarity intervals and conversion algorithms; the conversion algorithm is used to convert the reference similarity of the reference image in the search results corresponding to the target algorithm system into the benchmark similarity corresponding to the benchmark algorithm system; determine the conversion algorithm corresponding to the search results based on the similarity interval of the reference similarity in the search results of the target algorithm system; and convert the reference similarity of the reference image in the search results into the benchmark similarity based on the conversion algorithm corresponding to the search results.

[0018] The target algorithm system is identified by color, and the background color of the similarity fusion in the second interface is the same as the color of the identification information of the benchmark algorithm system.

[0019] A third aspect of this application provides an electronic device, including: one or more processors; one or more memories; wherein the one or more memories are used to store computer program code, the computer program code including computer instructions, and when the one or more processors execute the computer instructions, the electronic device performs the above-described image retrieval method.

[0020] The fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed on a computer, cause the computer to perform the aforementioned image retrieval method.

[0021] The beneficial effects described in aspects two through four can be referred to the analysis of the beneficial effects in aspect one, and will not be repeated here. Attached Figure Description

[0022] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of the present invention and do not constitute a limitation on the technical solutions of the present invention.

[0023] Figure 1 An overall architecture diagram of a cluster system provided in this application embodiment;

[0024] Figure 2 A flowchart of an image retrieval method provided in this application embodiment;

[0025] Figure 3 A flowchart illustrating a method for determining fusion similarity provided in this application embodiment;

[0026] Figure 4 A flowchart illustrating a method for determining benchmark similarity provided in this application embodiment;

[0027] Figure 5 A similarity diagram of search results provided for an embodiment of this application;

[0028] Figure 6 A schematic diagram of useful material data provided in an embodiment of this application;

[0029] Figure 7 A schematic diagram of a similarity conversion training algorithm provided in an embodiment of this application;

[0030] Figure 8 A schematic diagram illustrating a merged search result provided in an embodiment of this application;

[0031] Figure 9 This is a schematic diagram illustrating the ranking and sorting of search results according to an embodiment of this application.

[0032] Figure 10 A schematic diagram of a first interface provided in an embodiment of this application;

[0033] Figure 11 This application provides an embodiment of a schematic diagram illustrating the use of search results as cached data.

[0034] Figure 12 This is a schematic diagram illustrating the interception of cached data provided in an embodiment of this application;

[0035] Figure 13 This is a schematic diagram illustrating another method for intercepting cached data, as provided in an embodiment of this application.

[0036] Figure 14 This is a schematic diagram of the structure of an image retrieval device provided in an embodiment of this application;

[0037] Figure 15 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0040] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "connected" and "linked" should be interpreted broadly, for example, as a fixed connection, a detachable connection, or an integral connection. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, when describing pipelines, the terms "connected" and "linked" as used in this application have the meaning of establishing electrical connection. The specific meaning needs to be understood in conjunction with the context.

[0041] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0042] To address the issue of the inability to uniformly schedule multiple algorithm systems for image retrieval, this application provides an image retrieval method. This method identifies a target algorithm system based on at least one algorithm system identifier carried in the retrieval request, and then sends the image to be retrieved, carried in the retrieval request, to the target algorithm system. Each of the multiple target algorithm systems then uses its respective retrieval algorithm to retrieve the image. Upon receiving retrieval results from multiple target algorithm systems, the method determines the fusion similarity of reference images within the retrieval results. Since the fusion similarity is determined based on a unified standard benchmark similarity, the target image can be identified from the reference images corresponding to the multiple target algorithm systems based on the fusion similarity, thus enabling the unified display of the target image.

[0043] This way, when users want to use multiple algorithm systems for image retrieval, they don't need to switch between algorithm systems. They can simply select the target algorithm system they want to search for, and the search results from multiple algorithm systems will be presented on a single interface. Users can more intuitively compare the search results from multiple algorithm systems, thus improving the user experience.

[0044] It should be understood that the image retrieval method provided in this application embodiment can be applied to an electronic device or to a system consisting of at least two electronic devices to achieve the above method through information interaction between the devices.

[0045] In this application embodiment, the electronic device can be a terminal device, such as a personal computer (PC), laptop computer, mobile device, tablet computer, etc. This application embodiment does not limit the specific form of the electronic device. Alternatively, the electronic device can also be a single server or a server cluster composed of multiple servers. In some implementations, the server cluster can be a distributed cluster server.

[0046] Please see Figure 1 , Figure 1 For an overall architecture diagram of a cluster system provided in this application embodiment, please refer to [link / reference]. Figure 1 The cluster system provided in this application embodiment includes: a multi-algorithm application system 100, a multi-algorithm integration system 200, and multiple algorithm systems 300.

[0047] It should be understood that the image retrieval method provided in this application embodiment can be applied to the multi-algorithm integration system 200, or to a system composed of the multi-algorithm application system 100 and the multi-algorithm integration system 200, and the image retrieval method can be realized through information interaction between the systems.

[0048] Among them, the multi-algorithm application system 100 can uniformly schedule and manage multiple algorithm systems 300 through standard protocols of recognized organizations, and is a system that can use multiple algorithms to realize functions such as video image content analysis and description, retrieval, etc. For example... Figure 1As shown, the multi-algorithm application system 100 includes a retrieval request arrangement module 101 and an image retrieval display module 102. The retrieval request arrangement module 101 supports users in arranging retrieval requests, such as uploading images to be retrieved, selecting a retrieval time, selecting a retrieval range, setting the minimum similarity of the retrieval targets, selecting multiple algorithm systems, sorting rules, setting the benchmark vendor for similarity conversion (optional operation), and setting the similarity interval for image retrieval from each algorithm system (optional operation). After the retrieval request is arranged, it is sent to the multi-algorithm integration system 200. The image retrieval display module 102 receives the target data returned by the multi-algorithm integration system 200 and displays the fusion effect of the target data on the same interface according to reasonable rules.

