Information search method and device, electronic equipment and computer readable storage medium
By acquiring the reference area image from the target image and searching based on user information, the problem of low accuracy of information search in the prior art is solved, and efficient and accurate information search is achieved to meet the immediate needs of users.
Patent Information
- Application Number
- CN202310603164.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-25
- Publication Date
- 2024-11-26
AI Technical Summary
The existing information search methods are low in accuracy and cannot effectively meet the users' needs to easily find relevant information when they see a specific object.
By acquiring the reference area image from the target image, uploading it to the server, and determining candidate search results with similarity satisfies preset conditions based on the service category to which the user information and the reference area image belongs, improving the accuracy and efficiency of information search.
It improves the accuracy and efficiency of information search, ensures that users can quickly find information that meets their needs, meet users' immediate shopping and service needs, and improve service quality and shopping experience.
Smart Images

Figure CN119025696A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to information processing, and in particular to an information search method, device, electronic device and computer-readable storage medium. Background Art
[0002] Demands are everywhere in people's daily lives, but the demands that are stimulated instantly cannot always be met conveniently and effectively. Once users see a favorite hairstyle, a popular tourist spot, or tempting food, they need to open various corresponding apps to search, but they may not find what they want in the end; they see a celebrity's hairstyle in a magazine or a show that suits them very well, but they don't know where to get it done.
[0003] Traditional solutions can only search for relevant information through input information, but the accuracy of such information search is low. Summary of the invention
[0004] The embodiments of the present application provide an information search method, device, electronic device, and computer-readable storage medium, which can improve the accuracy of information search.
[0005] In a first aspect, the present application provides an information search method. The method comprises:
[0006] In response to an information search event for a target image, acquiring a reference area image from the target image;
[0007] Uploading the reference area image to a server;
[0008] Receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0009] The present application also provides an information search device, applied to a terminal, the device comprising:
[0010] An image acquisition module, configured to acquire a reference area image from the target image in response to an information search event for the target image;
[0011] An image uploading module, used for uploading the reference area image to a server;
[0012] A receiving module is used to receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0013] The present application also provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
[0014] In response to an information search event for a target image, acquiring a reference area image from the target image;
[0015] Uploading the reference area image to a server;
[0016] Receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0017] The present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0018] In response to an information search event for a target image, acquiring a reference area image from the target image;
[0019] Uploading the reference area image to a server;
[0020] Receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0021] The present application also provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0022] In response to an information search event for a target image, acquiring a reference area image from the target image;
[0023] Uploading the reference area image to a server;
[0024] Receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0025] In the above information search method, in response to an information search event for a target image, the terminal obtains a reference area image from the target image, the reference area image being a local image in the target image, which is helpful for targeted search; the reference area image is uploaded to a server, so that the server determines a candidate similar service whose similarity with the reference area image meets a preset similarity condition, and determines a candidate search result from similar candidate services of the service category to which the reference area image belongs according to user information of the terminal, the candidate search result better meets user needs, and improves the accuracy of information search. In addition, by obtaining the reference area image through the terminal, and determining the candidate search result from similar candidate services of the service category to which the reference area image belongs according to user information through the server, the search scope is narrowed and the efficiency of information search is improved.
[0026] In a second aspect, the present application provides an information search method. The method comprises:
[0027] Acquire a reference area image; the reference area image is acquired from the target image;
[0028] Determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category;
[0029] Calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity;
[0030] Acquire user information, and filter candidate search results from the similar candidate services according to the user information;
[0031] The candidate search results are sent to the terminal.
[0032] The present application also provides an information search device, applied to a server, comprising:
[0033] An image acquisition module, used to acquire a reference area image; the reference area image is acquired from a target image;
[0034] A service determination module, used to determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category;
[0035] A service query module, used for calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity;
[0036] A result generation module, used to obtain user information, and filter candidate search results from the similar candidate services according to the user information;
[0037] The search feedback module is used to send the candidate search results to the terminal.
[0038] The present application also provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
[0039] Acquire a reference area image; the reference area image is acquired from the target image;
[0040] Determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category;
[0041] Calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity;
[0042] Acquire user information, and filter candidate search results from the similar candidate services according to the user information;
[0043] The candidate search results are sent to the terminal.
[0044] The present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0045] Acquire a reference area image; the reference area image is acquired from the target image;
[0046] Determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category;
[0047] Calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity;
[0048] Acquire user information, and filter candidate search results from the similar candidate services according to the user information;
[0049] The candidate search results are sent to the terminal.
[0050] The present application also provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0051] Acquire a reference area image; the reference area image is acquired from the target image;
[0052] Determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category;
[0053] Calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity;
[0054] Acquire user information, and filter candidate search results from the similar candidate services according to the user information;
[0055] The candidate search results are sent to the terminal.
[0056] In the above information search method, the server obtains a reference area image; determines the service category to which the reference area image belongs, and obtains each candidate service corresponding to the service category; then calculates the similarity between the reference area image and each candidate service, and determines similar candidate services among the candidate services based on the similarity; then obtains user information, and obtains candidate search results from similar candidate services based on the user information, and finally sends the candidate search results to the terminal; thereby, from each candidate service corresponding to the service category, to similar candidate services among the candidate services, and then to the candidate search results, the service scope is gradually narrowed, and candidate search results that better meet user needs are returned to the terminal, thereby improving the accuracy of information search, and the server determines the candidate search results from similar candidate services of the service category to which the reference area image belongs according to the user information, thereby narrowing the search scope and improving the efficiency of information search. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0058] Figure 1 An application environment diagram of an information search method in an embodiment;
[0059] Figure 2 is a flow chart of a method for searching information of a terminal in one embodiment;
[0060] Figure 3is a flow chart of a process for obtaining a reference area image in one embodiment;
[0061] Figure 4 A schematic diagram of interface changes for a selected operation in an embodiment;
[0062] Figure 5 A flowchart of an information search method of a server in one embodiment;
[0063] Figure 6 is a schematic diagram of the structure of a target detection model in one embodiment;
[0064] Figure 7 A schematic diagram showing changes in a reference avatar image in one embodiment;
[0065] Figure 8 A schematic diagram of changes in a rating information interface in an embodiment;
[0066] Fig. 9 is a flow chart of an image retrieval technique in one embodiment;
[0067] Fig.10 is an interactive interface flow chart of an information search method in one embodiment;
[0068] Fig.11 A structural block diagram of an information search method in an embodiment;
[0069] Fig.12 is a flow chart of an information search method in one embodiment;
[0070] Fig.13 is a timing diagram of an information search method in one embodiment;
[0071] Fig.14 A structural block diagram of an information search method in an embodiment;
[0072] Fig.15 A schematic diagram of interface changes of an information search method in one embodiment;
[0073] Fig.16 A schematic diagram of interface changes of an information search method in another embodiment;
[0074] Fig.17 A flowchart of feature extraction and similarity calculation in one embodiment;
[0075] Fig.18 is a schematic diagram of the structure of an object category in one embodiment;
[0076] Fig.19 A schematic diagram of search results for subject detection in one embodiment;
[0077] Fig. 20 is a schematic diagram of information search results in one embodiment;
[0078] Fig.21 A schematic diagram of a structure of a specific service category in an embodiment;
[0079] Fig. 22 is a structural block diagram of an information search device in an embodiment;
[0080] Fig.23 is a structural block diagram of an information search device in another embodiment;
[0081] Fig.24 is a schematic diagram of a server structure in an embodiment;
[0082] Fig.25 It is a schematic diagram of the terminal structure in one embodiment. DETAILED DESCRIPTION
[0083] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0084] The information search method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the terminal 102 communicates with the server 104 through a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers.
[0085] The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster or cloud consisting of multiple servers.
[0086] In one embodiment, Figure 2 As shown, an information search method is provided, which is applied to Figure 1 The terminal 102 in the example is used as an example to illustrate, and the following steps are included:
[0087] Step 202: In response to an information search event for a target image, a reference area image is acquired from the target image.
[0088] The target image includes at least part of the selectable area, and the selectable area can be used as a reference area image under the influence of different information search events. Optionally, the target image includes at least one identifiable object of a service category, and under the influence of a specific information search event, the area where the identifiable object is located can serve as a reference area image. The identifiable object is an object that can be identified by target detection, including but not limited to objects, avatar hairstyles, attractions, food or other identifiable objects of other service categories.
[0089] The information search event is an event for information search processing on the target image. This event is used to control the terminal to select the selectable area in the target image as the reference area image. The reference area image is the area image to be searched for information obtained from the target image. The reference area image is used as the input data for information search, and the information search is performed in the image.
[0090] Step 204: Upload the reference area image to the server.
[0091] The server is used to identify the information of the reference area image and search according to the information of the reference area image to obtain candidate search results. Optionally, the server stores a service resource, and can query the corresponding candidate search results in the service resource according to the information of the reference area image, and return the candidate search results to the terminal.
[0092] In an optional embodiment, uploading the reference area image to the server includes: the terminal uploads the reference area image to the cloud to instruct the cloud to send the candidate service information adapted by the reference area image to the terminal as a candidate search result; wherein the candidate service information includes, but is not limited to, information about physical products, virtual product information, product resource point information for providing physical products, or product resource point information for providing service products. Product resource point information for providing physical products includes information about online stores or physical stores that provide physical products by mailing, distributing physical products, or delivering them offline; product resource point information for providing service products includes information about online stores or physical stores that provide such service products for remote or offline services, virtual props services such as phone bills or game coins, scenic spot planning services, or processing services for physical products.
[0093] Exemplarily, product resource point information for providing physical products includes restaurant A11 information generated by a resource point push application for restaurant A1 for food A; product resource point information for providing service products includes barber shop B11 information generated by a resource point push application for barber shop B1 for hairstyle B.
[0094] Step 206, receiving the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services of the service category to which the reference area image belongs according to the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0095] The candidate search results can be determined as the target search results based on the selection instructions sent by the terminal, and the corresponding services are provided to the terminal through the target search results. The candidate search results are resource information determined by the server from similar candidate services in the service category to which the reference area image belongs based on the user information of the terminal. If the server screens the similar candidate services obtained by searching based on the reference area image according to the user information of the terminal, and obtains similar candidate services applicable to the user information; the similar candidate services applicable to the user information are used as candidate search results, the candidate search results are more in line with user needs, the accuracy of information search is improved, and the search scope is narrowed, and the efficiency of information search is improved accordingly.
[0096] The service category is a set of categories divided according to a pre-set service category. The service category is obtained by classifying various types of services that can be provided to users. In the process of searching according to the reference area image, similar candidate services in the service category to which the reference area image belongs are identified, and the service range with appropriate data volume is determined by the similar candidate services. Similar candidate services are candidate services in the service category to which the reference area image belongs, and the similarity with the reference area image meets the preset similarity condition. Candidate services include product resource points provided for physical products and services under the service category to which the reference area image belongs, and the candidate services correspond to application programs such as resource point push applications and path planning applications, so as to execute the corresponding services through the application programs corresponding to the candidate services. Optionally, the preset similarity condition is a similarity condition for a feature dimension, and the feature dimension can be image similarity, semantic similarity, a similarity condition for a feature dimension obtained by combining the two, or a similarity condition for other feature dimensions. For example: if the similarity between a candidate service and the reference area image is greater than a preset similarity threshold, it is determined that the candidate service meets the preset similarity condition, that is, the candidate service is determined to be a similar candidate service.
