Archive information retrieval method and device, equipment and medium

Abstract

The invention provides an archive information retrieval method and device, equipment and a medium, and belongs to the technical field of data processing.The method comprises the steps that current to-be-retrieved target image data is obtained; performing feature extraction on the target image data to obtain a target feature group, and determining a quality score of a plurality of dimension features included in the target feature group; according to the quality scores of the plurality of dimension features included in the target feature group, determining weights of the plurality of dimension features included in the target feature group; and respectively matching the target feature group with feature groups pre-bound with each piece of archive information, and determining an archive information retrieval result according to the feature matching condition of each obtained feature group and the weight of the multiple dimension features contained in the target feature group.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an archive information retrieval method, device, equipment and medium. Background Art

[0002] The traditional archival information retrieval method is to randomly select a picture from the archival information as the image feature bound to the archival information during the initial archiving. In the subsequent image search, the input picture is compared with the image features bound to each archival information to obtain the archival information retrieval result.

[0003] However, since the characteristics of archival information that can be described by a randomly bound single image feature are limited, false detection or missed detection is prone to occur when retrieving archival information, resulting in low accuracy of archival information retrieval results. Summary of the invention

[0004] In view of the above problems, the embodiments of the present application provide an archive information retrieval method, apparatus, device and medium to overcome the above problems or at least partially solve the above problems.

[0005] In a first aspect of an embodiment of the present application, a method for retrieving archive information is provided, the method comprising: Get the target image data to be retrieved; Performing feature extraction on the target image data to obtain a target feature group, and determining quality scores of multiple dimensional features included in the target feature group; Determining weights of the multiple dimensional features included in the target feature group according to the quality scores of the multiple dimensional features included in the target feature group; The target feature group is matched with the feature groups pre-bound to each piece of archive information respectively, and the archive information retrieval result is determined according to the feature matching conditions of each feature group obtained and the weights of multiple dimensional features included in the target feature group.

[0006] As a possible implementation, the archive information retrieval result is determined according to the feature matching conditions of each feature group and the weights of multiple dimensional features included in the target feature group, including: Determine a first matching score corresponding to each of the feature groups according to feature matching conditions of the feature groups and weights of multiple dimensional features included in the target feature group; When the number of feature groups meeting the first screening condition exceeds a preset number, each feature group meeting the first screening condition is used as a candidate feature group, and the weights of the multiple dimensional features included in the target feature group are adjusted to increase the weight ratio of the target dimensional features in the multiple dimensional features, wherein the first screening condition is used to screen the feature group according to the first matching score; Determining second matching scores corresponding to each of the candidate feature groups according to feature matching conditions of each of the candidate feature groups and the adjusted weights of the multiple dimensional features; Determine a score difference between a first matching score and a second matching score corresponding to each of the candidate feature groups; The archive information retrieval result is determined according to the score differences corresponding to each of the candidate feature groups.

[0007] As a possible implementation, determining the archive information retrieval result according to the score differences corresponding to each of the candidate feature groups includes: In the case that the number of candidate feature groups that meet the second screening condition does not meet the preset number, repeatedly increasing the weight ratio of the new target dimensional features in the multiple dimensional features to re-determine the second matching scores corresponding to each of the candidate feature groups, and re-determine the score difference corresponding to each of the candidate feature groups based on each of the second matching scores obtained last time and each of the second matching scores obtained currently, until the number of candidate feature groups that meet the second screening condition meets the preset number, the repetitive execution of the steps is terminated, wherein the second screening condition is used to screen feature groups according to the score difference and the preset score difference threshold; The archive information pre-bound to each candidate feature group that meets the second screening condition is determined as the archive information retrieval result.

[0008] As a possible implementation, the target dimension feature is determined by the following steps: Receiving indication information of target dimension features from a user; Alternatively, the target dimension feature is determined based on a preset list, where the preset list includes: multiple feature dimensions and their respective corresponding priorities.

[0009] As a possible implementation, the method further includes: For the currently collected image data, feature extraction is performed on the image data to obtain a feature group corresponding to the image data, wherein the feature group corresponding to the image data includes: features with the highest quality scores extracted from multiple feature dimensions respectively; Determine whether there is file information corresponding to the image data in the database; If the archive information corresponding to the image data does not exist in the database, generating the archive information according to the image data, binding the generated archive information with the feature group corresponding to the image data and storing them in the database; When archival information corresponding to the image data exists in a database, the image data is used to update the archival information, and a quality score comparison is performed between a plurality of dimensional features included in a feature group corresponding to the image data and a plurality of dimensional features included in a feature group pre-bound to the archival information corresponding to the image data. The plurality of dimensional features with the highest quality scores obtained after comparison are used to update the feature group pre-bound to the archival information corresponding to the image data.

[0010] As a possible implementation manner, the feature groups pre-bound to each of the archive information are obtained from a cache; and the method further includes: Recording the binding relationship between different archive information and different feature groups, and obtaining the access frequency of the different feature groups; The recorded binding relationship and the hotspot feature group whose access frequency exceeds a preset frequency threshold among the different feature groups are stored in the cache.

[0011] As a possible implementation, the method further includes: Merging multiple archive information retrieval requests received within a set time period into a batch request, where different archive information retrieval requests carry different target image data to be retrieved; The batch requests are processed in parallel using multiple threads to obtain and return the archive information retrieval results corresponding to each of the multiple archive information retrieval requests.