[0049] The multi-algorithm integration system 200 is essentially a subset of the multi-algorithm application system 100, but it is a system separated from the multi-algorithm application system specifically responsible for protocol interfacing and data forwarding. For example... Figure 1 As shown, the multi-algorithm integration system 200 includes: a task scheduling module 201, a data filtering module 202, a similarity conversion module 203, a data caching module 204, a data merging module 205, a secondary data merging module 206, a secondary data filtering module 207, a data sorting module 208, and a data slicing module 209.

[0050] The task scheduling module 201 is used to receive the retrieval request sent by the multi-algorithm application system 100 and send the retrieval request to the target algorithm system among the multiple algorithm systems 300 according to the algorithm system identifier in the retrieval request.

[0051] The data filtering module 202 is used to filter the search results corresponding to each target algorithm system according to the filtering rules corresponding to each target algorithm system carried in the search request when receiving search results returned by multiple target algorithm systems.

[0052] The similarity conversion module 203 is used to convert the similarity of the search results returned by multiple target algorithm systems into a baseline similarity.

[0053] The data caching module 204 is used to store the search results after similarity conversion as cached data.

[0054] The data merging module 205 is used to merge cached data with the same image identifier, and the merged cached data includes multiple similarities.

[0055] The data secondary merging module 206 is used to merge multiple similarities of cached data so that each cached data corresponds to a similarity in each algorithm system.

[0056] The secondary data filtering module 207 is used to process the similarity of each data in the cached data according to the similarity value rules to obtain the fusion similarity of the cached data; and then filter the cached data according to the fusion similarity.

[0057] The data sorting module 208 is used to sort the cached data according to sorting rules and fusion similarity.

[0058] The data slicing module 209 is used to extract target data from cached data according to the display rules carried by the retrieval request and output the target data to the algorithm application system 100.

[0059] It should be noted that, Figure 1 This embodiment merely illustrates an application scenario of multiple algorithm systems 300, including three algorithm systems: Algorithm System A, Algorithm System B, and Algorithm System C. The image retrieval algorithms used by each algorithm system include, but are not limited to, face retrieval algorithms and body retrieval algorithms. However, in actual use, the number of algorithm systems and the algorithms included in each algorithm system can be adjusted according to user needs and actual circumstances; this application does not impose any limitations in this regard.

[0060] Figure 2 The diagram shown is a flowchart illustrating an image retrieval method provided in an embodiment of this application. Figure 2 As shown, the image retrieval method provided in this application embodiment includes S201-S204:

[0061] S201. In response to the user's operation of selecting multiple target algorithm systems to search for the image on the first interface, the image to be searched is sent to multiple target algorithm systems so that the multiple target algorithm systems can use their respective search algorithms to search for the image.

[0062] When a user needs to search for images, they can select multiple target algorithm systems on the first interface to search for the image. The image to be searched is then sent to the user-selected target algorithm systems, allowing each system to use its own search algorithm to retrieve the image. This way, users don't need to switch between multiple algorithm vendors' platforms; they can simply use the first interface provided in this application to select and use multiple algorithm systems for image retrieval, significantly improving the user experience.

[0063] It should be understood that each target algorithm system has multiple retrieval algorithms. In one implementation, before a user performs an image search on the first interface, they need to select the image type to be searched. Then, based on the image type selected by the user, the retrieval algorithm used by the target algorithm system is determined. For example, if the user determines that the image type to be searched is a face, multiple target algorithm systems will use their respective face recognition algorithms to search for the image.

[0064] S202. Receive retrieval results sent by multiple target algorithm systems and determine the fusion similarity of reference images in the retrieval results.

[0065] The reference image is the image retrieved by the target algorithm system, and the fusion similarity is determined based on the benchmark similarity; the benchmark similarity is the similarity based on the benchmark algorithm system.

[0066] Since each target algorithm system has different internal algorithms and rules when performing image retrieval, the similarity calculation standards of the reference images in the retrieval results of each algorithm system are different, which are not compatible with each other and there are differences in algorithm accuracy. Therefore, it is necessary to first convert the similarity of the reference images of each retrieval result into a unified standard benchmark similarity, and then determine a fusion similarity based on the benchmark similarity.

[0067] As one feasible approach, please refer to Figure 3 The step of determining the fusion similarity of reference images in the search results may include S301-S303:

[0068] S301. Convert the reference similarity of the reference images in the search results into the baseline similarity.

[0069] The reference similarity is the similarity between the reference image calculated by the target algorithm system and the image to be retrieved. The reference similarity of the reference image for each retrieval result is converted into a unified standard benchmark similarity, and then the fusion similarity is determined based on the benchmark similarity.

[0070] As one feasible approach, please refer to Figure 4 S301 can be specifically implemented as S401-S404:

[0071] S401, Obtain the benchmark algorithm system.

[0072] It should be understood that the first interface may include a control for selecting a benchmark algorithm system from multiple algorithm systems. In some embodiments, when a user arranges a search task on the first interface, they can select a benchmark algorithm system from the multiple algorithm systems. The user-selected algorithm system is then used as the benchmark algorithm system, converting the reference similarity in the search results of other algorithm systems into a similarity based on the user-selected benchmark algorithm system. In other embodiments, if the user does not select a benchmark algorithm system from the multiple algorithm systems when arranging a search task on the first interface, any one of the multiple algorithm systems can be used as the benchmark algorithm system.