[0097] Exemplarily, if it is determined that the reference area image is an unknown food X, and the user information includes the teenager's food preference, the service category is the food category, the candidate services include food A, food B, food C and food D, similar candidate services are food A, food B and food D whose similarity to food X meets corresponding conditions, the candidate search results are food A and food B determined by the server according to the teenager's food preference, and the target search result is food A selected according to the instruction issued by the terminal; the resource point push application corresponding to food A can be a restaurant push application where food A is purchased or processed, and the path planning application corresponding to food A is used to plan the path between the terminal location and the restaurant that provides food A.
[0098] In the above information search method, the terminal responds to the information search event of the target image, obtains the reference area image from the target image, and the reference area image is a local image in the target image, which is helpful for targeted search; uploads the reference area image to the server, so that the server can determine the candidate search results from the similar candidate services of the service category to which the reference area image belongs according to the user information of the terminal, and the candidate search results are more in line with the user's needs, and the accuracy of the information search is improved. Moreover, due to the small search range, the efficiency of the information search is high; the terminal receives the candidate search results returned by the server. Thus, the terminal captures the image, and the server determines the candidate search results from the similar candidate services of the service category to which the reference area image belongs according to the user information, so as to accurately and efficiently conduct information search.
[0099] In one embodiment, the overall process of obtaining input data is described; in response to an information search event of a target image, a reference area image is obtained from the target image, including: obtaining a target image obtained by a preset operation method; in response to an information search event generated by a trigger operation on the target image, a reference area image is obtained from the target image.
[0100] The trigger operation on the target image can be an operation before, during or after the target image acquisition process, and the trigger operation is used to trigger the information search event. Optionally, the preset operation mode and the trigger operation can be operation instructions triggered by different buttons, threads or modules of the terminal or other methods, so that the two are performed synchronously. Optionally, the trigger operation can be triggered by a specific key or a specific touch action. The trigger operation includes but is not limited to triggering by clicking or double-clicking a certain position of the target image, and triggering by pressing the volume "+" key while taking a photo to acquire the target image.
[0101] In an optional implementation, acquiring the target image obtained by a preset operation mode includes: receiving the target image transmitted by other electronic devices, or acquiring the target image from an album of the terminal, or acquiring the target image recorded in real time by the terminal.
[0102] The process of acquiring the target image and the reference area image is executed through the same terminal. After acquiring the target image, the reference area image can be quickly acquired from the target image. The data processing link is short and the data processing efficiency is high. Moreover, information search events can be directly generated for the target object. With the mobile phone platform in a neutral position, it is easy to access a variety of third-party services, meet the more personalized needs of users, and improve the efficiency of information search.
[0103] Optionally, the preset operation mode includes a shooting or screenshot operation. Thus, by shooting or screenshotting with the same terminal, the target image can be obtained through the camera or through the content on the terminal screen, which makes the data processing link relatively shorter and the information search more efficient.
[0104] In one embodiment, Figure 3 As shown, the process of obtaining the reference area image is described. In response to the information search event of the target image, the reference area image is obtained from the target image, including:
[0105] Step 302: In response to the information search event of the target image, upload the target image to the server.
[0106] Optionally, the server includes a server for detecting object location information, and a server for filtering candidate search results from similar candidate services based on user information. These two types of servers are professionally maintained by different personnel to ensure service quality. Exemplarily, the server for detecting object location information is an intelligent cloud that provides a target detection model; the server for filtering candidate search results from similar candidate services based on user information is a traditional cloud service platform, which is used for data cloud storage.
[0107] In another optional embodiment, the same server is used to detect the object location information and to filter the candidate search results from similar candidate services according to the user information.
[0108] Step 304: receiving the object position information returned by the server, wherein the object position information is a position mark obtained by the server performing target detection on the target image.
[0109] The object location information is a location mark obtained by performing target detection on the target image, which can be a location box of the candidate area, or a center coordinate point of the candidate area, or a set of region coordinates of the detected candidate area in the target image. Optionally, the object location information returned by the server includes a location mark for representing each candidate area, and can also include a location mark for representing the entire target image.
[0110] The candidate area is an image area obtained by performing feature recognition according to a preset service category based on a target detection model. Each candidate area is a selectable area and can be used as a reference area image. It is obtained by detecting objects, scenic spots, hairstyles or foods in the process of detecting the target detection model.
[0111] Step 306: Display the candidate area corresponding to the object position information in the target image.
[0112] In an optional implementation, the candidate area corresponding to the object position information is displayed in the target image, including: if the object position information returned by the server is the position frame of the candidate area, the center coordinate point of the candidate area is determined according to the position frame of the candidate area, and the center coordinate point of each candidate area is displayed in the target image; if the object position information returned by the server is the center coordinate point of the candidate area, the center coordinate point of each candidate area is displayed in the target image; if the object position information returned by the server is the region coordinate set of the candidate area, the center coordinate point of each candidate area is determined according to the region coordinate set, and the center coordinate point of each candidate area is displayed in the target image. In this way, each area in the target image is displayed more concisely, so as to more efficiently obtain the reference area image.
[0113] Step 308: Acquire a reference region image from the target image marked with the candidate region.
[0114] Optionally, the object location information is the server using its own or a third-party target detection model to perform target detection on the image uploaded by the terminal, and feedback the detected target coordinate points and other location marks to the terminal, and then display the location marks of the objects such as objects indicated by the target coordinate points on the mobile phone target picture. Exemplarily, when the object location information is the location mark of the object, a white dot is marked at the center of the corresponding object to indicate that the area where the object is located can be used as a reference area image as a selectable area, and a white dot can also be marked in the center of the target image to indicate the complete target image.
[0115] In an optional implementation, obtaining a reference area image from a target image marked with a candidate area includes: displaying the target image marked with the candidate area through a screen of a terminal; and segmenting a candidate area that triggers a selection instruction into a reference area image.
[0116] In one feasible implementation, during the acquisition of the reference area image, after receiving the function request from the terminal side, the terminal cloud side will perform target detection on the target image to determine the selection box of the candidate area. Optionally, a general target detection model can be used to determine the selection box of the candidate area, and only the position of the selection box is detected without recognition, so as to improve the detection speed.
[0117] The target detection reference area image is performed through the server to obtain the object position information, so that the data processing efficiency is improved, the candidate area corresponding to the object position information is displayed in the target image, and the reference area image is obtained from the target image marked with the candidate area, so that each candidate area can be clearly displayed without interfering with each other through the terminal, so as to efficiently screen out the reference area image from the target image.
[0118] In a feasible implementation, obtaining a reference region image from a target image marked with candidate regions includes: determining the center point of each candidate region and the center point of the entire image of the target image; calculating the distance between the center point of each region and the center point of the entire image; determining the reference region with the shortest distance to the center of the entire image from each candidate region based on the distance between the center point of each region and the center point of the entire image; and segmenting the reference region image from the reference region of the target image. This achieves automatic determination of the reference region image.
[0119] In a feasible implementation, in order to make the reference area image more in line with user needs and improve the accuracy of candidate search results; the reference area image is obtained from a target image marked with a candidate area, including: in response to a selection operation on a candidate area in the target image, the candidate area affected by the selection operation is determined as a reference area; the reference area is extracted from the target image to obtain a reference area image.
[0120] The selection operation of the candidate area is to treat the candidate area as a combined area. This process can select the object location information in the target image for more efficient detection, or add a selection box to the target image to autonomously divide the reference area in the target area. Since the candidate area is directly selected, it can better meet the needs of users and further improve the accuracy of the search.
[0121] In an optional embodiment, a region identifier is displayed in a target image, and the region identifier includes a candidate region identifier and a target image identifier; in response to a selection operation on a candidate region in the target image, the candidate region affected by the selection operation is determined as a reference region, including: in response to a selection operation on the region identifier, the region identifier affected by the selection operation is determined as a reference region identifier, and the region represented by the reference region identifier is determined as a reference region.
[0122] The region identifier is used to determine the selectable range in the target image. The global image or local image represented by the region identifier can be converted into a reference region under the action of the selection operation. The candidate region identifier is the position mark of each candidate region, and the target image identifier is the position mark of the entire target image.
[0123] When the selection operation acts on a candidate region identifier, the candidate region identifier acted on by the selection operation is the reference region identifier, and the region represented by the candidate region identifier is determined as the reference region. Figure 4 (a) shown.
[0124] When the selection operation acts on the target image identifier, the target image identifier acted on by the selection operation is the reference area identifier, and the entire image of the target image is determined as the reference area. Figure 4 (b) as shown.
[0125] Therefore, the range of the target image that can be used as a reference area image is clarified through area identification, and a candidate area in the target image can be selected as a reference area through candidate area identification. The entire target image can also be determined as a reference area through target image identification, so as to determine the reference area image more efficiently.
[0126] In an optional embodiment, if Figure 4 As shown in (c), in response to a selection operation on a candidate area in a target image, the candidate area affected by the selection operation is determined as a reference area, including: in response to a selection box added in the target image, the area contained in the selection box is determined as the reference area. Thus, when the object that the user wants to select is not in the candidate area obtained by target detection, the user can add a selection box to select the area of interest, and can also manually adjust the width and height of the box to customize the reference area when the target detection is incomplete.
[0127] In an optional embodiment, if the target image identifier acted upon by the selection operation, the entire image is initially presented as a frame selection by default, and corresponds to the middle white dot; if the selection operation acts upon the candidate area identifier obtained by target detection, the central white dot of the candidate area acted upon by the selection operation is selected, and the brightness of the candidate area is increased to a preset value, and when the position of the object area is correct, no adjustment is required, and when it is incorrect, manual adjustment can be made.
[0128] In a possible implementation, extracting a reference area from a target image to obtain a reference area image includes: determining a container of the reference area; filling the container with pixel values of the reference area of the target image according to the coordinate correspondence between the reference area and the container to obtain a container filled with the reference area; and using the container filled with the reference area as the reference area image. The reference area container is a running environment that performs pixel values on the reference area of the target image and configures computing resources in units of processes, and the running environment complements and interferes with other running environments to stably generate the reference area image.
[0129] In a possible implementation, extracting the reference area from the target image to obtain the reference area image includes: cropping the target image according to the reference area to obtain the reference area image. Thus, the target object is cropped and the reference area image is generated efficiently through the preset operating environment of the terminal.
[0130] In one embodiment, uploading the target image to the server includes: scaling the target image to obtain a target image of a target size, and uploading the target image of the target size to the server, wherein the target size is obtained based on a preset side length and a screen size ratio.
[0131] The target size is designed based on the preset side length and screen size ratio. The target image of the target size is less deformed than the unmodified target image and can be applied to the respective screen ratios of different models. The target image of the target size is uploaded to the server in a shorter time, which can improve the efficiency of data transmission and thus improve the efficiency of information search. The preset side length is the benchmark data for calculating the screen size ratio, which is used to determine the size of the target image. For example, the target size is 512*256mm, the preset long side is 512mm, and it is compressed in jpg format according to the target size and uploaded to the cloud.
[0132] The target size may be calculated before the target image is scaled, or may be data stored in the terminal, and the preset side length is data stored in the terminal.