[0012] A second aspect of the embodiments of the present application provides an archive information retrieval device, the device comprising: A data acquisition module is used to acquire the target image data to be retrieved; A feature extraction module, used to extract features from the target image data to obtain a target feature group, and determine the quality scores of multiple dimensional features included in the target feature group; A weight determination module, used to determine the weights of the multiple dimensional features included in the target feature group according to the quality scores of the multiple dimensional features included in the target feature group; The information retrieval module is used to match the target feature group with the feature groups pre-bound to each piece of archival information, and determine the archival information retrieval result according to the feature matching conditions of each feature group obtained and the weights of multiple dimensional features contained in the target feature group.

[0013] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the archival information retrieval method described in the first aspect are implemented.

[0014] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the archival information retrieval method described in the first aspect are implemented.

[0015] The embodiments of the present application include the following advantages: the archival information is bound to the feature group, so that the archival information characteristics can be described more comprehensively from multiple dimensions with the help of the feature group, which can avoid the influence of the limitations of the archival information characteristics described by the single image feature on the retrieval results, and the weights of the multiple dimensional features are determined according to the quality scores of the multiple dimensional features contained in the target feature group, which is conducive to increasing the influence of the features with higher quality scores (i.e., the features that describe the characteristics of the target image data more accurately) on the feature matching results, thereby improving the accuracy of feature matching. Therefore, the present application can effectively reduce the possibility of false detection and missed detection by converting the comparison link for a single image feature in the archival information retrieval process into a comprehensive matching link for multiple dimensional features in the feature group, thereby improving the accuracy of the archival information retrieval results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application 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 labor.

[0017] Figure 1 It is an implementation flow chart of an archive information retrieval method in an embodiment of the present application; Figure 2 It is a schematic diagram of an implementation process of an archive information retrieval method in an embodiment of the present application; Figure 3 It is a structural schematic diagram of an archive information retrieval device in an embodiment of the present application; Figure 4 It is a schematic diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In application scenarios such as security and traffic management, image data of target objects (such as people or vehicles) with specific characteristics (such as specific image characteristics, specific behavior pattern characteristics, etc.) will be identified and aggregated (i.e., archived) to generate archival information of the target objects, so as to monitor and analyze the target objects or provide other personalized services to users.

[0019] To support the retrieval of archival information, the system will randomly select a picture from the archival information as the image feature bound to the archival information during the initial archiving.

[0020] The user can then search the archive information by image search, that is, by inputting a picture of the target object (such as a person or a car) so that the system can compare the picture with the single image feature bound to each archive information (i.e. 1:N comparison), and then return the archive information with the highest image similarity as the search result to the user.

[0021] For example, after a user uploads a face picture (or face identifier), the system will identify one or more similar face pictures in the face database and determine their credibility. It will then return the profile information bound to the N face pictures with the highest credibility (i.e., similarity) as retrieval results to the user, where N is the preset number of returned information.

[0022] However, since the characteristics of the archival information that can be described by a randomly bound single image feature are limited, for example, the single image feature is usually a picture captured from a certain angle of a part of the target object (such as the back or face of a person), it is easy to make false detections or missed detections when retrieving archival information, which leads to low accuracy of the archival information retrieval results.

[0023] In view of the problems existing in the above-mentioned related technologies, the embodiments of the present application propose a method, device, equipment and medium for archival information retrieval, which converts the comparison link for a single image feature in the archival information retrieval process into a comprehensive matching link for multiple dimensional features in a feature group, and applies the quality scores of the multiple dimensional features to the comprehensive matching link, thereby avoiding the influence of the limitations of the archival information characteristics described by the single image feature on the retrieval results, and increasing the influence of the features with higher quality scores on the feature matching results, that is, increasing the influence of the features that more accurately describe the characteristics of the target image data on the feature matching results, thereby improving the accuracy of feature matching. As a result, the possibility of false detection and missed detection can be effectively reduced, thereby improving the accuracy of archival information retrieval results.

[0024] In conjunction with the accompanying drawings, an archival information retrieval method, apparatus, device and medium provided in the embodiments of the present application are described in detail through some embodiments and their application scenarios.

[0025] First, refer to Figure 1 FIG. 1 is a flowchart of an implementation of an archive information retrieval method provided in an embodiment of the present application, and the method comprises the following steps: Step S11: Acquire the target image data to be retrieved.

[0026] In specific implementation, image data (such as pictures or videos) input by the user can be received through an interface such as a web page.

[0027] The currently received image data is used as the target image data to be retrieved; alternatively, the target image data may be preprocessed first to reduce noise and enhance useful information therein, and then the preprocessed image data is used as the target image data to be retrieved to improve the accuracy of subsequent image processing.

[0028] Step S12: extracting features from the target image data to obtain a target feature group, and determining quality scores of multiple dimensional features included in the target feature group.

[0029] In a specific implementation, the quality scores of the multiple dimensional features included in the target feature group may be determined based on the information integrity and image quality of the target image data.