[0073] S402. Determine the similarity conversion rules between the target algorithm system and the benchmark algorithm system.

[0074] S403. Based on the similarity conversion rules corresponding to each target algorithm system, convert the reference similarity of the reference images in the search results into the baseline similarity.

[0075] For example, if the target algorithm system includes algorithm system A, algorithm system B, and algorithm system C, and algorithm system A is used as the baseline algorithm system, then the similarity scores in the search results of algorithm systems B and C need to be converted using a similarity conversion method. Please refer to [link to relevant documentation]. Figure 5 After converting the similarity scores in the search results of the algorithm system to the similarity scores corresponding to the benchmark algorithm system, each search result corresponds to two similarity scores. Here, `oldSimilarity` represents the similarity score retrieved by the original algorithm system, and `newSimilarity` represents the converted similarity score corresponding to the benchmark algorithm system. Please continue reading. Figure 5 Since algorithm system A is the baseline algorithm system, the oldSimilarity and newSimilarity in the search results of algorithm system A are the same after conversion; the newSimilarity obtained after conversion of the similarity in the search results of algorithm system B is 0.912524; and the newSimilarity obtained after conversion of the similarity in the search results of algorithm system C is 0.883452.

[0076] It should be noted that the training data for the algorithm in this application comes from two sources: one is the retrieval results from multiple algorithm systems; the other is the retrieval results obtained by periodically retrieving a certain number of images in the background and calling the retrieval functions of multiple algorithm systems. The multi-algorithm integration system saves the data in a database. Table 1 shows an example of the data collected in this application stored in the database table. Please refer to Table 1. The database table can store the algorithm system identifier, retrieval algorithm, retrieved image ID, image ID, similarity value, and retrieval time for each piece of data.

[0077] Table 1 Database Tables

[0078]

[0079] The background of the multi-algorithm integration system 200 can periodically clear out useless and duplicate material data, and then perform pairwise matching on the material data to obtain useful material data. Here, algorithm system A is temporarily used as the baseline algorithm system, and other algorithm systems are aligned with algorithm system A.

[0080] Please see Figure 6The obtained useful sets of source data are {Algorithm System B, Algorithm System A}[0.928, 0.931], {Algorithm System C, Algorithm System A}[0.985, 0.993], [0.921, 0.927]. The backend of the multi-algorithm integration system 200 can train the similarity conversion formula based on these source data on a daily schedule.

[0081] Please see Figure 7 Using algorithm system A as the baseline algorithm system, the formula for similarity conversion of algorithm system C obtained by training with univariate linear regression is Y = 1.010466456413269X - 0.0026736529543995857. That is to say, if the similarity in the search results of algorithm system C is 0.866, the similarity obtained after similarity conversion is 0.872.

[0082] It should be understood that since the univariate linear regression algorithm can utilize a large number of samples (known data) to generate a fitting equation, thereby predicting unknown data, the more samples there are, the more accurate the prediction of unknown data will be. Therefore, as a feasible implementation method, the similarity conversion rule can adopt the univariate linear regression algorithm. However, the embodiments of this application do not limit the algorithm of the similarity conversion rule. As long as it meets the business requirements, there is no problem in using other algorithms.

[0083] It should be understood that in this embodiment, a conversion algorithm is trained for all algorithm systems across multiple algorithm systems. This ensures that regardless of which algorithm system the user selects as the baseline algorithm system, other algorithm systems have corresponding similarity conversion rules. Thus, when a user performs a search, even if the baseline algorithm system is changed, the similarity in the search results of each algorithm system can be converted, thereby enabling a better comparison of the search results from multiple algorithm systems.

[0084] In some embodiments, a conversion algorithm can be used when performing similarity conversion so that the similarity of the retrieval results of different algorithms has a unified benchmark. In order to improve the accuracy of algorithm training, the algorithm training can be divided into multiple segments. The trained algorithm can automatically learn and calculate, and simulate training can be performed on the critical value of the similarity of different algorithm systems to find the threshold interval where the fitting degree is infinitely close to 1, that is, the similarity interval with the smallest error, thereby realizing the conversion of similarity in multiple segments.

[0085] For example, please refer to Table 2, which shows a similarity conversion self-learning method provided in an embodiment of this application. Taking the material data of algorithm system C as being divided into two segments according to similarity as an example, the similarity conversion self-learning method in Table 2 is used to find the threshold interval with the best fit of 1, that is, the similarity interval with the smallest error, thereby realizing the conversion of similarity in multiple segments.

[0086] Table 2. Similarity conversion self-learning method

[0087]

[0088] As shown in Table 2, on the first day, if it is the first training, the source data of algorithm system C is divided into two segments according to the similarity of the source data. The first segment of source data falls within the similarity range of [0-0.5]. The similarity of the first segment is shifted 100 times to the left, decreasing by 0.001 each time, to find the similarity corresponding to the closest fit of 1. The second segment of source data falls within the similarity range of [0.5-1]. The similarity of the second segment is shifted 100 times to the right, increasing the threshold by 0.001 each time, to find the similarity corresponding to the closest fit of 1. Finally, the closest similarity to 1 is found between the first and second segments of source data; here, we assume it to be 0.6. On the second day, the source data of algorithm system C can be divided into two segments [0-0.6] and [0.6-1] based on 0.6 for algorithm training. The similarity of the first segment of source data is shifted to the left 100 times, with the threshold decreasing by 0.001 each time, to find the similarity with the closest fit to 1. The similarity of the second segment of source data is shifted to the left 100 times, with the threshold increasing by 0.001 each time, to find the similarity with the closest fit to 1. Finally, the closest similarity to 1 is found in the algorithms trained on both sides, which is assumed to be 0.7 here... On each subsequent day, similarity transformation self-learning is performed based on the source data of algorithm system C to make the obtained similarity interval more accurate, thereby making the similarity transformation result more accurate and reliable.