[0133] In one embodiment, the method further includes: in response to a selection instruction for a candidate search result, determining a target search result according to the selection instruction, and calling an application corresponding to the target search result to execute a corresponding service; in response to a service information synchronization instruction generated during the execution of the corresponding service by the application, transmitting the service information to a server to instruct the server to synchronize the service information to an associated device associated with the terminal.
[0134] The target search result is a service provided by at least one application determined from the candidate search results according to the terminal's selection instruction. Since most applications can be jumped and entered from the terminal, the corresponding service can be executed through the application corresponding to the target search result; that is, the terminal directly executes the service corresponding to the target search result, and during the execution of the service, the service information is synchronized to the terminal's associated device through the server, so as to efficiently realize multi-terminal data synchronization.
[0135] The associated device associated with the terminal is a device that has data association with the terminal, which can be a device corresponding to the same user information as the terminal, so that the user can obtain the synchronization information of related orders and services on any terminal; the associated device can also be used to provide a server corresponding to the target search result, so as to synchronize the server data of different applications. Exemplarily, if the terminal is a mobile phone that is browsing the food A takeaway order or order page, the associated devices associated with the terminal include but are not limited to computers, wearable devices, car computers, and smart TVs that belong to the same user account as the mobile phone; after information synchronization, the user can continue to browse the food A takeaway order or order page through any associated device.
[0136] In a feasible implementation, the target search result is to determine the resource point A3 that provides food A near the terminal location through a certain resource point push application. After the target search result is determined according to the selection instruction, the resource point push application corresponding to the target search result is called to perform the services corresponding to the target search result such as resource transfer or other orders of food A for resource point A3; and during the execution of these services, the information of resource transfer or other orders will be synchronized to the terminal's associated device. Among them, the user can reach a mailing agreement for food A with the merchant of resource point A3 through the resource point push application of the terminal, so that food A can be delivered to the location of the terminal.
[0137] Therefore, cross-end services are provided for the personality matching the user information, and the data of the cross-end services is obtained by combining the data of multiple terminals and the data of each user scenario. With the help of the cloud basic platform, a multi-dimensional fusion analysis model can be built to realize user personalized services or product recommendations, which can provide users with more personalized services. For example, recommendations can be made based on the user's daily usage habits, user portraits, and user scenarios, and customized recommendations that fit the user can be provided; at the same time, all terminals with photo and screenshot functions can use this function, and related order, logistics, browsing and other information can be shared across terminals to form a seamless intelligent cross-end service; for example, personalized push based on geographic location and time, for example, when a food picture is detected, the user can push takeout at home and push nearby shops outdoors; nearby supermarket information can be added to the shopping candidate list during the day, etc.
[0138] In one embodiment, the method also includes: responding to a selection instruction for a candidate search result, determining a target search result according to the selection instruction, determining a resource point push application and a path planning application between the terminal and the target search result according to the resources corresponding to the target search result; calling at least one of the resource point push application and the path planning application to execute corresponding services.
[0139] The target search result is determined from the candidate search results and still represents certain resource information; this resource information is used to indicate the resource owner of a certain resource, which can be the geographical location of a resource point that provides a certain resource, or the resource point identifier or name that provides a certain resource, or the resource point link of a certain resource.
[0140] The resource point push application is used to determine the resource owner corresponding to the target search result; the path planning application is used to act as a bridge between the terminal and the target search result so that the user can reach the resource owner of the target search result. Optionally, some applications have the functions of both the resource point push application and the path planning application. For example: in the case where the target search result belongs to a physical store, the resource point push application is a physical store push application, and the path planning application can be a map application; in the case where the target search result is an online store, the resource point push application is an online store push application, and the path planning application is an application for web page jumps.
[0141] Taking the target result belonging to a physical store as an example, when the resource point push application is called, the specific information of the resource owner can be obtained, and the service details can be determined based on the specific information, and services corresponding to the target search results such as coupon purchase and evaluation information viewing can be executed; when the route planning application is called, the location information of the resource owner can be obtained, and the route can be planned based on the location information. For example: when a user takes a headshot photo and clicks on the hair, he can access a resource point push application to search for nearby related hair salons, and when taking photos of scenic spots, he can access the tourist attraction page of the resource point push application; for example, if the user usually likes to use resource point push application A, then if it is a commodity, he can jump to the shopping page of the resource point push application first. Similarly, if the photo image is of food or hairdressing, you can jump to the nearby resource point push application A to view the evaluation of nearby resource points; if it is a tourist attraction picture, you can jump to the attraction introduction and ticket purchase page related to each resource point push application.
[0142] The system's photo taking and screenshot functions can be used as an entry point to more conveniently access other systems, thus providing richer services.
[0143] In addition, based on the massive cloud data provided by the server as the basis for user information, user images and user intention data can be modeled in the cloud to build a user intention prediction model for complex scenarios. This can provide users with more accurate candidate search results and make more accurate service recommendations.
[0144] In summary, the mobile phone image acquisition system is combined with the information search process to meet the user's instant shopping and service needs, improve service quality and shopping experience; use multi-terminal data to build a multi-dimensional fusion analysis model to realize user personalized services or product recommendations, and resource point push applications include applications for recommending shopping information, local shops, attractions, geographic locations or other applications. Moreover, the mobile platform can access multiple third-party services and shopping apps to meet users' more personalized needs and bring a more extreme user experience. At the same time, providing comprehensive services can meet more types of more personalized needs; innovatively open up a fast direct connection channel between instantaneous needs and services in reality, and at the same time, combined with a cross-platform and cross-terminal system program, transmit service information to the server to instruct the server to synchronize service information to the associated device associated with the terminal to understand the service status at any time; and form a user intention recognition system through massive data to meet users' more accurate needs based on user information.
[0145] In one embodiment, Figure 5 As shown, the present application also provides an information search method, applied to a server, the method comprising:
[0146] Step 502, obtaining a reference area image; the reference area image is obtained from the target image.
[0147] Optionally, the reference area image may be directly uploaded by the terminal, forwarded from other servers, or obtained from a server based on certain instructions. The reference area image is input data for the information search process.
[0148] Step 504: determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category.
[0149] In an optional embodiment, determining the service category to which the reference area image belongs includes: judging the service category to which the reference area image belongs according to feature information of the reference area image in preset service categories.
[0150] In an optional embodiment, obtaining candidate services corresponding to a service category includes: determining a service set corresponding to the service category, and determining services in the service set as candidate services; wherein the service set may be a service list or a service array.
[0151] Step 506: Calculate the similarity between the reference area image and each candidate service, and determine similar candidate services among the candidate services based on the similarity.
[0152] In an optional embodiment, calculating the similarity between the reference area image and each candidate service includes: extracting at least one of image features and semantic features from the reference area image, performing similarity calculation based on the extracted features and features of each candidate service, to obtain the similarity between the reference area image and each candidate service.
[0153] Correspondingly, determining similar candidate services among the candidate services based on the similarity includes: if the similarity between the reference area image and each candidate service meets a preset similarity condition, determining the candidate service as a similar candidate service.
[0154] In an optional implementation, the image search service of a third-party resource point push application can be called; or similar candidate services can be determined according to semantic similarity and image similarity. The process of searching for goods is completed by calling the image search service of the resource point push application, which saves manpower and reduces costs. At the same time, the third-party resource point push application's own algorithm is more timely in updating goods. Determining similar candidate services according to semantic similarity and image similarity in this case can better control the overall data flow.
[0155] Step 508: Obtain user information, and filter similar candidate services according to the user information to obtain candidate search results.
[0156] In one feasible implementation, the candidate search results are obtained by filtering similar candidate services according to the user information, including: matching similar candidate services according to user preferences corresponding to the user information; and filtering similar candidate services matching the user preferences as candidate search results. Optionally, the candidate search results are displayed in a list, chart, QR code, or hyperlink.
[0157] Step 510: Send the candidate search results to the terminal.
[0158] In an optional implementation manner, sending the candidate search results to the terminal includes: forwarding the candidate search results to the terminal via other servers, or directly transmitting the candidate search results to the terminal.
[0159] In the above information search method, the server obtains a reference area image; determines the service category to which the reference area image belongs, and obtains each candidate service corresponding to the service category, and each candidate service corresponding to the service category is a candidate service with the largest range in this embodiment; then calculates the similarity between the reference area image and each candidate service, and determines similar candidate services among the candidate services based on the similarity, and the similar candidate services are candidate services with a middle range in this embodiment; then obtains user information, and filters candidate search results from similar candidate services according to the user information, and the candidate search results are candidate services with a smaller range, and finally sends the candidate search results to the terminal to return candidate search results that better meet user needs to the terminal, thereby improving the accuracy of information search, and the server determines the candidate search results from similar candidate services of the service category to which the reference area image belongs according to the user information, thereby narrowing the search range and improving the efficiency of information search.
[0160] In one embodiment, the method also includes: acquiring a target image uploaded by a terminal; performing target detection on the target image to obtain object location information; returning the object location information to the terminal, wherein the object location information is used to indicate that the target image displays a candidate area corresponding to the object location information, and the reference area image is acquired from the target image marked with the candidate area.
[0161] In a feasible implementation, performing target detection on a target image to obtain object location information includes: determining a target detection model adapted to the target image; performing target detection on the target image according to the identifiable objects of the target detection model to obtain location information of the identifiable objects. The identifiable objects may be objects whose image features or semantic features can be identified by other target detection models, so that the location information and service category of the identifiable objects can be detected respectively by different target detection models.
[0162] Since the server has more computing resources, it can perform efficient target detection on the target image and obtain the object location information more quickly, thereby efficiently indicating the candidate area corresponding to the object location information displayed in the target image, making it easier for users to obtain the reference area image.
[0163] In one embodiment, the object position information is obtained by detecting the target image through a pre-trained target detection model, and the training method of the target detection model includes: determining different training image data sets, and the respective detection heads of each training image data set in the target detection model to be trained; predicting each training image data set through different detection heads to obtain the prediction value of each training image data set; aligning the label values of different training image data sets according to the prediction value of each training image data set to obtain the mapping label set of each training image data set; training the target detection model to be trained according to each training image data set and the corresponding mapping label set to obtain a trained target detection model.
[0164] The training image dataset is a dataset from a data source, which can be but not limited to the Open Images dataset, MS-COCO dataset, Objects365 dataset, and traditional datasets such as Pascal VOC and COCO. Among them, the OpenImages dataset is a large-scale dataset, containing 6,000 categories of labels, with an average of 8 labels per image, 1,910K images, and 15,440K annotated objects; ImageNet is a large-scale image library based on the WordNet backbone structure, with a total of 14,197,122 images, divided into 21,841 categories of labels; compared with traditional datasets, the MS-COCO dataset is currently the most challenging target detection dataset, which contains 80 categories, 164K images, and 897K annotated objects; the Objects365 dataset contains a total of 630,000 images, covering 365 categories, up to 10 million frames, and has the characteristics of large scale, high quality, and strong generalization ability.
[0165] Among them, the current open source data sets basically include precise frame annotations of various items in more than 14 categories, such as people, clothing, rooms, bathrooms, kitchens, offices, electrical appliances, transportation, food, fruits, vegetables, animals, sports, and musical instruments. They can basically meet the basic requirements of training data for various object frame annotations required in this case. The construction of the general object detection model in this case can be built based on these open source training data sets, and the data set can also be built in a customized way.