[0030] Taking the facial feature as an example, if the target image data lacks the facial part or the image of the facial part is blurry, the facial features contained in the target feature group obtained for the target image data will have null values ​​or low accuracy. At this time, the quality score determined for the facial features will also be correspondingly low, and the quality score of the facial features obtained when the facial part is missing will also be lower than the quality score of the facial features obtained when the image of the facial part is blurry.

[0031] Therefore, the present application introduces the quality score to quantify the accuracy of each dimensional feature in describing the characteristics of the target image data (such as facial features), so as to facilitate the subsequent reasonable determination of the feature weight based on the quality score.

[0032] Optionally, to improve processing efficiency, a deep learning model can be pre-trained to perform feature extraction operations and quality score scoring operations based on multiple preset feature dimensions. In this way, by subsequently inputting image data into the deep learning model, the corresponding feature group and the quality scores of the multiple dimensional features contained in the feature group can be quickly obtained; and with the help of the continuous learning ability of the deep learning model, the accuracy of the feature group and quality score output by the model can also be ensured.

[0033] Step S13: Determine the weights of the multiple dimensional features included in the target feature group according to the quality scores of the multiple dimensional features included in the target feature group.

[0034] In a specific implementation, the multiple dimensional features can be sorted in order of quality scores from high to low, and weights can be set from high to low for the sorted multiple dimensional features, that is, the features with higher quality scores have a larger proportion of weight in the weights of multiple dimensions, so as to increase the influence of features with higher quality scores (i.e., features that more accurately describe the characteristics of the target image data) on the feature matching results (such as the comprehensive matching score).

[0035] Alternatively, the quality score and other factors (such as important features additionally specified by the user) may be comprehensively considered to determine the weights of the multiple dimensional features included in the target feature group.

[0036] Step S14: Match the target feature group with the feature groups pre-bound to each piece of archival information respectively, and determine the archival information retrieval result according to the feature matching conditions of each feature group obtained and the weights of multiple dimensional features included in the target feature group.

[0037] In a specific implementation, for each piece of archival information, the feature group pre-bound to the archival information is matched with the target feature group. For example, based on algorithms such as feature distance measurement, the similarity between the feature group and each dimensional feature in the target feature group is calculated to obtain the feature matching status of the feature group (i.e., the similarity score corresponding to each dimensional feature in the feature group).

[0038] Alternatively, it is also possible to determine whether the match is successful based on the similarity score and a preset score threshold, thereby obtaining the feature matching status of the feature group (ie, a matching status characterization value, such as 1 for successful matching and 0 for failed matching).

[0039] After the feature matching status of each feature group is obtained, the comprehensive matching score of each feature group is determined according to the feature matching status of each feature group and the weights of multiple dimensional features included in the target feature group.

[0040] For example, for each feature group, the similarity score corresponding to each dimensional feature in the feature group and the weight of the corresponding dimensional feature in the target feature group are weighted and calculated to obtain a comprehensive matching score of the feature group.

[0041] For example, the feature group includes three dimensional features, namely, face features, body posture features and gait features. The face features, body posture features and gait features included in the feature group are matched with the face features, body posture features and gait features included in the target feature group (i.e., feature similarity calculation is performed), and the similarity scores C1, C2 and C3 corresponding to the face features, body posture features and gait features in the feature group are obtained.

[0042] Then, the similarity scores C1, C2 and C3 corresponding to the facial features, human posture features and gait features in the feature group are weighted and calculated with the weights W1, W2 and W3 of the facial features, human posture features and gait features in the target feature group to obtain the comprehensive matching score of the feature group (i.e., C1*W1+C2*W2+C3*W3).

[0043] After obtaining the comprehensive matching scores of the feature groups, the feature groups are sorted in descending order of the comprehensive matching scores, and the top N feature groups are selected from the sorted feature groups, and finally the archive information pre-bound to the top N feature groups is used as the archive information retrieval result; wherein N is the preset number of returned information, which can be 1, 10 or 100, etc. (corresponding to the requirements of the search result range such as top1, top10 or top100).

[0044] By adopting the technical solution of the embodiment of the present application, the archival information is bound to the feature group, so that the archival information characteristics can be described more comprehensively from multiple dimensions with the help of the feature group, which can avoid the influence of the limitations of the archival information characteristics described by a single image feature on the retrieval results, and according to the quality scores of the multiple dimensional features contained in the target feature group, the weights of the multiple dimensional features are determined, which is conducive to increasing the influence of features with higher quality scores (i.e., features that describe the characteristics of the target image data more accurately) on the feature matching results, thereby improving the accuracy of feature matching. Therefore, the present application can effectively reduce the possibility of false detection and missed detection by converting the comparison link for a single image feature in the archival information retrieval process into a comprehensive matching link for multiple dimensional features in the feature group, thereby improving the accuracy of the archival information retrieval results.

[0045] As a possible implementation, the archive information retrieval result is determined according to the feature matching conditions of each feature group and the weights of multiple dimensional features included in the target feature group, including: Step S141: determining first matching scores corresponding to each of the feature groups according to feature matching conditions of the feature groups and weights of multiple dimensional features included in the target feature group.

[0046] In a specific implementation, the first matching score of each feature group can be determined by calculating a weighted sum according to the feature matching conditions (such as similarity scores or matching condition representation values) of each feature group and the weights of multiple dimensional features included in the target feature group.