[0089] As a feasible implementation method, the similarity conversion rule includes the correspondence between similarity intervals and conversion algorithms; the conversion algorithm is used to convert the reference similarity of the target algorithm system corresponding to the reference image in the retrieval results into the benchmark similarity corresponding to the benchmark algorithm system. S403 can be specifically implemented as (11) and (12):

[0090] (11) Determine the conversion algorithm corresponding to the search results based on the similarity interval of the reference similarity in the search results.

[0091] (12) Based on the conversion algorithm corresponding to the search results, convert the reference similarity of the reference images in the search results into the baseline similarity.

[0092] The algorithm training is divided into multiple segments, and the corresponding conversion algorithm is determined according to the similarity interval of the reference similarity in the retrieval results, so that the similarity conversion results are more accurate and reliable.

[0093] Please refer to Table 2. Since the similarity conversion self-learning method yields a value of 0.7 on the second day, when converting the similarity in the search results of algorithm system C to the similarity in algorithm system A on the third day, the search results of algorithm system C can be divided into two segments for similarity conversion. One conversion algorithm is used when the similarity in the search results of algorithm system C is between 0 and 0.7, and another conversion algorithm is used when the similarity in the search results is between 0.7 and 1. In this way, the corresponding conversion algorithm can be determined based on the similarity range in the search results, making the similarity conversion results more accurate and reliable.

[0094] S302. Merge search results with the same image identifier.

[0095] Search results with the same image identifier are merged, so that the merged search results include multiple benchmark similarities, and each benchmark similarity corresponds to a target algorithm system.

[0096] Because each algorithm in the target algorithm system may retrieve duplicate reference images when searching for the image, meaning the same reference image is retrieved by multiple algorithm systems simultaneously, failing to merge these duplicate reference images will inevitably affect the user's viewing experience.

[0097] As a feasible implementation method, image identifiers can serve as Uniform Resource Locators (URLs) for cached data. A URL is a concise representation of the location and access method of a resource obtainable from the internet; it is the address of a standard resource on the internet. Every file on the internet has a unique URL, containing information indicating the file's location and how the browser should handle it. In other words, each search result has its corresponding URL. If the URLs are the same, it indicates that the reference images in the two search results are identical. Search results with the same URL are merged, and the merged search results include multiple benchmark similarities to ensure that there are no duplicate images in the merged search results.

[0098] As another feasible implementation method, the image identifier can be an image ID. Each search result has its corresponding image ID. If two search results have the same image ID, it means that the images corresponding to the two search results are the same. The search results with the same image ID are merged. The merged search results include multiple benchmark similarities, so that there are no duplicate images in the merged search results.

[0099] For example, please refer to Figure 8Starting from the first search result and working rightwards, the results are compared. If data with the same image ID exists in the search results, it is merged into one data entry. Figure 8 As shown, the search results of algorithm system A have the same image ID as the search results of algorithm system B and algorithm system C (data number 41 in the figure). Therefore, the duplicate data in the search results of algorithm system B and algorithm system C will be removed. The search result at number 124 will then include the baseline similarity in the search results of algorithm system A, algorithm system B, and algorithm system C. If there are other algorithm systems with the same image, they will be merged, and an additional similarity value will be added.

[0100] Because merging search results with the same image identifier can result in two scenarios: one is that the same algorithm system may produce multiple identical baseline similarities; the other is that the same algorithm system may produce different baseline similarities.

[0101] As a feasible approach, if multiple identical baseline similarities appear in a single search result from the same algorithm system, duplicate similarity data can be directly removed.

[0102] For example, if multiple benchmark similarity data corresponding to a single search result are shown below:

[0103]

[0104] Since the search results contain baseline similarity data for two identical algorithm systems B, the duplicate baseline similarity data is directly deleted. After deletion, the corresponding baseline similarity data in the search results are shown below:

[0105]

[0106] As another feasible approach, when different baseline similarities exist within the same algorithm system, the data with lower similarity can be filtered out, and the data with higher baseline similarity can be selected.

[0107] For example, if multiple benchmark similarity data corresponding to a single search result are shown below:

[0108]

[0109]

[0110] Since the search results contain baseline similarity data for two different algorithm systems B, the baseline similarity data with the lower similarity between the two algorithm systems B is directly deleted. After deletion, the corresponding baseline similarity data in the search results are as follows:

[0111]

[0112] Search results with the same image identifier are merged to ensure no duplicate images exist in the merged results. Then, multiple baseline similarity scores are combined, so that each search result corresponds to a specific baseline similarity score in each target algorithm system. This process ensures that there are no duplicate images in the search results and that the output target data is also free of duplicate data, preventing users from seeing duplicate search results. Furthermore, users can intuitively see which algorithm systems retrieved the same image and directly compare the similarity scores calculated by each algorithm system, thus improving the user experience.

[0113] S303. Determine the fusion similarity based on the similarity value rules and multiple benchmark similarities.