[0166] In an optional embodiment, a fully automatic method of visual data is used to unify the labels of different image training sets to obtain a more robust output result; wherein each data set has its own label set, and the label sets of each data set are predicted and aligned through the detection heads of each data set respectively, so as to jointly learn a label mapping set for all data sets, as well as a one-to-one mapping relationship between the labels of each data set and the labels in the label mapping set.
[0167] In the process of jointly learning a label mapping set for all data sets, each data set has its own label set for prediction, and the predicted values form a public label space L. There is a mapping relationship between the public label space L and the label set Tk of the label mapping set, and the label set Tk of the label mapping set is the aligned public label space L. The one-to-one mapping relationship between the labels of each data set and the labels in the label mapping set means that the label c of the service category in the public label space L is mapped to at most one label ^c of the service category in the label mapping set Tk. The labels of the label mapping set are all accurately matched to the labels in the label mapping set.
[0168] In the process of training the target detection model to be trained, a set of bounding boxes and their predicted values of each training image data set can be given, and the prediction score of the neural network model can be determined by the output result of the neural network model. The prediction score can be used to control the training progress of the target detection model and can be used to evaluate the prediction ability of the trained target detection model.
[0169] Specifically, each training image dataset has its own label set 1...label set k (labelspace L1...Lk), and each training image dataset has its own label set for prediction, and the predicted values form a common label space L (common label space), and a common label space L (common label space) is obtained, and a mapping from the common label space L (common label space) to the label mapping set (dataset-specific label space) is defined: Tk:L→Lk, which can be a Boolean linear transformation. In other words, after obtaining the initial training image dataset from the data source, the initial training image dataset is predicted through the detection head to obtain the result of a projection - the label in the common space; and the labels in the common space are aligned to obtain the result of a secondary projection - the label in the mapping label set.
[0170] Among them, each label joint label:c∈L in the public label space is mapped to at most one label dataset-specific label c^∈Lk in the mapping label set; and the inner label dataset-specificlabel in the mapping label set just matches a label joint label in the public label space; that is, the label in the public label space is aligned with the mapping label set, the two do not contain repeated labels, and the labels are in a one-to-one correspondence. Given a set of detection heads (partitioned detectors) output for classification of different training image datasets (bboxbi), a joint detection score di is constructed by averaging the output common classes. Exemplarily, the score for classification prediction of the training image dataset L1 is represented as di1∈R|L1|, the score for classification prediction of the training image dataset L2 is represented as di2∈R|L2|... The score of the training image dataset Lk is represented as dik∈R|Lk|, and so on. After the scores of each training image dataset are comprehensively calculated, the score (joint detection score) of the joint training of the target detection model is obtained.
[0171] The advantage of using multiple detection heads is that the same target detection model can adapt to the detection of multiple training image data sets at the same time, forming a common label space for the images of each data set, so that the trained target detection model can be applicable to various types of reference area images, accurately identify the location information of multiple types of identifiable objects, and have high robustness, and can maintain and guarantee the accuracy of recognition under certain parameter perturbations.
[0172] In an exemplary embodiment, the scores of each training image data set are comprehensively calculated to obtain a score of the joint training of the target detection model (joint detection score), and the score of the joint training of the target detection model is represented by di, and its calculation formula is as follows:
[0173]
[0174] Among them, di is the joint training score of the target detection model, Tk is the prediction value matrix of the k-th image training set, is the predicted value matrix after the k-th image training set is transposed, and L is the loss function of the target detection model.
[0175] Among them, the target detection model includes two parts: a partitioned detector and a unified detector; the partitioned detector refers to the detection heads of the training image data sets in the target detection model to be trained; the unified detector is trained by the backbone network of the shared neural network model of the target detection model to be trained, and the backbone network may include a network layer (neck) that mixes and combines image features.
[0176] In an exemplary embodiment, target detection is performed on C class labels through a training set of k images, and its loss function includes two loss functions, which are respectively related to distortion and average precision. The distortion measures the difference in detection scores between the partitioned detector and the unified detector:
[0177]
[0178] The distortion is is the predicted value of the detection head in the k-th image training set in the C category label; is the mapping label of the detection head of the k-th image training set to the C category label.
[0179] The average precision is the average precision measurement of the output results of the mapped label set on the validation set corresponding to the image training set through each label output in class c:
[0180]
[0181] in, is the average precision, |L k | is the number of labels in the k-th image training set, is the averaged result of the predicted value of the detection head in the k-th image training set in the C category label, It is the mapping label averaging result of the detection head of the k-th image training set in the C category label.
[0182] Each partition detector is trained together through these two loss functions, and their output spaces are merged after training. For this process, different detection heads are used, and different detection heads adopt their own training protocols and losses for each image training set; different detection heads are used, and different detection heads adopt their own training protocols and loss functions for each image training set, and the outputs share a common detection architecture to automatically integrate the outputs of different detection heads into common semantics to form a mapping label set. And when there is a mapping label set, a fully automatic method using only visual data can be used to unify the output space of the multi-dataset detection system. Similar new targets are detected using target detectors of similar concepts from different data sets, thereby achieving a general target detection model of this embodiment.
[0183] The effect of this embodiment can be explained by comparing different images. The most traditional target detection model is Figure 6 As shown in (a), the traditional solution uses training image datasets such as COCO, Objects365, and OpenImages. Each training image dataset corresponds to a single model. In this way, models trained with different training image datasets cannot be used to detect targets in other datasets. They all detect targets of the same target vehicle category and output three repeated detection frames. The slightly improved target detection model is shown in Figure 6 As shown in method (b), a multi-head model is trained, and the target detection model to be trained is a shared model. Different detection heads are used for different data sets, and the loss function (loss) of each data set is calculated separately. In this way, the trained model can recognize targets in multiple data sets, but it does not achieve a unified label space. In this embodiment, the prediction values generated by different detection heads are aligned to the mapping label sets of each training image data set to form a unified label space for all image training sets, and the characteristics of each image training set are fully utilized, so that the trained neural network model has high robustness and can maintain and guarantee the recognition accuracy under certain parameter perturbations.
[0184] In one embodiment, each candidate service is represented by its own candidate service image, and the candidate service image contains both semantic and image dimensional information, so as to more accurately calculate the similarity and determine similar candidate services from the candidate services.
[0185] Correspondingly, the similarity between the reference area image and each candidate service is calculated, and similar candidate services among the candidate services are determined based on the similarity, including: determining the corresponding candidate service images according to the service category; determining the semantic similarity and image similarity of each candidate service image based on the semantic features and image features of the reference area image and each candidate service image; and determining the candidate service images whose semantic similarity satisfies a preset semantic similarity condition and whose image similarity satisfies a preset image similarity condition as candidate service images representing similar candidate services.
[0186] The candidate service images are determined according to the service category, and the category set that can be provided to users is circled. The category set directly divides the candidate services for similarity calculation, and services of other service categories will not be designated as candidate services, reducing the possibility of erroneous information search.
[0187] Semantic features are features extracted from the semantic information contained in an image, and image features are visual features of an image. Both semantic features and image features are divided according to service categories. For example, if an image is of a red object A, its semantic features include features that can be represented semantically, such as red, object A, and parts of object A. Its image features include a hue feature distribution map of different positions of red, a structural feature map of object A, a feature map of parts of object A, and a feature map of the parts installation structure of object A.
[0188] In a feasible implementation, based on the semantic features and image features of the reference area image and each candidate service image, the semantic similarity and image similarity of each candidate service image are determined, including: extracting the semantic features and image features of the reference area image to obtain reference semantic features and reference image features; acquiring the semantic features and image features of the candidate service image to obtain candidate semantic features and candidate image features; calculating the similarity between the reference semantic features and the candidate semantic features to obtain the semantic similarity of each candidate service image; and calculating the similarity between the reference image features and the candidate image features to obtain the image similarity of each candidate service image.
[0189] Among them, in order to improve the efficiency of feature extraction, each service category can have its own corresponding feature extraction model; the semantic features and image features of the reference area image are extracted by the feature extraction model of each service category to obtain the reference semantic features and reference image features of the reference area image; and the semantic features and image features of the candidate service image can be obtained in advance.
[0190] In one embodiment, the similarity includes a semantic similarity calculated based on a feature dictionary; the feature dictionary includes the semantics of each candidate service, and the semantics of each candidate service belongs to at least one service type; according to each reference semantic feature extracted from the reference area image, each dictionary semantic feature whose similarity with each reference semantic feature meets a preset dictionary semantic similarity condition can be searched in the semantic features of the feature dictionary, and then each dictionary semantic feature is combined into each candidate service dictionary set; each candidate service is determined according to each candidate service dictionary set, and the semantic similarity of each candidate service is determined according to the similarity between each semantic feature of each candidate service dictionary set and the reference semantic feature, so as to improve processing efficiency.
[0191] The process of generating a feature dictionary includes: dividing the dictionary construction images into candidate service image sets according to preset service categories; determining the dictionary construction images in the candidate service image sets as candidate service images respectively; and generating feature dictionaries for each candidate service set based on semantic features extracted from each candidate service image.
[0192] The dictionary construction image is an image that represents a specific service of a preset service category and is used to construct a semantic dictionary. The data source of the dictionary construction image is not specifically limited, and it can be part of the images in the training image set, images from other image sets, or images constructed from a graph feature set. Optionally, the dictionary construction image includes specific services that represent at least one level of service in the service category. Exemplarily, the specific services of a service category A may include A1 level services and A11 level services under the A1 level services, and the dictionary construction image may be an image that represents the A1 level service, or an image that represents the A11 level service.
[0193] The candidate service image includes a dictionary construction image divided according to preset service categories, and there is a certain dictionary correspondence between the candidate service image and the dictionary construction image. Through the dictionary correspondence, the dictionary construction image can be determined according to the candidate service dictionary set composed of the candidate service image, and then the specific candidate service is determined by the dictionary construction image.
[0194] In an optional implementation, dividing the dictionary construction image into candidate service image sets according to preset service categories includes: dividing the dictionary construction image into candidate service image sets corresponding to the preset service categories one by one according to the preset service categories. The candidate service image sets corresponding to the preset service categories one by one include candidate service images of at least one service level.
[0195] In an optional embodiment, based on the semantic features extracted from each candidate service image, a feature dictionary for each candidate service set is generated respectively, including: according to a preset service category, using the semantic features extracted from each candidate service image as the feature dictionary mapping value of each candidate service set, and generating an index of the feature dictionary mapping value.
[0196] According to the preset service category, the dictionary construction image is divided into a set of candidate service images, which realizes the semantic feature splitting of the dictionary construction image. Based on the semantic features extracted from each candidate service image, a feature dictionary for each candidate service set is generated respectively, which is a reorganization of the semantic features of the dictionary construction image. By splitting and reorganizing the semantic features of the dictionary construction image, the semantic features of each dictionary construction image are arranged in an orderly manner in the feature dictionary, which helps to more efficiently calculate the semantic similarity between the reference area image and each candidate service.
[0197] In one embodiment, the similarity includes an image similarity calculated based on a preset service image feature set; the preset service image feature set includes preset image features of each candidate service, and the preset image features of each candidate service belong to at least one service type; according to each reference image feature extracted from the reference area image, each preset image feature whose similarity with each reference image feature satisfies the preset image feature similarity condition can be searched in the image features of the preset service image feature set, and then each preset image feature is combined into each candidate service image set; each candidate service is determined according to each candidate service image set, and the image similarity of each candidate service is determined according to the similarity between each preset image feature of each candidate service image set and the reference image feature, so as to improve processing efficiency.