[0047] For example, for each feature group, a weighted sum is calculated for the matching situation representation value corresponding to each dimensional feature in the feature group and the weight of the corresponding dimensional feature in the target feature group to obtain a first matching score for the feature group.

[0048] Step S142: When the number of feature groups that meet the first screening condition exceeds a preset number, each feature group that meets the first screening condition is used as a candidate feature group, and the weights of the multiple dimensional features included in the target feature group are adjusted to increase the weight ratio of the target dimensional feature in the multiple dimensional features.

[0049] Among them, the first filtering condition is used to filter the feature group according to the first matching score. The first filtering condition can be determined according to actual needs. For example, the first filtering condition can be: in the order of the first matching scores from high to low, select the feature groups corresponding to the first N first matching scores, and N is equal to the preset number.

[0050] In a specific implementation, when the number of feature groups meeting the first screening condition meets a preset number, the pre-bound archive information of each feature group meeting the first screening condition can be directly used as the archive information retrieval result.

[0051] Taking into account the situation that different feature groups may have the same corresponding first matching scores (this may be caused by the low quality scores of some dimensional features in the target feature group involved in the matching or the feature group pre-bound to the archive information), and thus the number of feature groups that meet the first screening condition exceeds the preset number, that is, exceeds the preset number of returned information, at this time it can be regarded that the amount of information contained in the obtained search results (that is, the range of the search results) does not meet the user's needs).

[0052] At this time, the weights of the multiple dimensional features included in the target feature group can be adjusted to increase the weight ratio of the target dimensional features in the multiple dimensional features. Thus, two sets of feature weights focusing on different dimensional features (i.e., the weights of the multiple dimensional features before and after adjustment) can be obtained to facilitate subsequent optimization of the search result range.

[0053] Step S143: determining a second matching score corresponding to each of the candidate feature groups according to the feature matching conditions of the respective candidate feature groups and the adjusted weights of the multiple dimensional features.

[0054] In a specific implementation, the second matching score of each candidate feature group can be obtained by calculating a weighted sum according to the feature matching conditions (such as similarity scores or matching condition representation values) of each candidate feature group and the weights of multiple dimensional features included in the adjusted target feature group.

[0055] Step S144: determining the score difference between the first matching score and the second matching score corresponding to each of the candidate feature groups.

[0056] In specific implementation, by determining the score difference between the first matching score and the second matching score corresponding to each of the candidate feature groups, the degree of difference in the comprehensive matching situations obtained when focusing on feature matching of different dimensional features can be determined, and the search result range can be subsequently optimized based on the degree of difference.

[0057] Step S145: Determine the archive information retrieval result according to the score differences corresponding to each of the candidate feature groups.

[0058] In a specific implementation, the candidate feature groups may be sorted in descending order of the score difference, and the first N (i.e., the preset number) candidate feature groups may be selected, and then the archive information pre-bound to the selected N candidate feature groups may be used as the archive information retrieval result.

[0059] It can be understood that if the score difference corresponding to the candidate feature group is small, it means that when focusing on different dimensional features for feature matching, the degree of difference in the comprehensive matching situation obtained for the candidate feature group is small, which means that the matching degree of the candidate feature group when focusing on different dimensional features for feature matching is stably maintained at a high level.

[0060] Therefore, the present application optimizes the scope of search results based on the score difference, and can ensure accurate search capabilities for archival information when the first matching scores corresponding to different feature groups are the same.

[0061] Optionally, determining the archive information retrieval result according to the score differences corresponding to each of the candidate feature groups includes: In the case that the number of candidate feature groups that meet the second screening condition does not meet the preset number, repeatedly increasing the weight ratio of the new target dimensional features in the multiple dimensional features to re-determine the second matching scores corresponding to each of the candidate feature groups, and re-determine the score difference corresponding to each of the candidate feature groups based on each of the second matching scores obtained last time and each of the second matching scores obtained currently, until the number of candidate feature groups that meet the second screening condition meets the preset number, the repetitive execution of the steps is terminated, wherein the second screening condition is used to screen feature groups according to the score difference and the preset score difference threshold; The archive information pre-bound to each candidate feature group that meets the second screening condition is determined as the archive information retrieval result.

[0062] In specific implementation, a preset score difference threshold (such as one percent) is introduced to control the accuracy and range of the search results.

[0063] Specifically, based on the score differences corresponding to the respective candidate feature groups currently obtained, it is determined whether the number of candidate feature groups that meet the second screening condition meets the preset number.

[0064] If the number of candidate feature groups meeting the second screening condition meets the preset number, the archive information pre-bound to each candidate feature group meeting the second screening condition is determined as the archive information retrieval result.

[0065] If the number of candidate feature groups that meet the second screening condition does not meet the preset number, for example, the number of candidate feature groups whose corresponding score differences do not exceed the preset score difference threshold is too small, so that the number of candidate feature groups that meet the second screening condition is lower than the preset number; or, there are candidate feature groups with the same score difference among the candidate feature groups that do not exceed the preset score difference threshold, so that the number of candidate feature groups that meet the second screening condition exceeds the preset number; at this time, continue to select new target dimensional features, and increase the weight ratio of the new target dimensional features to perform feature matching focusing on different dimensional features again.