[0114] After merging the search results in S302, a single search result may include the baseline similarity scores of multiple algorithm systems. However, when displaying data, multi-algorithm application systems require a certain similarity criterion, and the baseline similarity scores for each search result may be for different algorithm systems. For example, cached data 'a' includes the similarity score of algorithm system A, while cached data 'b' includes the similarity scores of algorithm systems B and C. In this case, multi-algorithm application systems lack a definite similarity criterion when displaying data.

[0115] The similarity scoring rules specify how to process multiple similarities. These rules include, but are not limited to, taking the average, the maximum, and the median. First, the multiple baseline similarities corresponding to the search results are processed according to the similarity scoring rules to obtain the fused similarity of the cached data.

[0116] For example, if the search results include the benchmark similarity between algorithm system A and algorithm system B, as shown below:

[0117]

[0118] According to the similarity value rules, after averaging the two baseline similarities, the fusion similarity of the cached data is 0.930599.

[0119] Multiple baseline similarities are processed according to similarity value rules to obtain a fused similarity, so that each retrieved data corresponds to a fused similarity. In this way, when displaying data, each search result can be displayed based on the fused similarity.

[0120] S203. Based on fusion similarity, determine multiple first target images from the reference images corresponding to multiple target algorithm systems.

[0121] Since the fusion similarity is determined based on a benchmark similarity of a unified standard, it would be more reasonable to select multiple first target images based on the fusion similarity.

[0122] As a feasible implementation method, S203 can be specifically implemented as follows: Select the reference images corresponding to the top K images with the highest fusion similarity, ranked from highest to lowest, as the first target images. In other words, use the K reference images with the highest fusion similarity in the search results as the first target images.

[0123] In some embodiments, after obtaining the fusion similarity of each search result, useless search results can be filtered out. For example, if a user requests a fusion similarity greater than or equal to 0.75 when arranging search tasks on the first interface, then all search results with a fusion similarity less than 0.75 in the search results corresponding to that search request will be deleted. Filtering search results based on fusion similarity can filter out some useless data, avoiding resource waste caused by excessive data volume.

[0124] In some embodiments, prior to S203, the method further includes: sorting the search results according to sorting rules and fusion similarity.

[0125] The sorting rules include, but are not limited to, fusion sorting and ranking sorting. Fusion sorting involves sorting all search results for the image to be retrieved in ascending or descending order based on fusion similarity. Ranking sorting requires recording the historical order in which the corresponding algorithm system retrieved the images in the cache, and then sorting them by fusion similarity while maintaining the historical order retrieved in the corresponding algorithm system.

[0126] by Figure 9 For example, in this implementation, the original search results are usually sorted according to the reference similarity corresponding to each algorithm system. The historical order in which the search results for algorithm system A and algorithm system B are retrieved is as follows: Figure 9 (I) and Figure 9 (II) Figure 9 (iii) refers to the result of sorting the search results based on ranking and fusion similarity.

[0127] Please see Figure 9 When sorting the search results based on ranking and fusion similarity, the first data from Algorithm System A (DID0000001) and the first data from Algorithm System B (DID0000002) are compared for fusion similarity. The data with the higher fusion similarity is ranked first, and the data with the lower fusion similarity is ranked second. Then, the second data from Algorithm System A (DID0000002) and the second data from Algorithm System B are compared for fusion similarity. Since the image IDs of the second data from Algorithm System A and the first data from Algorithm System B are the same, the second data from Algorithm System A (DID0000002) and the first data from Algorithm System B (DID0000002) are directly ranked first. Figure 9 The position of the second data in (iii) is determined, and the two data are merged, with the second data from algorithm system B becoming the third data after sorting. For example... Figure 9 As shown, the order of the search results obtained in this way is the same as the historical order in which the search results corresponding to algorithm system A and algorithm system B were retrieved.

[0128] As can be seen, after sorting the cached data using ranking sorting, no matter how the target data is extracted from the search results, the difference in the amount of data retrieved by each algorithm system in the target data is not significant. There will be no situation where a certain algorithm system retrieves a lot of data, which allows users to more intuitively compare the search results of each algorithm system.

[0129] S204. Display the second interface, which includes multiple first target images, as well as the identification information and fusion similarity of the target algorithm system corresponding to each first target image.

[0130] In this context, a first target image corresponds to one or more target algorithm systems.

[0131] Please see Figure 10 , Figure 10 The diagram shown is a schematic representation of a second interface provided in an embodiment of this application. After determining the first target image, multiple first target images, along with their corresponding algorithm system identification information and fusion similarity, are displayed, allowing users to intuitively see the retrieved results.

[0132] It should be noted that, Figure 10 This is merely an example illustrating the display of the algorithm system's identification information and fusion similarity on the first target image. However, in practical applications, the algorithm system's identification information and fusion similarity can be displayed next to or below the first target image, and this application does not impose any limitations on this.

[0133] It should be understood that the identification information of the target algorithm system can be a color identifier, a text identifier, etc., and this application embodiment does not impose any limitation on it. As a feasible implementation method, the identification information of the target algorithm system is a color identifier, and the background color of the second interface that integrates similarity is the same as the color of the identification information of the benchmark algorithm system.

[0134] By matching the background color of the fusion similarity score with the color of the identifier information of the benchmark algorithm system, the fusion similarity score of the first target image can be more clearly displayed, providing users with a more intuitive visual experience.

[0135] The image retrieval method provided in this application, when a user selects multiple target algorithm systems to retrieve an image on a first interface, sends the image to be retrieved to multiple target algorithm systems for retrieval. Then, it determines the fusion similarity of reference images in the retrieval results of multiple target algorithm systems, and selects the first target image from the retrieval results based on the fusion similarity, and displays it accordingly. This way, when a user wants to use multiple algorithm systems for image retrieval, they do not need to switch between algorithm systems; they can simply select the target algorithm system they want to retrieve from. This enables unified scheduling of multiple algorithm systems, improving the user experience.