[0198] The steps for generating a preset service image feature set include: dividing the image constructed by the graph feature set into a set of candidate service images according to the preset service category; determining the images constructed by the graph feature set in the set of candidate service images as candidate service images respectively; and clustering the candidate service images according to similar features to generate a preset service image feature set.
[0199] The graph feature set construction image is an image that represents a specific service of a preset service category and is used for the preset service image feature set. The data source of the graph feature set construction image is not specifically limited, and it can be part of the images in the training image set, images in other image sets, or dictionary construction images. Optionally, the graph feature set construction image includes a specific service that represents at least one level of service in the service category.
[0200] The candidate service image includes an image constructed by a graph feature set divided according to preset service categories, and there is a certain graph set correspondence between the candidate service image and the image constructed by the graph feature set. Through this graph set correspondence, the graph feature set construction image can be determined according to the candidate service dictionary set composed of the candidate service image, and then the specific candidate service is determined through the graph feature set construction image.
[0201] In an optional implementation, dividing the image constructed by the graph feature set into a set of candidate service images according to the preset service category includes: dividing the image constructed by the graph feature set into a set of candidate service images corresponding to the preset service category one by one according to the preset service category. The set of candidate service images corresponding to the preset service category one by one includes candidate service images of at least one service level.
[0202] In an optional embodiment, clustering is performed based on similar features between the candidate service images to generate a preset service image feature set, including: according to the preset service category, using the image features extracted from each candidate service image as the preset atlas mapping value of each candidate service set, and generating an index of the preset atlas mapping value.
[0203] According to the preset service category, the image constructed by the graph feature set is divided into a set of candidate service images, which realizes the image feature splitting of the image constructed by the graph feature set. The images constructed by the graph feature set in the set of candidate service images are respectively determined as candidate service images, which is a reorganization of the image features of the image constructed by the graph feature set. By splitting and reorganizing the image features of the image constructed by the graph feature set, the image features of each image constructed by the graph feature set are arranged in an orderly manner into the preset service image feature set, which helps to more efficiently calculate the image similarity between the reference area image and each candidate service.
[0204] In an optional implementation, the service categories include at least one of items, attractions, hairstyles, and food; each candidate service includes at least one resource of each candidate item service corresponding to items, each candidate hairstyle service corresponding to hairstyles, each candidate attraction service corresponding to attractions, and each candidate food service corresponding to food. By dividing the service categories, candidate services, and candidate service resources in this way, similarity calculation can be performed more efficiently, similar candidate services of the reference area image can be determined more accurately, and the resources of the similar candidate services can be determined.
[0205] Optionally, objects, attractions, hairstyles and foods respectively have similarity calculation models corresponding to their categories. The similarity calculation models of the above categories can be used to more accurately determine the similar candidate services and the resources of the similar candidate services of the reference area image, and then the resource points can be pushed more efficiently for the resources of the similar candidate services, and the paths to the resource points can be planned.
[0206] In an optional implementation, the reference area image is a reference avatar image, the service category includes a hairstyle category, and each candidate service includes candidate hairstyles corresponding to the hairstyle category. The reference avatar image includes but is not limited to a human or animal avatar. In the process of searching based on the reference avatar image, feature extraction can be performed based on each candidate hairstyle corresponding to the hairstyle category, rather than feature extraction based on the face image, so as to increase the resources required by the terminal.
[0207] Based on the semantic features and image features of the reference area image and each candidate service image in the candidate service set, the image similarity and image similarity of each candidate service image are determined, including: extracting the reference hairstyle features of the reference avatar image; and determining the semantic similarity and image similarity of each candidate hairstyle according to the reference hairstyle features and the semantic features and image features of each candidate hairstyle.
[0208] The reference hairstyle feature is a feature extracted from the hair part of the reference head image, which involves the shape, length, distribution, color and other aspects of the hair to obtain the candidate hairstyle corresponding to the reference area image.
[0209] Optionally, the reference avatar image that is not segmented is as follows: Figure 7 As shown in (a), the segmentation of the reference hairstyle feature is as follows Figure 7 (b) As shown in FIG. Correspondingly, extracting the reference hairstyle features of the reference avatar image includes: identifying the semantic features and image features of the reference avatar image; inputting the semantic features and image features of the reference avatar image into the avatar segmentation model for hairstyle segmentation, so as to obtain the reference hairstyle semantic features and reference hairstyle image features from the semantic features and image features of the reference avatar image.
[0210] After determining the semantic similarity and image similarity of each candidate hairstyle, the method further includes: determining the candidate hairstyle images whose semantic similarity satisfies a preset semantic similarity condition and whose image similarity satisfies a preset image similarity condition as candidate hairstyle images representing similar candidate services.
[0211] By finely segmenting the avatar, information search is performed on the avatar hairstyle of the reference area image to more efficiently retrieve the specific hairstyle. Candidate search results such as barber shops, hair salons, beauty salons or hair care salons that can provide hairstyle-related services are searched to determine the corresponding resource point push program or path planning program. Exemplarily, through the resource point push program, the hair salons that provide hairstyle services are displayed to the terminal in the form of a list. At the same time, the geographical location information and navigation buttons of the barber shops near the terminal are displayed. The user can directly click the navigation button to jump to the path planning program to navigate to the corresponding barber shop location through the path planning program.
[0212] In an exemplary embodiment, after determining the candidate services according to the service category, the server determines similar candidate services according to the similarity between the reference area image and the candidate services, and then weights the similar candidate services according to the user information to obtain the scoring information of the similar candidate services, arranges the scoring information of the similar candidate services from high to low, and determines the top five candidate services with the scoring information, forms a candidate service list with the top five candidate services, and feeds the candidate service list back to the terminal as the candidate search results.
[0213] In one embodiment, user information includes user portraits and user scenarios; user portraits are used to determine group characteristics of users to which the terminal belongs, and are long-term user information; they include but are not limited to user attributes such as user age, user ID, user preferences such as food cuisine preferences, clothing styles, living habits such as common means of transportation, frequency of use of items, user behaviors such as online shopping frequency and offline shopping, and other information abstracted from the user feature model; user scenarios are used to determine the environmental factors for the terminal to search for information, and are short-term user information or real-time user information; they include but are not limited to information such as the type of user's location, the time for searching the reference area image, and the weather environment of the reference area image.
[0214] Candidate search results are obtained by screening similar candidate services according to user information, including: weighting each similar candidate service according to the user portrait and the user scenario to obtain score information of each similar candidate service; the weight coefficient of the user portrait is greater than the weight coefficient of the user scenario; and similar candidate services whose score information meets preset conditions are determined as candidate search results.
[0215] The scoring information of similar candidate services is information obtained by evaluating similar candidate services according to user portraits and user scenarios after similar candidate services are determined according to similarity with reference area images. Optionally, the scoring information of similar candidate services is scoring information used to adjust the similarity according to user information; when there are similar candidate services of multiple service categories, the weight coefficient of the service category is greater than the weight coefficient of the user portrait, and the weight coefficient of the user portrait is greater than the weight coefficient of the user scenario.
[0216] On the premise that the weight coefficient of the user portrait is greater than the weight coefficient of the user scenario, each similar candidate service is weighted according to the user portrait and the user scenario. The user's long-term habits are mainly considered, supplemented by short-term user information or real-time user information, to evaluate each similar candidate service to obtain the scoring information of each similar candidate service.
[0217] In an optional implementation, the scoring information is a matching score obtained by comprehensive evaluation based on the weight coefficients of the service category, user portrait and user scenario, and similar candidate services whose scoring information meets preset conditions are determined as candidate search results, including: similar candidate services whose matching scores meet the matching score range are determined as candidate search results; wherein the matching score range can be the matching scores arranged in sequence within the preset range.
[0218] By weighting similar candidate services through user portraits and user scenarios, the arrangement of similar candidate services is adjusted to make the candidate search results more in line with user needs and improve the efficiency of user information search. Therefore, cross-end services are provided for the personality of user information matching, and the data of cross-end services is a combination of multi-end data and data from various user scenarios. With the help of the cloud basic platform, a multi-dimensional fusion analysis model can be built to realize user personalized services or product recommendations, which can provide users with more personalized services.
[0219] In an optional implementation, the user scenario includes indoor and outdoor scenario features of the user, and distance scenario features between the terminal and the resource points corresponding to the similar candidate services. The indoor and outdoor scenario features of the user belong to one of the location types, and are used to determine whether the user is indoors or outdoors. They belong to the scenario information within the terminal location range, so as to provide similar candidate services for indoor and outdoor scenarios; the distance scenario features belong to one of the environmental factors, and the distance scenario features can reflect the distance between the user and each resource point, so as to provide similar candidate services considering the path, and belong to the path scenario information between the terminal and the resource point.
[0220] Based on the user portrait and the user scenario, each similar candidate service is weighted to obtain the scoring information of each similar candidate service, including: obtaining the portrait weight coefficient of each candidate service based on the user portrait; determining the scene weight coefficient of each similar candidate service based on the user's indoor and outdoor scene characteristics and the distance scene characteristics; and weighting each similar candidate service based on the portrait weight coefficient and the scene weight coefficient to obtain the scoring information of the similar candidate services.
[0221] The scenario weight coefficient of each similar candidate service is the result of comprehensive consideration of the scenario information within the terminal location range and the path scenario information between the terminal and the resource point. It can more specifically screen similar candidate services for users to obtain exclusive rating information for the group to which the user belongs, thereby improving the efficiency of information search.
[0222] In an optional implementation, candidate services whose scoring values meet preset conditions are determined as candidate search results, including: sorting similar candidate services from high to low according to their scoring information to obtain sorted similar candidate services; among the sorted similar candidate services, similar candidate services that meet preset search conditions are used as candidate search results.
[0223] The sorted similar candidate service is at least one similar candidate service obtained by sorting according to the scoring information. Since the sorted similar candidate services are sorted from high to low according to the scoring information of the similar candidate services, if there is a similar candidate service that will meet the preset search condition among the sorted similar candidate services, it can be directly used as a candidate search result to improve the sorting efficiency, thereby improving the information search efficiency.
[0224] In an optional implementation, the preset search conditions include a stop search condition and a preset range condition of the location; among the sorted similar candidate services, the similar candidate services that meet the preset search conditions are used as candidate search results, including: determining whether the sorted similar candidate services meet the stop search conditions, and if so, using the similar candidate services ranked within the preset range as candidate search results.
[0225] In a feasible implementation, each similar candidate service is weighted according to the user portrait and the user scenario to obtain the scoring information of each similar candidate service, including: obtaining a similarity threshold; selecting from candidate services above the similarity threshold, and merging repeated similar candidate services to obtain screened similar candidate services; determining the initial weight information of the screened similar candidate services according to the service category, and then weighting the screened similar candidate services according to the respective weight coefficients of the user portrait and the user scenario to obtain the scoring information of each similar candidate service.
[0226] Correspondingly, similar candidate services are sorted from high to low according to their score information to obtain sorted similar candidate services, including: sorting the similar candidate services from high to low according to their score information, and using the knowledge graph to add significant common sense to obtain sorted similar candidate services, so as to help the sorting be more reasonable, thereby returning candidate services that are more targeted to the needs of the user.