[0066] Thus, by repeatedly executing the aforementioned feature matching steps focusing on features of different dimensions, the matching results are iteratively optimized until, based on the currently determined score difference, it is determined that the number of candidate feature groups currently meeting the second screening condition meets the preset number, and then the aforementioned steps are repeated. In this way, archive information retrieval results that meet the retrieval result range set by the user (i.e., the preset number) and have a high matching degree can be returned to the user.

[0067] Optionally, the target dimension feature is determined by the following steps: Receiving indication information of target dimension features from a user; Alternatively, the target dimension feature is determined based on a preset list, where the preset list includes: multiple feature dimensions and their respective corresponding priorities.

[0068] In specific implementations, when the system needs to perform feature matching that focuses on features of different dimensions, it can ask the user about the target dimension features that currently need to have their weight ratio increased, and receive indication information about the target dimension features from the user. For example, if the user is actually more concerned about the facial feature of wearing glasses (or the gait feature when walking), the system can indicate the corresponding dimension features. The system can then increase the weight ratio of the features indicated by the user compared to other dimensional features (such as human posture, etc.) to achieve more accurate retrieval that meets user needs.

[0069] Alternatively, it is also possible to set priorities according to the importance of multiple preset feature dimensions associated with the search scenario (such as the search for vehicle-related archival information, the search for person-related archival information, etc.), and establish a preset list to record the priority of each preset feature dimension. Thus, when the system needs to perform feature matching focusing on different dimensional features, it can automatically determine the target dimensional features that need to increase the weight ratio according to the preset list, thereby achieving more accurate retrieval that is more in line with the characteristics of the scenario.

[0070] As a possible implementation, the method further includes: For the currently collected image data, feature extraction is performed on the image data to obtain a feature group corresponding to the image data, wherein the feature group corresponding to the image data includes: features with the highest quality scores extracted from multiple feature dimensions respectively; Determine whether there is file information corresponding to the image data in the database; If the archive information corresponding to the image data does not exist in the database, generating the archive information according to the image data, binding the generated archive information with the feature group corresponding to the image data and storing them in the database; When archival information corresponding to the image data exists in a database, the image data is used to update the archival information, and a quality score comparison is performed between a plurality of dimensional features included in a feature group corresponding to the image data and a plurality of dimensional features included in a feature group pre-bound to the archival information corresponding to the image data. The plurality of dimensional features with the highest quality scores obtained after comparison are used to update the feature group pre-bound to the archival information corresponding to the image data.

[0071] In a specific implementation, at the aggregation stage, for the currently collected image data (such as the currently captured picture or video), firstly, feature extraction is performed on the image data to determine the feature group corresponding to the image data.

[0072] Among them, when multiple features are extracted from a feature dimension, the feature with the highest quality score among the multiple features extracted from the feature dimension can be used as the optimal feature for constructing a feature group, thereby ensuring that the obtained feature group can contain the features with the highest quality scores extracted from multiple feature dimensions respectively.

[0073] Then, it is determined whether there is archival information corresponding to the image data in the database. If there is no archival information corresponding to the image data in the database, a filing operation is performed on the image data to generate archival information containing the image data.

[0074] Then, the generated archive information is bound to the feature group corresponding to the image data and stored in the database (for example, both are stored in the database in an associated storage manner) to support multi-dimensional feature comprehensive matching in the subsequent image search stage.

[0075] If the archive information corresponding to the image data exists in the database, the image data may be added to the archive information to update the archive information.

[0076] At the same time, a quality score comparison is performed on each feature dimension of the feature group pre-bound to the archival information and the feature group corresponding to the image data, and then new features with higher quality scores in the feature group corresponding to the image data are used to replace old features with lower quality scores in the feature group pre-bound to the archival information.

[0077] Therefore, with the continuous capture and update of the archival information, the feature group bound to the archival information can be gradually optimized, thereby ensuring the timeliness and accuracy of the feature group bound to the archival information.

[0078] In this embodiment, the traditional single image feature is expanded to a feature group containing multi-dimensional features (including but not limited to multi-dimensional features such as face, human posture, gait, etc.), and in the archiving stage, it is divided into the feature group with the best archival information binding according to the quality and the feature group is progressively optimized, so as to ensure the timeliness and accuracy of the feature group bound to the archival information, which is beneficial to improve the accuracy of archival information retrieval.

[0079] As a possible implementation manner, the feature groups pre-bound to each of the archive information are obtained from a cache; and the method further includes: Recording the binding relationship between different archive information and different feature groups, and obtaining the access frequency of the different feature groups; The recorded binding relationship and the hotspot feature group whose access frequency exceeds a preset frequency threshold among the different feature groups are stored in the cache.

[0080] In specific implementation, a mapping table may be established to associate and store identifiers of different archive information and identifiers of feature groups to which they are bound, and the mapping table may be stored in the cache to implement cache storage of the recorded binding relationships.

[0081] At the same time, hot feature groups that are frequently accessed are stored in the cache. For example, if archive information associated with a certain geographical area is frequently retrieved, the feature group pre-bound to the archive information associated with the geographical area can be stored in the cache as a hot feature group.

[0082] In this embodiment, the recorded binding relationship and characteristic feature group are stored in a cache layer so that the aforementioned information can be quickly accessed when performing a 1VN search, thereby improving the search speed.