[0136] In some embodiments, since multiple algorithm systems may be searching in the same image database, multiple target algorithm systems may retrieve the same reference image, that is, a first target image may correspond to multiple target algorithm systems.

[0137] As a feasible implementation method, if a first target image corresponds to multiple target algorithm systems, the second interface displays the identification information of the multiple target algorithms corresponding to each first target image.

[0138] The second interface displays the identification information of multiple target algorithms corresponding to each first target image, which makes it more intuitive to see which algorithm system the retrieved image comes from and which algorithm systems retrieved the same image at the same time.

[0139] In some embodiments, to compare the retrieval results of multiple algorithm systems, it is possible to... Figure 10 Use the options at the top of the page to select and switch the target data to be displayed. Clicking the "Fusion" option will display the first target image based on the fusion similarity of the first target image; clicking the "Algorithm System A" option will display the data retrieved by Algorithm System A in the search results separately; clicking the "Algorithm System B" option will display the data retrieved by Algorithm System B in the search results separately; clicking the "Algorithm System C" option will display the data retrieved by Algorithm System C in the search results separately.

[0140] As a feasible implementation method, the second interface also includes: display controls for each target algorithm system; the method also includes: after detecting the user's trigger operation on the display controls of the target algorithm system, displaying a third interface, which includes multiple second target images and reference similarities corresponding to the second target images.

[0141] The second target image is determined based on the reference image and reference similarity in the search results corresponding to the target algorithm system.

[0142] After the user clicks the display control of the target algorithm system, the second target image is determined based on the reference image and reference similarity in the search results corresponding to the target algorithm system. Then, the second target image and the reference similarity between the second target image are displayed on the third interface.

[0143] By selecting the display controls for different target algorithm systems, users can compare the search results from multiple algorithm systems from different perspectives, providing users with different evaluation criteria. This can improve the display effect of multi-algorithm image retrieval and bring a better user experience.

[0144] In some embodiments, since users may perform multiple searches for the same image to be searched, the search results of multiple target algorithm systems can be cached as cached data. This can avoid multiple searches for the same image to be searched, thus avoiding waste of resources.

[0145] For example, combining Figure 11 As shown, the target algorithm system includes algorithm system A, algorithm system B, and algorithm system C. After the reference similarity of the retrieval results of the three algorithm systems A, B, and C is converted into the benchmark similarity, the retrieval results of the three systems are appended to the cache data.

[0146] As a feasible implementation, upon receiving an instruction to search for an image, the system first determines whether the image has already been searched and whether corresponding cached data exists. If the amount of cached data is greater than the amount of data to be returned, the first target image is directly determined from the cached data; if the amount of cached data is less than the amount of data to be returned, the image to be searched is sent to multiple target algorithm systems, allowing these systems to search for the image.

[0147] For example, when a user performs an image search for the first time, there is no corresponding cached data for the image. Therefore, the target algorithm system needs to be determined for the search. If the search yields 52 results, but the user only requested the first 20, then the first 20 results are selected as the target data. When the user wants to view the cached data on the next page (assuming 20 results are displayed per page), it's equivalent to resubmitting the search. This time, there is corresponding cached data for the image, and the amount of cached data is greater than the amount of data to be returned. Therefore, the 21st to 40th results from the 52 cached results are selected as the target data. When the user wants to continue searching for cached data on the next page, since only 12 cached results remain, which is less than the amount of data to be returned, another search is required.

[0148] In some embodiments, since the number of first target images that the second interface can display is limited, the step of determining multiple first target images from reference images corresponding to multiple target algorithm systems based on fusion similarity can be specifically implemented as follows: determining multiple first target images from cached data according to display rules and fusion similarity.

[0149] The display rules specify how cached data is extracted. Since the content of the target data to be displayed in a multi-algorithm application system is limited, all cached data cannot be sent directly to the system. Instead, based on the display rules carried in the retrieval request, the target data is extracted from the cached data and sent to the multi-algorithm application system so that the second interface displays the first target image of the target data.

[0150] For example, as a feasible implementation method, after sorting the cached data in descending or ascending order based on fusion similarity, please refer to [link to relevant documentation]. Figure 12 The display rule can be a pagination parameter. Based on the pagination parameter (2,25), data from the cache number 26 to 50 can be extracted as the target data. For another feasible implementation method, please refer to [link to relevant documentation]. Figure 13 The display rule can be the total number of data points. Based on the total number of data points of 100, the first 100 data points in the cached data are selected as the target data.

[0151] In some embodiments, when a user clicks on a first target image on the second interface, the first target image and the image to be searched are displayed side-by-side, along with detailed information about the search results corresponding to the first target image, such as image content, image information, similarity before and after conversion, etc. This allows users to more intuitively compare the search results and the image to be searched, and to see detailed information about the search results corresponding to the first target image.

[0152] In some embodiments, due to differences between different algorithm systems, there is a great deal of controversy regarding the similarity values ​​in the search results. The reference similarity in the search results of some algorithm systems is more accurate in a certain range, but less accurate in other ranges, resulting in less accurate search results after fusion. In this case, it is necessary to set the filtered similarity range for the target algorithm system when the user arranges the search task on the first interface.