[0227] In one feasible implementation, the preset search conditions include a preset search quantity; among the sorted similar candidate services, similar candidate services that meet the preset search conditions are used as candidate search results, including: if the sorted similar candidate services reach the preset search quantity, all the sorted similar candidate services are used as candidate search results. Exemplarily, since the scoring information of the similar candidate services is sorted from high to low, if there are 5 sorted similar candidate services, all the 5 sorted similar candidate services are used as candidate search results, and this information search is stopped. Thus, according to the scoring information of the similar candidate services from high to low, and among the sorted similar candidate services, similar candidate services that meet the preset search conditions are used as candidate search results. There is no need to sort the scoring information of all similar candidate services, only a certain range of scoring information needs to be selected for sorting. The candidate search results have a higher search efficiency. The scoring information of different reference area images, such as Figure 8 shown.
[0228] In an exemplary embodiment, the candidate search results are implemented using image retrieval technology, and its architecture is mainly composed of offline processes and online processes. The offline process mainly constructs an index for calculating the similarity of each candidate service in the two dimensions of semantics and image, and the online process mainly determines the retrieval candidate service based on the index.
[0229] like Fig. 9 As shown in (a), in the offline process, we first need to create a category list and build a set of candidate service images grouped by category. The candidate service image set is similar to a container with a large amount of data. Each candidate service image in the container needs to be feature learned and extracted to generate a semantic dictionary. In the process of calculating image similarity, we mainly train the image feature search model for different features of the product image, such as color, shape, and patterns, and then cluster the similar features between the candidate service images to generate a preset service image feature set. The semantic dictionary and the preset service image feature set are stored in the cloud database, and there is an index for calculating the similarity of each candidate service. The above construction process can refer to the face library construction process of the face recognition algorithm, except that the candidate service images need to be carefully maintained to ensure their integrity and timeliness.
[0230] The online similarity calculation process is as follows Fig. 9(b) is shown; when the user retrieves an image, the server obtains the target image uploaded by the system, extracts features of the target image, obtains the feature value of the target image, and estimates the group belonging to the service category according to the feature value of the target image, so as to realize the category analysis of the candidate service, so that the terminal responds to the information search event generated by the trigger operation of the target image and obtains the reference area image from the target image. Then, based on the semantic dictionary of the service category and the preset image set, the similarity index with the reference area image is performed to determine the candidate service image whose semantic similarity meets the preset semantic similarity condition and whose image similarity meets the preset image similarity condition as the candidate service image representing the similar candidate service; then, according to the score information of the similar candidate service, the similar candidate service is sorted from high to low to obtain the sorted similar candidate service, and the data form of the sorted similar candidate service can be a candidate service ranking list; then, among the sorted similar candidate services, the similar candidate service that meets the preset search condition is used as the candidate search result to realize the re-arrangement of the candidate service ranking list, and finally, the candidate search result is returned to the terminal. Optionally, the candidate search results in the list can be filtered according to user information, and the candidate search results can be fed back to the user; the interface of the push application of each resource point such as the shopping website can also be called, and then the interface fed back by the push application of each resource point can be filtered and optimized, and recommendations can be made based on the filtered and optimized interface.
[0231] Among them, the similarity index means that according to the visual image features and the semantic features of the visual connotation, based on the above-mentioned "dictionary" and the preset image feature set, the candidate service images whose semantic similarity meets the preset semantic similarity conditions and the image similarity meets the preset image similarity conditions are determined as candidate service images representing similar candidate services, thereby determining similar candidate services among the candidate services.
[0232] In an optional implementation, the method further includes: recording the target search results selected from the candidate search results and the access data of the target search results according to the user information; adjusting the user portrait in the user information according to the target search results and the access data; and determining the portrait weight coefficient of each candidate service according to the adjusted user portrait.
[0233] The target search result is the selected candidate service, and the selected candidate service corresponds to the matching location data, and the access data is used to execute the service provided by the target search result. Exemplarily, the target search result is food A selected according to the instruction issued by the terminal, and the resource point push application corresponding to food A is the restaurant takeaway push application of food A, and the access data of the target search result includes the takeaway page access data of the restaurant takeaway push application, and the order page access data.
[0234] In one feasible implementation, the target search result selected from the candidate search results and the access data of the target search result are recorded according to the user information, including: according to the user identifier as an index, the target search result selected from the candidate search results and the access data of the target search result are recorded to generate the target search result selected by the user identifier and the access data of the target search result.
[0235] According to user information, the target search results and the access data of the target search results are recorded, and the user portrait is adjusted accurately and timely. Then, based on the adjusted user portrait, the portrait weight information of each candidate service is made closer to the needs of the user of the terminal, thereby improving the efficiency of the user's information search. Personalized highly relevant service recommendations are made to users to meet their needs of comparing candidate services of the same service category, increase their options for candidate services, and improve information search efficiency.
[0236] Based on this, the image search function mobile phone camera and screenshot system is used to meet the user's instant shopping and service needs, improve service quality and shopping experience; use multi-terminal data to build a multi-dimensional fusion analysis model to determine the candidate services for user personalized services or product recommendations, and the candidate services are executed through resource point push applications and route planning applications. Resource point push applications and route planning applications include but are not limited to applications for providing shopping links, local store information, scenic spot information, map information, etc. for candidate services. In the multi-dimensional fusion analysis model, mobile phones can be used as terminals and cloud servers to cooperate. Both mobile phones and cloud servers can serve as good intermediary media for accessing third-party services. They can fairly access a variety of third-party services and shopping apps, meet users' more personalized needs, and bring a more extreme user experience, so as to comprehensively improve different services, meet more types of more personalized needs, open up a fast direct connection channel between instantaneous needs and services in reality, and combine cross-platform and cross-terminal system programs to understand the service status at any time; and through massive data, a user intention recognition system is formed, and combined with the user's personal portrait, it can meet the user's more accurate needs.
[0237] In summary, the information search method is used as a window to quickly capture user needs. Through the user information of the terminal, the user browsing is increased, and the end-cloud algorithm is used to accurately guide the traffic to various resource points to push applications, so that users can obtain accurate candidate search results. In addition, the user's product click information and search information can be used as the basis for information recommendation, so as to push more accurate information to users.
[0238] In an alternative embodiment, Fig.10As shown, first, a target image is photographed or captured by a terminal, and in response to an information search event generated by a trigger operation on the target image, the target image is uploaded to a server, and the object location information returned by the server is received; then, in response to a selection operation on an area identifier, the area identifier acted upon by the selection operation is determined as a reference area identifier, and the area represented by the reference area identifier is determined as a reference area; or, in response to a selection box added to the target image, the area contained in the selection box is determined as a reference area; then, based on the user portrait and the user scenario, each similar candidate service is weighted to obtain scoring information for each similar candidate service; based on the scoring information for each similar candidate service, candidate search results are screened from the similar candidate services; and the candidate search results are sent to the terminal.
[0239] In an optional embodiment, the terminal embeds the "image-guided service" into the mobile phone's camera and screenshot system to improve information search efficiency, meet user needs instantly, and enhance the user's service quality and shopping experience; the process is as follows:
[0240] First, when the user has a need for immediate shopping in life, he takes out his mobile phone to take a photo of the item or takes a screenshot of the mobile phone, and when a service request is triggered, the system automatically uploads the picture to the cloud server. This process belongs to step 202 and step 302 therein.
[0241] Secondly, the cloud service receives the request, performs object detection on the image, marks the location of the main objects, and synchronizes the information to the terminal side. This process belongs to step 304, step 306, step 502 and corresponding embodiments.
[0242] Again, the user selects items based on the identified objects. If none of them are what the user wants, the user can frame the items by himself. The coordinates of the selected item area frame will be uploaded to the cloud again. This process belongs to step 308 and corresponding embodiments.
[0243] Then, the cloud calls the image search service algorithm to perform a similarity search on the items selected by the user, obtains a list of services with high similarity, and identifies candidate services to obtain service categories. This process belongs to step 504 and corresponding embodiments.
[0244] Then, the cloud calculates the similarity between the reference area image and each candidate service among the candidate services corresponding to the service category; this process is embodied in step 506 .
[0245] Next, based on the user information, a candidate service list that is most likely to satisfy the user is selected from the product list. This process is reflected in step 508.
[0246] Finally, the candidate service list is fed back to the client; at the same time, the user's transaction and traffic diversion status are counted, uploaded to the cloud, and multi-terminal information is circulated and synchronized.
[0247] In one embodiment, Fig.11 As shown, the information search method includes three parts: capturing user needs on the terminal side, intelligent personalized analysis of users on the cloud side, and generating personalized candidate services on the terminal. These three parts together realize seamless intelligent cross-terminal services. Product or service information will be synchronized to smart watches, car terminals, etc., and service information such as products, logistics, takeout, travel, and users' immediate preferences will be synchronized and updated to multiple related devices in real time.
[0248] Optionally, the terminal captures user needs, including: capturing user needs instantly through screenshots and photo taking functions; intelligent personalized analysis of users on the cloud side, including: combining geographic location, time, scene, user portrait, and image information to analyze the products that users are interested in, or other services they expect; generating personalized candidate services on the terminal: the terminal side determines the resource points with the highest frequency of use in the service category according to the user's frequency of use of the application, and pushes applications or route planning applications through the resource points with the highest frequency of use in the service category to quickly determine the target search results. Exemplarily, the terminal selects a third-party APP jump or quick order to allow users to quickly determine the target search results.
[0249] Among them, the intelligent personalized analysis of users on the cloud side is based on the capture of user needs on the terminal side and the results of scene perception; the generation of personalized candidate services on the terminal side is based on the results of intelligent personalized analysis of users on the cloud side and the results of scene perception. Scene perception is based on information such as user geographic location, time, external environment, and user APP usage habits.
[0250] In an exemplary embodiment, Fig.12 As shown, the user takes a photo or screenshot through the terminal, and the terminal function triggers an event to upload the target image to the cloud, and the cloud image object is detected and fed back to the terminal; the user clicks on the corresponding object or directly selects the corresponding object; the terminal uploads the image of the reference area divided by the image to the cloud; the cloud target recognizes the name of the object and determines the service link of similar candidate services with the image search service; the cloud combines the user portrait, user location, and the similarity of similar candidate services to filter and sort the service links of similar candidate services to obtain the top 5 candidate service list; the client receives and presents the top 5 service images or related service links, the user selects the service, jumps to the third-party APP to the corresponding service page, and feedbacks the order information; the user portrait is updated; and information flows across multiple terminals.
[0251] In an exemplary embodiment, Fig.13As shown, the target detection model is an object detection model, the candidate services are items, the selection operation includes clicking on the item of interest or self-selecting the item, the reference area image is the image of interest, the candidate services are service links, the candidate services are represented by candidate service images, and the candidate service images are associated with the service links.