[0083] As a possible implementation, the method further includes: Merging multiple archive information retrieval requests received within a set time period into a batch request, where different archive information retrieval requests carry different target image data to be retrieved; The batch requests are processed in parallel using multiple threads to obtain and return the archive information retrieval results corresponding to each of the multiple archive information retrieval requests.

[0084] In the specific implementation, multiple 1VN requests (i.e., archive information retrieval requests) received from various users within a set time period are merged into a batch request for unified processing to reduce the number of input / output (I / O) operations; and the batch requests are processed in parallel using multiple threads to shorten the processing time, thereby improving the retrieval speed.

[0085] For example, refer to Figure 2 The schematic diagram of the implementation process of the archive information retrieval method shown in FIG. 1 mainly includes the following steps: Step 1: Bind a feature group to each profile information.

[0086] In this step, different from the traditional clustering operation, the present application will bind a feature group to the current file (ie, the currently generated file information) when archiving (ie, initial clustering), instead of just binding a single image feature.

[0087] It can be understood that compared with the single image feature, the present application can achieve a more detailed and comprehensive description of the characteristics of the archival information (that is, the target object) from different dimensions (such as facial features, body features, and different characteristics, etc.) with the help of feature groups.

[0088] Step 2: Dynamically adjust the feature group bound to the archive information.

[0089] In this step, considering that the optimal feature group may not be obtained during the initial aggregation, that is, it is difficult to obtain relatively accurate feature data for all preset feature dimensions, the present application will perform bypass feature group detection updates during the later real-time aggregation, that is, continuously obtain new features with higher quality scores for each preset feature dimension to replace the old features with lower quality scores in the feature group.

[0090] Therefore, this application continuously improves the feature fullness and accuracy of the feature group bound to the archival information based on image data such as pictures or videos captured by different acquisition devices at different locations and times, so as to gradually optimize the feature group, thereby providing strong support for archival retrieval optimization.

[0091] Step 3: Continuously adjust the matching results to gradually improve the accuracy of the retrieval.

[0092] In this step, in the retrieval stage of searching for files with images, the feature group of the current image data (i.e., the target feature group) is first obtained, and reasonable feature weights are assigned according to the quality scores of each dimensional feature in the target feature group. For example, if the face is not clear enough, the weights of the dimensional features related to the human body can be appropriately increased.

[0093] Execute the 1VN search operation, that is, obtain the feature group pre-bound to each file information and perform feature matching with the target feature group, and then calculate the comprehensive matching score (i.e., the first matching score) of each feature group according to the set feature weight, and cache each first matching score.

[0094] Then, based on the first screening condition, it is determined whether to perform optimization of the matching result. If it is determined to perform optimization of the matching result, the weight ratio of the target feature is increased to perform a new calculation on the previous matching result to obtain a second matching score.

[0095] Repeat the above steps until it is determined that the optimization of the matching results can be terminated according to the second screening condition. For example, when the highest score difference between the two matches does not exceed the preset score difference threshold (such as 1%), it can be considered that the best matching result has been obtained, and then it is determined that the optimization of the matching results can be terminated. Finally, according to the score difference of each matching result (i.e., matching score), the corresponding archive information retrieval results such as top10 and top100 are determined.

[0096] Optionally, since the aforementioned 1VN search operation occupies the main time of searching for files with images, the hotspot feature group can be pre-added to the cache to increase the search speed.

[0097] Optionally, during retrieval, multiple 1VN requests can be merged into one batch request to reduce the number of I / O operations, and multi-threaded parallel processing can be used to further shorten the processing time, thereby effectively improving the retrieval speed.

[0098] Based on the above embodiments and examples, the archival information retrieval solution provided by this application can achieve at least the following optimization effects: 1. Enhanced search accuracy: Based on the feature group binding and weight distribution mechanism, the retrieval results are more accurate and the possibility of false detection and missed detection is reduced; and the dynamic optimization mechanism based on feature groups can also ensure the timeliness and accuracy of the description of the characteristics of archival information, thereby further improving the retrieval accuracy.

[0099] 2.Flexible adaptability: The system can flexibly adjust feature groups and weight settings according to different application scenarios, is suitable for a variety of business needs, and can improve the scalability and adaptability of the system.

[0100] 3. Improve search speed: Through caching technology and batch processing, the frequency of access to the database can be reduced, greatly shortening the time required for each search. For example, in a big data environment, it is often necessary to retrieve a large amount of archive information from multiple databases. At this time, using a caching strategy can significantly reduce the system load and improve the overall response speed.

[0101] 4. Optimize user experience: This application can greatly improve the efficiency and accuracy of image-based file search, thereby providing users with faster and more accurate retrieval services. The faster retrieval speed and higher retrieval accuracy can significantly improve the user experience when using the image-based file search function. For example, it can significantly reduce invalid searches and user waiting time, thereby enhancing the practicality of the system and user satisfaction.