[0153] As a feasible implementation method, the first interface also includes: a similarity filtering interval corresponding to each target algorithm system. After receiving the search results sent by multiple target algorithm systems, the method further includes: deleting the search results corresponding to each target algorithm system whose reference similarity falls outside the corresponding similarity filtering interval.

[0154] In other words, after obtaining the search results, the search results returned by each target algorithm system are first filtered according to the similarity filtering interval corresponding to each target algorithm system, removing search results outside the similarity filtering interval, and then the fusion similarity of the reference images in the search results is determined. This makes the similarity in the obtained search results more in line with user needs, and more refined and accurate.

[0155] In some embodiments, users may issue numerous search requests, each with its own corresponding cached data. This large amount of cached data can consume significant amounts of memory. Therefore, it is crucial to focus on the timeliness of cached data and remove it if it has not been used for an extended period to prevent it from consuming memory.

[0156] As a feasible implementation method, the approach also includes: tracking the caching time of cached data; and deleting cached data if the caching time reaches a preset time.

[0157] It should be understood that the preset time is pre-set by the system. In actual application, the preset time can be set according to needs, and this application embodiment does not impose any limitations on it. For example, as a feasible implementation method, the preset time is 30 minutes. If the caching time of the cached data reaches 30 minutes, the cached data is deleted.

[0158] If, at any point before the cached data reaches the preset time, it is queried or modified again, the time when the cached data is processed again is used as the starting time to recalculate the cache time. If the cached data reaches the preset time, it means that the cached data has not changed or been used within the preset time, so the cached data is deleted to prevent it from consuming memory.

[0159] This application also provides an image retrieval device; please refer to [link / reference]. Figure 14The device 140 includes: a sending module 141, configured to send the image to be retrieved to the plurality of target algorithm systems, so that the plurality of target algorithm systems can retrieve the image to be retrieved using their respective retrieval algorithms; a determining module 142, configured to receive the retrieval results sent by the plurality of target algorithm systems and determine the fusion similarity of reference images in the retrieval results; wherein the reference images are images retrieved by the target algorithm systems, and the fusion similarity is determined based on a benchmark similarity; the benchmark similarity is a similarity based on a benchmark algorithm system; the determining module 142 is further configured to determine a plurality of first target images from the reference images corresponding to the plurality of target algorithm systems based on the fusion similarity; and a display module 143, configured to display a second interface, the second interface including the plurality of first target images, and the identification information of the target algorithm system corresponding to each first target image and the fusion similarity; wherein one first target image corresponds to one or more target algorithm systems.

[0160] As a feasible implementation method, if a first target image corresponds to multiple target algorithm systems, the second interface displays the identification information of the multiple target algorithm systems corresponding to each first target image;

[0161] The second interface also includes: display controls for each target algorithm system; the display module is also used to display a third interface, which includes multiple second target images and reference similarities corresponding to the second target images. The second target images are determined based on the reference images and reference similarities in the search results corresponding to the target algorithm system.

[0162] The module is specifically used to convert the reference similarity of reference images in the search results into benchmark similarity, where the reference similarity is the similarity between the reference image and the image to be searched calculated by the corresponding target algorithm system; merge search results with the same image identifier, the merged search results include multiple benchmark similarities, each benchmark similarity corresponds to a target algorithm system; and determine the fusion similarity based on the similarity value rules and multiple benchmark similarities.

[0163] The module is specifically used to: acquire the benchmark algorithm system; determine the similarity conversion rules between the target algorithm system and the benchmark algorithm system; the similarity conversion rules include the correspondence between similarity intervals and conversion algorithms; the conversion algorithm is used to convert the reference similarity of the reference image in the search results corresponding to the target algorithm system into the benchmark similarity corresponding to the benchmark algorithm system; determine the conversion algorithm corresponding to the search results based on the similarity interval of the reference similarity in the search results of the target algorithm system; and convert the reference similarity of the reference image in the search results into the benchmark similarity based on the conversion algorithm corresponding to the search results.

[0164] The target algorithm system is identified by color, and the background color of the similarity fusion in the second interface is the same as the color of the identification information of the benchmark algorithm system.

[0165] This application also provides an electronic device; please refer to [link / reference]. Figure 15 The electronic device 150 includes: one or more memories 151; one or more processors 152, wherein the one or more memories 151 are used to store computer program code, the computer program code including computer instructions; when the one or more processors 152 execute the computer instructions, the electronic device 150 performs the image retrieval method provided in the above embodiments.

[0166] Optionally, the memory 151 may be a non-transitory computer-readable storage medium, such as a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. The embodiments of this application do not impose any limitations on this.

[0167] The processor 152 may be a central processing unit (CPU), a general-purpose processor, a network processor (NP), a digital signal processor (DSP), a microprocessor, a microcontroller, a programmable logic device (PLD), or any combination thereof, and the embodiments of this application do not impose any limitations on this.

[0168] This application also provides a computer program product comprising one or more instructions, which are stored in the memory of a computer device and executed by a processor to complete the various processes described in the above embodiments.