[0252] Correspondingly, the information search method includes: the terminal user takes a photo or screenshot with a mobile phone to obtain a target image; uploads the target image to the cloud; the cloud analyzes the objects in the target image through an object detection model; the cloud feeds back the object location information to the terminal; the user clicks on the object of interest or selects the object by himself through the terminal to obtain the image of the area of interest; the cloud recognizes the object in the image through an object recognition model to obtain the image of the area of interest and the name of the object; the cloud can choose to develop its own algorithm or directly call a third-party API interface to execute the corresponding process of the image search service model; after executing the image search service model, the cloud can obtain the service link, and then calculate the similarity through the services related to the service link to obtain similar candidate services and product links of the reference area image, that is, all matched service and product links; then, the cloud screens and sorts all matched service and product links according to the user portrait, object location information, user location information, and product similarity to obtain the top 5 ranked service links and candidate service images; then, the cloud sends the top 5 ranked service links and candidate service images to the terminal. The terminal then displays the top 5 service links and candidate service images to the user through a pop-up window; when the user selects a service link and candidate service image, the target service result is obtained; based on the target service result, the user is redirected to the corresponding product page of the third-party application, and after the user places an order, information is transferred across multiple terminals.
[0253] In an exemplary embodiment, the user obtains the target image by taking photos or screenshots of various terminal devices (such as mobile phones, tablets, and watches), and triggers the information search service of the target image. The terminal uploads the target image to the cloud in real time, and the cloud detects the main items in the photos uploaded by the user and feeds the results back to the mobile phone. The user clicks or selects the items of interest in the image on the terminal on the mobile phone image to obtain the reference area image, which is uploaded to the cloud again. The cloud recognizes the image of the area of interest to the user, and then classifies it according to the image type.
[0254] Regardless of the above, image retrieval is required to match the corresponding relevant product image or service image, and feedback the service image to the corresponding product link list; at the same time, the retrieved product list is personalized filtered and sorted and recommended by the user in combination with scene perception and cloud user portraits, and synchronized with the client. Finally, the user makes a choice based on the feedback results and personal preferences, and clicks to jump to the third-party application. For example, if a user takes a picture and clicks on an item, hairstyle, attraction or food in the picture, then the corresponding physical service link or non-physical service link of the application is recommended based on the personalized best matching service resource point, or the system application is called through the deep personalized customization system in the cloud to directly jump or quickly place an order; at the same time, the corresponding order information is pushed to the user's multi-end associated devices through the cloud system, and the user can follow and modify his or her order and service information through any of his or her end-side devices. For example, if a user takes a picture of someone and taps the hair area, then we can recommend the nearby Meituan barber shop link and barber shop location to the user based on the user's location, or the deep personalized customization system in the cloud can ask the user to navigate directly to the store location;
[0255] In one embodiment, Fig.14 As shown, the terminal application performs image acquisition by taking photos or screenshots, and uploads the acquired target image to the object detection platform of the cloud server. The object detection platform detects and feeds back the object location information of the object to the terminal. When the user of the terminal clicks or selects the target image, a screenshot of the area of interest is taken to obtain a reference area image; the terminal uploads the reference area image to the cloud according to the user ID, user location and reference area image; the cloud then determines a candidate service list from similar candidate services of the service category to which the reference area image belongs based on the user information of the terminal, and returns the candidate service list to the terminal; the terminal displays the candidate service list in a pop-up window for the user to call a third-party program.
[0256] In an exemplary embodiment, Fig.15 As shown, the method includes: after the terminal obtains the reference area image, the terminal responds to the information search event of the target image, takes the target image as the reference area image, uploads the reference area image to the server, and receives the candidate search results returned by the server; the candidate search results include the candidate service images for which a resource point pushes the application program; wherein the reference area image is as follows Fig.15 As shown in (a), the resource point push application is Fig.15 As shown in (b), the candidate service images are Fig.15 (c) As shown. It can be understood that the reference area image can also be Fig.16 (a) Using the reference area image segmented from the target image, the resource point push application can also Fig.16 (b) shows that multiple service categories are included.
[0257] In an exemplary embodiment, Fig.17 As shown in the figure, the offline process and online process of this method refer to: the offline process refers to the entire process of offline index library construction, involving image selection of each category, offline image cutting, offline feature extraction, semantic dictionary and index construction of preset image set. Online process: includes key steps such as online category recognition, online image cutting, online feature extraction and online index query.
[0258] In an exemplary embodiment, the service category is an object category, and the object category includes 14 object categories, and the 14 object categories and their diagrams are as follows: Fig.18 As shown, it includes men's tops, skirts, women's bottoms, luggage, shoes, bottled drinks, snacks, furniture, beauty products, toys, digital products, underwear and other categories of items.
[0259] In an exemplary embodiment, because the background of the commodity is complex and the main object is often small, in order to reduce the influence of a large amount of background interference and multiple subjects, the search target needs to be extracted from the image. The search results of the subject detection are used to determine the candidate area. The search results of the subject detection are as follows: Fig.19 shown.
[0260] In an exemplary embodiment, an image is used as an input medium of a terminal, and a target search result is determined through the terminal to be pushed to an application at a resource point such as a social communication platform or a shopping platform, and the application at the resource point is called to perform information mining, and valuable information in the content ecology of the WeChat platform is obtained, and then displayed to the user in the form of e-commerce + encyclopedia + information. Fig. 20 shown.
[0261] In an exemplary embodiment, whether it is a code in various forms or text characters, it can be considered as an image in a specific coding form, which is recognizable. Images of objects or natural scenes can be used as target images or reference area images, and the service understanding ability of natural scene images, such as scanning wine, scanning cars, scanning plants, scanning faces, etc., can also be used as target images or reference area images. The specific service categories are as follows: Fig.21 shown.
[0262] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0263] Based on the same inventive concept, the embodiment of the present application also provides an information search device for implementing the above-mentioned information search method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more information search device embodiments provided below can refer to the limitations on the information search method above, and will not be repeated here.
[0264] In one embodiment, Fig. 22 As shown, an information search device is provided, which is applied to a terminal and includes:
[0265] An image acquisition module 2202, configured to acquire a reference area image from the target image in response to an information search event for the target image;
[0266] An image uploading module 2204 is used to upload the reference area image to a server;
[0267] The receiving module 2206 is used to receive the corresponding candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services of the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
[0268] In one embodiment, the image acquisition module 2202 is used to:
[0269] Acquire a target image obtained by a preset operation mode;
[0270] In response to an information search event generated by a trigger operation on the target image, a reference area image is acquired from the target image.
[0271] In one embodiment, the preset operation mode includes a shooting or screenshot operation.
[0272] In one embodiment, the image acquisition module 2202 is used to:
[0273] In response to an information search event of the target image, uploading the target image to the server;
[0274] Receiving the object position information returned by the server, wherein the object position information is a position mark obtained by the server performing target detection on the target image;
[0275] Displaying a candidate area corresponding to the object position information in the target image;
[0276] A reference region image is obtained from a target image marked with the candidate region.
[0277] In one embodiment, the image acquisition module 2202 is used to:
[0278] In response to a selection operation on a candidate region in the target image, determining the candidate region on which the selection operation acts as a reference region;
[0279] The reference area is extracted from the target image to obtain the reference area image.
[0280] In one embodiment, the target image displays a region identifier, which includes a candidate region identifier and a target image identifier; the image acquisition module 2202 is used to:
[0281] In response to a selection operation on the region identifier, the region identifier affected by the selection operation is determined as a reference region identifier, and the region represented by the reference region identifier is determined as a reference region.
[0282] In one embodiment, the image acquisition module 2202 is used to: in response to a check box added to the target image, determine the area included in the check box as a reference area.
[0283] In one embodiment, the image upload module 2204 is used to:
[0284] The target image is scaled to obtain a target image of a target size, and the target image of the target size is uploaded to the server, wherein the target size is obtained based on a preset side length and a screen size ratio.
[0285] In one embodiment, the device further comprises a service execution module, configured to:
[0286] In response to a selection instruction for the candidate search result, determining a target search result according to the selection instruction, and calling an application corresponding to the target search result to execute a corresponding service;
[0287] In response to a service information synchronization instruction generated during the execution of the corresponding service by the application, the service information is transmitted to the server to instruct the server to synchronize the service information with an associated device associated with the terminal.
[0288] In one embodiment, the service execution module is used to:
[0289] In response to a selection instruction for the candidate search result, determining a target search result according to the selection instruction, determining a resource point push application and a path planning application between the terminal and the target search result according to the resource corresponding to the target search result;
[0290] At least one application program among the resource point push application and the path planning application is called to execute corresponding services.
[0291] In one embodiment, Fig.23 As shown, an information search device is provided, which is applied to a server and includes:
[0292] An image acquisition module 2302 is used to acquire a reference area image; the reference area image is acquired from a target image;
[0293] A service determination module 2304 is used to determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category;
[0294] A service query module 2306 is used to calculate the similarity between the reference area image and each of the candidate services, and determine similar candidate services among the candidate services based on the similarity;
[0295] A result generation module 2308 is used to obtain user information and filter candidate search results from the similar candidate services according to the user information;
[0296] The search feedback module 2310 is configured to send the candidate search results to the terminal.
[0297] In one embodiment, the image acquisition module 2302 is used to:
[0298] Acquire the target image uploaded by the terminal;
[0299] Performing target detection on the target image to obtain object position information;
[0300] The object position information is returned to the terminal, the object position information is used to indicate that the target image displays a candidate area corresponding to the object position information, and the reference area image is obtained from the target image marked with the candidate area.
[0301] In one embodiment, the object position information is obtained by detecting the target image using a pre-trained target detection model, and the training method of the target detection model includes:
[0302] Determining different training image data sets and respective detection heads of the training image data sets in the object detection model to be trained;
[0303] Predicting each of the training image data sets through different detection heads to obtain a predicted value of each of the training image data sets;
[0304] Aligning label values of different training image data sets according to the predicted values of each training image data set to obtain a mapping label set of each training image data set;
[0305] The target detection model to be trained is trained according to each of the training image data sets and the corresponding mapping label sets to obtain a trained target detection model.
[0306] In one embodiment, each of the candidate services is represented by a respective candidate service image, and the service query module 2306 is used to:
[0307] Determine the corresponding candidate service images according to the service category;
[0308] Determining semantic similarity and image similarity of each of the candidate service images based on semantic features and image features of the reference area image and each of the candidate service images;
[0309] The candidate service images whose semantic similarity satisfies a preset semantic similarity condition and whose image similarity satisfies a preset image similarity condition are determined as candidate service images representing similar candidate services.
[0310] In one embodiment, the similarity includes semantic similarity calculated based on a feature dictionary; the information search device further includes a dictionary generation module, the dictionary generation module is used to:
[0311] According to the preset service categories, the dictionary construction images are divided into a set of candidate service images;
[0312] Determine the dictionary construction images in the candidate service image set as candidate service images respectively;
[0313] Based on the semantic features extracted from each of the candidate service images, a feature dictionary for each of the candidate service sets is generated respectively.
[0314] In one embodiment, the similarity includes an image similarity calculated based on a preset service image feature set, and the information search device further includes an image set generation module, the image set generation module is used to:
[0315] According to the preset service category, the image constructed by the graph feature set is divided into a set of candidate service images;
[0316] Determine the graph feature set constructed images in the candidate service image set as candidate service images respectively;
[0317] Clustering is performed according to similar features between the candidate service images to generate the preset service image feature set.
[0318] In one embodiment, the service categories include at least one of items, attractions, hairstyles and food; the candidate services include at least one resource among the candidate item services corresponding to the items, the candidate hairstyle services corresponding to the hairstyles, the candidate attraction services corresponding to the attractions and the candidate food services corresponding to the food.