[0102] For the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the order of the actions described, because according to the embodiments of the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0103] Second, Figure 3: is a schematic diagram of the structure of an archive information retrieval device according to an embodiment of the present application, the device comprising: The data acquisition module 310 is used to acquire the target image data to be retrieved; A feature extraction module 320 is used to extract features from the target image data to obtain a target feature group, and determine quality scores of multiple dimensional features included in the target feature group; A weight determination module 330, configured to determine the weights of the multiple dimensional features included in the target feature group according to the quality scores of the multiple dimensional features included in the target feature group; The information retrieval module 340 is used to match the target feature group with the feature groups pre-bound to each piece of archival information, and determine the archival information retrieval result according to the feature matching conditions of each feature group and the weights of multiple dimensional features included in the target feature group.

[0104] By adopting the technical solution of the embodiment of the present application, the archival information is bound to the feature group, so that the archival information characteristics can be described more comprehensively from multiple dimensions with the help of the feature group, which can avoid the influence of the limitations of the archival information characteristics described by a single image feature on the retrieval results, and according to the quality scores of the multiple dimensional features contained in the target feature group, the weights of the multiple dimensional features are determined, which is conducive to increasing the influence of features with higher quality scores (i.e., features that describe the characteristics of the target image data more accurately) on the feature matching results, thereby improving the accuracy of feature matching. Therefore, the present application can effectively reduce the possibility of false detection and missed detection by converting the comparison link for a single image feature in the archival information retrieval process into a comprehensive matching link for multiple dimensional features in the feature group, thereby improving the accuracy of the archival information retrieval results.

[0105] Optionally, the information retrieval module 340 includes: A first retrieval module, configured to determine a first matching score corresponding to each of the feature groups according to feature matching conditions of the feature groups and weights of multiple dimensional features included in the target feature group; a second retrieval module, configured to, when the number of feature groups meeting the first screening condition exceeds a preset number, use the feature groups meeting the first screening condition as candidate feature groups, and adjust the weights of the multiple dimensional features included in the target feature group to increase the weight ratio of the target dimensional feature in the multiple dimensional features, wherein the first screening condition is used to screen the feature group according to the first matching score; A third retrieval module is used to determine a second matching score corresponding to each of the candidate feature groups according to the feature matching conditions of each of the candidate feature groups and the adjusted weights of the multiple dimensional features; A fourth retrieval module, used to determine the score difference between the first matching score and the second matching score corresponding to each of the candidate feature groups; The fifth retrieval module is used to determine the archive information retrieval result according to the score difference corresponding to each of the candidate feature groups.

[0106] Optionally, the fifth retrieval module includes: A first retrieval submodule is used for repeatedly performing the step of increasing the weight ratio of the new target dimensional features in the multiple dimensional features to re-determine the second matching scores corresponding to each of the candidate feature groups, and re-determining the score difference corresponding to each of the candidate feature groups according to each of the second matching scores obtained last time and each of the second matching scores obtained currently, when the number of the candidate feature groups meeting the second screening condition does not meet the preset number, and terminating the repetitive execution of the steps, wherein the second screening condition is used to screen feature groups according to the score difference and the preset score difference threshold; The second search submodule is used to determine the archive information pre-bound to each candidate feature group that meets the second screening condition as the archive information search result.

[0107] Optionally, the device further comprises a feature determination module, configured to perform the following steps: Receiving indication information of target dimension features from a user; Alternatively, the target dimension feature is determined based on a preset list, where the preset list includes: multiple feature dimensions and their respective corresponding priorities.

[0108] Optionally, the device further comprises: A first processing module is used to perform feature extraction on the currently collected image data to obtain a feature group corresponding to the image data, wherein the feature group corresponding to the image data includes: features with the highest quality scores extracted from multiple feature dimensions; A second processing module, used for determining whether there is file information corresponding to the image data in the database; A third processing module, configured to generate archival information according to the image data if the archival information corresponding to the image data does not exist in the database, bind the generated archival information with the feature group corresponding to the image data, and store the generated archival information in the database; The fourth processing module is used to update the archival information using the image data when archival information corresponding to the image data exists in the database, and compare the quality scores of multiple dimensional features included in the feature group corresponding to the image data with the multiple dimensional features included in the feature group pre-bound to the archival information corresponding to the image data, and use the multiple dimensional features with the highest quality scores obtained after comparison to update the feature group pre-bound to the archival information corresponding to the image data.

[0109] Optionally, the feature groups pre-bound to each of the archive information are obtained from a cache; and the device further comprises: A first storage module is used to record the binding relationship between different archive information and different feature groups, and obtain the access frequency of the different feature groups; The second storage module is used to store the recorded binding relationship and the hot spot feature group whose access frequency exceeds a preset frequency threshold in the different feature groups into the cache.

[0110] Optionally, the device further comprises: A request merging module, used to merge multiple archive information retrieval requests received within a set time period into a batch request, where different archive information retrieval requests carry different target image data to be retrieved; The parallel processing module is used to use multiple threads to process the batch requests in parallel, obtain the archive information retrieval results corresponding to each of the multiple archive information retrieval requests, and return them.

[0111] It should be noted that the device embodiment is similar to the method embodiment, so the description is relatively simple, and the relevant parts can be referred to the method embodiment.

[0112] The present application also provides an electronic device, referring to Figure 4 , Figure 4 Schematic diagram of an electronic device proposed in an embodiment of the present application. Figure 4 As shown, the electronic device 100 includes: a memory 110 and a processor 120. The memory 110 and the processor 120 are connected via a bus communication. The memory 110 stores a computer program, which can be run on the processor 120 to implement the steps in the archival information retrieval method disclosed in the embodiment of the present application.