[0169] This application also provides a computer-readable storage medium, which includes computer-executable instructions. When the computer-executable instructions are run on a computer, the computer performs the image retrieval method as provided in the above embodiments.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0171] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0172] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0173] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0174] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, etc.) or a processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0175] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An image retrieval method, characterized in that, include: In response to the user's operation of selecting multiple target algorithm systems to search for an image on the first interface, the image to be searched is sent to the multiple target algorithm systems so that the multiple target algorithm systems can use their respective search algorithms to search for the image to be searched; The system receives search results sent by the multiple target algorithm systems, the search results including reference images retrieved by the target algorithm systems and corresponding reference similarities; The reference similarity is the similarity between the reference image retrieved by the target algorithm system and the image to be retrieved; Based on the similarity interval of the reference similarity in the search results, the conversion algorithm corresponding to the search results is determined from the similarity conversion rules; the similarity conversion rules include the correspondence between similarity intervals and conversion algorithms; Based on the conversion algorithm corresponding to the search results, the reference similarity of the reference images in the search results is converted into a benchmark similarity; the benchmark similarity is a similarity based on a benchmark algorithm system. Based on the similarity value rules and the baseline similarity corresponding to the search results with the same image identifier, the fusion similarity of the reference images in the search results is determined; Based on the fusion similarity, multiple first target images are determined from the reference images corresponding to the multiple target algorithm systems; The second interface is displayed, which includes the plurality of first target images, as well as the identification information of the target algorithm system corresponding to each first target image and the fusion similarity; wherein, one first target image corresponds to one or more target algorithm systems.

2. The method according to claim 1, characterized in that, If one first target image corresponds to multiple target algorithm systems, the second interface displays the identification information of the multiple target algorithm systems corresponding to each first target image.

3. The method according to claim 1, characterized in that, The second interface also includes: display controls for each of the target algorithm systems; the method further includes: After detecting a user's trigger operation on the display control of the target algorithm system, a third interface is displayed. The third interface includes multiple second target images and reference similarities corresponding to the second target images. The second target images are determined based on reference images and reference similarities in the search results corresponding to the target algorithm system.

4. The method according to claim 1, characterized in that, The benchmark similarity based on similarity value rules and search results with the same image identifier includes: Search results with the same image identifier are merged. The merged search results include multiple benchmark similarities, and each benchmark similarity corresponds to a target algorithm system. The fusion similarity is determined based on the similarity value rules and the multiple benchmark similarities.

5. The method according to claim 4, characterized in that, The step of determining the conversion algorithm corresponding to the search result from the similarity conversion rules based on the similarity interval of the reference similarity in the search result includes: Obtain the benchmark algorithm system; A similarity conversion rule is determined between the target algorithm system and the benchmark algorithm system; the conversion algorithm is used to convert the reference similarity of the reference image in the retrieval result corresponding to the target algorithm system into the benchmark similarity corresponding to the benchmark algorithm system; Based on the similarity interval of the reference similarity in the retrieval results of the target algorithm system, the conversion algorithm corresponding to the retrieval results is determined.

6. The method according to claim 5, characterized in that, The identification information of the target algorithm system is a color identifier, and the background color of the second interface that integrates similarity is the same as the color of the identification information of the benchmark algorithm system.

7. An image retrieval device, characterized in that, The device includes: The sending module is used to send the image to be retrieved to the multiple target algorithm systems when the user selects multiple target algorithm systems to retrieve the image on the first interface, so that the multiple target algorithm systems can use their respective retrieval algorithms to retrieve the image to be retrieved; A determining module is configured to receive search results sent by the multiple target algorithm systems, wherein the search results include reference images retrieved by the target algorithm systems and corresponding reference similarities; the reference similarity is the similarity between the reference images retrieved by the target algorithm systems and the image to be searched; based on the similarity interval in the search results, a conversion algorithm corresponding to the search results is determined from similarity conversion rules; the similarity conversion rules include the correspondence between similarity intervals and conversion algorithms; based on the conversion algorithm corresponding to the search results, the reference similarity of the reference images in the search results is converted into a benchmark similarity; the benchmark similarity is the similarity based on a benchmark algorithm system; and based on similarity value rules and the benchmark similarity corresponding to search results with the same image identifier, the fusion similarity of the reference images in the search results is determined. The determining module is further configured to determine a plurality of first target images from the reference images corresponding to the plurality of target algorithm systems based on the fusion similarity. The display module is used to display a second interface, which includes the plurality of first target images, as well as the identification information of the target algorithm system corresponding to each first target image and the fusion similarity; wherein, one first target image corresponds to one or more target algorithm systems.

8. The apparatus according to claim 7, characterized in that, If one first target image corresponds to multiple target algorithm systems, the second interface displays the identification information of the multiple target algorithm systems corresponding to each first target image; The second interface also includes: a display control for each of the target algorithm systems; the display module is further configured to display a third interface, which includes multiple second target images and reference similarities corresponding to the second target images, wherein the second target images are determined based on reference images and reference similarities in the search results corresponding to the target algorithm systems; The determining module is specifically used to merge search results with the same image identifier, wherein the merged search results include multiple benchmark similarities, each benchmark similarity corresponding to a target algorithm system; and to determine the fusion similarity according to the similarity value rules and the multiple benchmark similarities. The determining module is specifically used to: acquire a benchmark algorithm system; determine a similarity conversion rule between the target algorithm system and the benchmark algorithm system; the similarity conversion rule includes the correspondence between similarity intervals and conversion algorithms; the conversion algorithm is used to convert the reference similarity of the reference image in the search results corresponding to the target algorithm system into the benchmark similarity corresponding to the benchmark algorithm system; and determine the conversion algorithm corresponding to the search results based on the similarity interval in the search results of the target algorithm system. The identification information of the target algorithm system is a color identifier, and the background color of the second interface that integrates similarity is the same as the color of the identification information of the benchmark algorithm system.

9. An electronic device, characterized in that, include: One or more processors; one or more memories; The one or more memories are used to store computer program code, which includes computer instructions. When the one or more processors execute the computer instructions, the electronic device performs the image retrieval method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed on a computer, cause the computer to perform the image retrieval method according to any one of claims 1 to 6.

Images