[0319] In one embodiment, the reference area image is a reference avatar image, the service category includes a hairstyle category, and each candidate service includes candidate hairstyles corresponding to the hairstyle category; the service query module 2306 is used to:
[0320] Extracting reference hairstyle features of the reference avatar image;
[0321] The semantic similarity and image similarity of each of the candidate hairstyles are determined according to the reference hairstyle feature and the semantic features and image features of each of the candidate hairstyles.
[0322] In one embodiment, the user information includes a user portrait and a user scenario; the result generation module 2308 is used to:
[0323] According to the user portrait and the user scenario, weighted processing is performed on each of the similar candidate services to obtain scoring information of each of the similar candidate services; the weight coefficient of the user portrait is greater than the weight coefficient of the user scenario;
[0324] The similar candidate services whose rating information meets the preset conditions are determined as candidate search results.
[0325] In one embodiment, the user scenario includes indoor and outdoor scenario features of the user; the result generation module 2308 is used to:
[0326] Obtaining a portrait weight coefficient of each of the candidate services according to the user portrait;
[0327] Determine a scene weight coefficient of each of the similar candidate services according to the indoor and outdoor scene characteristics of the user and the scene characteristics of the distance between the terminal and the resources corresponding to each of the similar candidate services;
[0328] Each of the similar candidate services is weighted according to the portrait weight coefficient and the scene weight coefficient to obtain the scoring information of the similar candidate services.
[0329] In one embodiment, the result generation module 2308 is used to:
[0330] Sorting the similar candidate services from high to low according to the score information of the similar candidate services to obtain sorted similar candidate services;
[0331] Among the ranked similar candidate services, similar candidate services that meet preset search conditions are used as candidate search results.
[0332] In one embodiment, the information search device further includes a coefficient adjustment module, the coefficient adjustment module is used to:
[0333] According to the user information, recording the target search result selected from the candidate search results and the access data of the target search result;
[0334] Adjusting the user profile in the user information according to the target search results and the access data;
[0335] According to the adjusted user portrait, the portrait weight coefficient of each candidate service is determined.
[0336] Each module in the above information search device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.
[0337] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.24As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store candidate service data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an information search method is implemented.
[0338] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.25 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, an information search method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.
[0339] Those skilled in the art will understand that Fig.24 and Fig.25The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0340] The embodiment of the present application also provides a computer-readable storage medium, one or more non-volatile computer-readable storage media containing computer-executable instructions, when the computer-executable instructions are executed by one or more processors, the processors execute the steps of the information search method.
[0341] The embodiment of the present application also provides a computer program product including instructions, which, when executed on a computer, enables the computer to execute the information search method.
[0342] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.
[0343] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. It is determined to be illustrative rather than limiting, and RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.
[0344] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0345] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. An information search method, characterized in that: Applied to terminals, including: In response to an information search event for a target image, acquiring a reference area image from the target image; Uploading the reference area image to a server; Receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
2. The method according to claim 1, characterized in that The step of acquiring a reference area image from the target image in response to an information search event for the target image comprises: Acquire a target image obtained by a preset operation mode; In response to an information search event generated by a trigger operation on the target image, a reference area image is acquired from the target image.
3. The method according to claim 2, characterized in that The preset operation mode includes a shooting or screenshot operation.
4. The method according to claim 1, characterized in that: The step of acquiring a reference area image from the target image in response to an information search event for the target image comprises: In response to an information search event of the target image, uploading the target image to the server; Receiving object position information returned by the server, wherein the object position information is an object position mark obtained by the server performing target detection on the target image; Displaying a candidate area corresponding to the object position information in the target image; A reference region image is obtained from a target image marked with the candidate region.
5. The method according to claim 4, characterized in that The step of acquiring a reference region image from a target image marked with the candidate region comprises: In response to a selection operation on a candidate region in the target image, determining the candidate region on which the selection operation acts as a reference region; The reference area is extracted from the target image to obtain the reference area image.
6. The method according to claim 5, characterized in that The target image displays a region identifier, wherein the region identifier includes a candidate region identifier and a target image identifier; In response to a selection operation on a candidate area in the target image, determining the candidate area affected by the selection operation as a reference area includes: in response to a selection operation on the area identifier, determining the area identifier affected by the selection operation as a reference area identifier, and determining the area represented by the reference area identifier as a reference area.
7. The method according to claim 5, characterized in that In response to a selection operation on a candidate area in the target image, determining the candidate area affected by the selection operation as a reference area includes: in response to a selection box added in the target image, determining the area included in the selection box as a reference area.
8. The method according to claim 1, characterized in that The step of uploading the target image to the server comprises: The target image is scaled to obtain a target image of a target size, and the target image of the target size is uploaded to the server, wherein the target size is obtained based on a preset side length and a screen size ratio.
9. The method according to any one of claims 1 to 8, characterized in that The method further comprises: In response to a selection instruction for the candidate search result, determining a target search result according to the selection instruction, and calling an application corresponding to the target search result to execute a corresponding service; In response to a service information synchronization instruction generated during the execution of the corresponding service by the application, the service information is transmitted to the server to instruct the server to synchronize the service information to an associated device associated with the terminal.
10. The method according to any one of claims 1 to 8, characterized in that The method further comprises: In response to a selection instruction for the candidate search result, determining a target search result according to the selection instruction, determining a resource point push application according to the resource corresponding to the target search result, and a path planning application between the terminal and the target search result; At least one application program among the resource point push application and the path planning application is called to execute corresponding services.
11. An information search method, characterized in that: Applied to a server, the method comprises: Acquire a reference area image; the reference area image is acquired from the target image; Determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category; Calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity; Acquire user information, and filter candidate search results from the similar candidate services according to the user information; The candidate search results are sent to the terminal.
12. The method according to claim 11, characterized in that The method further comprises: Acquire the target image uploaded by the terminal; Performing target detection on the target image to obtain object position information; The object position information is returned to the terminal, the object position information is used to indicate that the target image displays a candidate area corresponding to the object position information, and the reference area image is obtained from the target image marked with the candidate area.
13. The method according to claim 12, characterized in that The object position information is obtained by detecting the target image using a pre-trained target detection model, and the training method of the target detection model includes: Determining different training image data sets and respective detection heads of the training image data sets in the object detection model to be trained; Predicting each of the training image data sets through different detection heads to obtain a predicted value of each of the training image data sets; Aligning label values of different training image data sets according to the predicted values of each training image data set to obtain a mapping label set of each training image data set; The target detection model to be trained is trained according to each of the training image data sets and the corresponding mapping label sets to obtain a trained target detection model.
14. The method according to claim 11, characterized in that Each of the candidate services is represented by a respective candidate service image, and the calculating of the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity, includes: Determine the corresponding candidate service images according to the service category; Determining semantic similarity and image similarity of each of the candidate service images based on semantic features and image features of the reference area image and each of the candidate service images; The candidate service images whose semantic similarity satisfies a preset semantic similarity condition and whose image similarity satisfies a preset image similarity condition are determined as candidate service images representing similar candidate services.
15. The method according to claim 11, characterized in that The similarity includes semantic similarity calculated based on a feature dictionary; the generation process of the feature dictionary includes: According to the preset service categories, the dictionary construction images are divided into a set of candidate service images; Determine the dictionary construction images in the candidate service image set as candidate service images respectively; Based on the semantic features extracted from each of the candidate service images, a feature dictionary for each of the candidate service sets is generated respectively.
16. The method according to claim 11, characterized in that The similarity includes an image similarity calculated based on a preset service image feature set, and the step of generating the preset service image feature set includes: According to the preset service category, the image constructed by the graph feature set is divided into a set of candidate service images; Determine the graph feature set constructed images in the candidate service image set as candidate service images respectively; Clustering is performed according to similar features between the candidate service images to generate the preset service image feature set.
17. The method according to claim 11, characterized in that The service categories include at least one of items, attractions, hairstyles and food; the candidate services include at least one resource among the candidate item services corresponding to the items, the candidate hairstyle services corresponding to the hairstyles, the candidate attraction services corresponding to the attractions and the candidate food services corresponding to the food.
18. The method according to claim 14, characterized in that The reference area image is a reference avatar image, the service category includes a hairstyle category, and each of the candidate services includes candidate hairstyles corresponding to the hairstyle category; The determining, based on the semantic features and image features of the reference area image and each candidate service image in the candidate service set, the image similarity and image similarity of each candidate service image comprises: Extracting reference hairstyle features of the reference avatar image; The semantic similarity and image similarity of each of the candidate hairstyles are determined according to the reference hairstyle feature and the semantic features and image features of each of the candidate hairstyles.
19. The method according to any one of claims 11 to 18, characterized in that The user information includes a user portrait and a user scenario; the screening of candidate search results from the similar candidate services according to the user information includes: According to the user portrait and the user scenario, weighted processing is performed on each of the similar candidate services to obtain scoring information of each of the similar candidate services; the weight coefficient of the user portrait is greater than the weight coefficient of the user scenario; The similar candidate services whose rating information meets the preset conditions are determined as candidate search results.
20. The method according to claim 19, characterized in that The user scenario includes indoor and outdoor scenario characteristics of the user, and the distance scenario characteristics between the terminal and the resource points corresponding to each of the similar candidate services; The step of weighting each of the similar candidate services according to the user portrait and the user scenario to obtain score information of each of the similar candidate services includes: Obtaining a portrait weight coefficient of each of the candidate services according to the user portrait; Determining a scene weight coefficient of each of the similar candidate services according to the user's indoor and outdoor scene characteristics and the distance scene characteristics; Each of the similar candidate services is weighted according to the portrait weight coefficient and the scene weight coefficient to obtain the scoring information of the similar candidate services.
21. The method according to claim 19, characterized in that The step of determining the candidate service whose score value meets a preset condition as a candidate search result includes: Sorting the similar candidate services from high to low according to the score information of the similar candidate services to obtain sorted similar candidate services; Among the ranked similar candidate services, similar candidate services that meet preset search conditions are used as candidate search results.
22. The method according to claim 11, characterized in that The method further comprises: According to the user information, recording the target search result selected from the candidate search results and the access data of the target search result; Adjusting the user profile in the user information according to the target search results and the access data; According to the adjusted user portrait, the portrait weight coefficient of each candidate service is determined.
23. An information search device, characterized in that: Applied to terminals, including: An image acquisition module, configured to acquire a reference area image from the target image in response to an information search event for the target image; An image uploading module, used for uploading the reference area image to a server; A receiving module is used to receive the candidate search results returned by the server; wherein the candidate search results are determined by the server from similar candidate services corresponding to the service category to which the reference area image belongs based on the user information of the terminal, and the similarity between the similar candidate services and the reference area image meets a preset similarity condition.
24. An information search device, characterized in that: Applicable to servers, including: An image acquisition module, used to acquire a reference area image; the reference area image is acquired from a target image; A service determination module, used to determine the service category to which the reference area image belongs, and obtain candidate services corresponding to the service category; A service query module, used for calculating the similarity between the reference area image and each of the candidate services, and determining similar candidate services among the candidate services based on the similarity; A result generation module, used to obtain user information, and filter candidate search results from the similar candidate services according to the user information; The search feedback module is used to send the candidate search results to the terminal.
25. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the computer program is executed by the processor, the processor is caused to perform the steps of the information search method according to any one of claims 1 to 22.
26. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 22 are implemented.
Citation Information
Cited By
Image recognition method, terminal, server, electronic equipment and storage medium
CN120747887A