[0113] The embodiment of the present application also provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the archival information retrieval method disclosed in the embodiment of the present application is implemented.

[0114] The embodiment of the present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the archival information retrieval method disclosed in the embodiment of the present application.

[0115] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0116] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, systems, devices, storage media, and program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0120] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0121] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0122] The above is a detailed introduction to the archival information retrieval method, device, equipment and medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for retrieving archive information, characterized in that: The method comprises: Get the target image data to be retrieved; Performing feature extraction on the target image data to obtain a target feature group, and determining quality scores of multiple dimensional features included in the target feature group; Determining weights of the multiple dimensional features included in the target feature group according to the quality scores of the multiple dimensional features included in the target feature group; The target feature group is matched with the feature groups pre-bound to each piece of archive information respectively, and the archive information retrieval result is determined according to the feature matching conditions of each feature group obtained and the weights of multiple dimensional features included in the target feature group.

2. The method according to claim 1, characterized in that According to the feature matching conditions of each feature group obtained and the weights of multiple dimensional features contained in the target feature group, the archive information retrieval result is determined, including: Determine a first matching score corresponding to each of the feature groups according to feature matching conditions of the feature groups and weights of multiple dimensional features included in the target feature group; When the number of feature groups meeting the first screening condition exceeds a preset number, each feature group meeting the first screening condition is used as a candidate feature group, and the weights of the multiple dimensional features included in the target feature group are adjusted to increase the weight ratio of the target dimensional features in the multiple dimensional features, wherein the first screening condition is used to screen the feature group according to the first matching score; Determining second matching scores corresponding to each of the candidate feature groups according to feature matching conditions of each of the candidate feature groups and the adjusted weights of the multiple dimensional features; Determine a score difference between a first matching score and a second matching score corresponding to each of the candidate feature groups; The archive information retrieval result is determined according to the score differences corresponding to each of the candidate feature groups.

3. The method according to claim 2, characterized in that Determining the archive information retrieval result according to the score differences corresponding to each of the candidate feature groups includes: In the case that the number of candidate feature groups that meet the second screening condition does not meet the preset number, repeatedly increasing the weight ratio of the new target dimensional features in the multiple dimensional features to re-determine the second matching scores corresponding to each of the candidate feature groups, and re-determine the score difference corresponding to each of the candidate feature groups based on each of the second matching scores obtained last time and each of the second matching scores obtained currently, until the number of candidate feature groups that meet the second screening condition meets the preset number, the repetitive execution of the steps is terminated, wherein the second screening condition is used to screen feature groups according to the score difference and the preset score difference threshold; The archive information pre-bound to each candidate feature group that meets the second screening condition is determined as the archive information retrieval result.

4. The method according to claim 2, characterized in that: The target dimension feature is determined by the following steps: Receiving indication information of target dimension features from a user; Alternatively, the target dimension feature is determined based on a preset list, where the preset list includes: multiple feature dimensions and their respective corresponding priorities.

5. The method according to claim 1, characterized in that The method further comprises: For the currently collected image data, feature extraction is performed on the image data to obtain a feature group corresponding to the image data, wherein the feature group corresponding to the image data includes: features with the highest quality scores extracted from multiple feature dimensions respectively; Determine whether there is file information corresponding to the image data in the database; If the archive information corresponding to the image data does not exist in the database, generating the archive information according to the image data, binding the generated archive information with the feature group corresponding to the image data and storing them in the database; When archival information corresponding to the image data exists in a database, the image data is used to update the archival information, and a quality score comparison is performed between a plurality of dimensional features included in a feature group corresponding to the image data and a plurality of dimensional features included in a feature group pre-bound to the archival information corresponding to the image data. The plurality of dimensional features with the highest quality scores obtained after comparison are used to update the feature group pre-bound to the archival information corresponding to the image data.

6. The method according to any one of claims 1 to 5, characterized in that: The feature groups pre-bound to each of the archive information are obtained from the cache; the method further includes: Recording the binding relationship between different archive information and different feature groups, and obtaining the access frequency of the different feature groups; The recorded binding relationship and the hotspot feature group whose access frequency exceeds a preset frequency threshold among the different feature groups are stored in the cache.

7. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Merging multiple archive information retrieval requests received within a set time period into a batch request, where different archive information retrieval requests carry different target image data to be retrieved; The batch requests are processed in parallel using multiple threads to obtain and return the archive information retrieval results corresponding to each of the multiple archive information retrieval requests.

8. An archive information retrieval device, characterized in that: The device comprises: A data acquisition module is used to acquire the target image data to be retrieved; A feature extraction module, used to extract features from the target image data to obtain a target feature group, and determine the quality scores of multiple dimensional features included in the target feature group; A weight determination module, used to determine the weights of the multiple dimensional features included in the target feature group according to the quality scores of the multiple dimensional features included in the target feature group; The information retrieval module is used to match the target feature group with the feature groups pre-bound to each piece of archival information, and determine the archival information retrieval result according to the feature matching conditions of each feature group obtained and the weights of multiple dimensional features contained in the target feature group.

9. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the archival information retrieval method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the archival information retrieval method as described in any one of claims 1 to 7 is implemented.

Images