A method, apparatus, electronic device and storage medium for assessing the quality of archives.

Abstract

This invention provides a method, apparatus, electronic device, and storage medium for assessing the quality of archives. The method includes: acquiring a target archive, which includes a set of target face images and a set of target human body images associated with the target face images; calculating the similarity of human body features in the set of target human body images and calculating the similarity of facial features in the set of target human body images; if both the average similarity of human body features and the average similarity of facial features meet a preset similarity threshold, then the archive quality level of the target archive is determined to be a first archive quality level; if neither the average similarity of human body features nor the average similarity of facial features meets the preset similarity threshold, then the trajectory score and dwell time of the target archive are calculated respectively, and the archive quality level of the target archive is determined based on the trajectory score and dwell time. This invention can improve the accuracy of archive quality assessment.

Description

Technical Field

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

[0002] Traditional methods only consider facial and body quality, neglecting the concept of archival quality. This makes it impossible to measure the quality of archival filing results, hindering archival optimization and assessment. During the archival process, the accuracy of facial or body image filing directly determines archival quality. For example, existing image filing methods typically extract facial features from captured images, calculate the similarity between two images, and determine whether to group them into the same archive based on the similarity score. When the shooting angle changes or facial features are blurred, the calculated similarity score becomes less reliable, resulting in poor image filing. Similarly, human images also need to be archived. Without facial features, the accuracy of human image filing is also low, leading to low archival quality. Furthermore, existing archival quality assessment methods also have low accuracy. Summary of the Invention

[0003] This invention provides a method for assessing the quality of archives, aiming to address the problem that existing methods for assessing the quality of archives also have low accuracy.

[0004] In a first aspect, embodiments of the present invention provide a method for assessing the quality of archives, the method comprising:

[0005] Obtain a target file, the target file including a set of target face images and a set of target human body images associated with the set of target face images;

[0006] Calculate the similarity of human features between any two different human images in the target human image set, and calculate the similarity of facial features between any two different face images in the target human image set;

[0007] If the average similarity of the human body features and the average similarity of the face features both meet the preset similarity threshold condition, then the file quality level of the target file is determined to be the first file quality level.

[0008] If the average similarity of the human body features and the average similarity of the facial features do not meet the preset similarity threshold, then the trajectory score and dwell time of the target file are calculated respectively, and the file quality level of the target file is determined based on the trajectory score and dwell time of the target file.

[0009] Optionally, calculating the similarity of human features between any two different human images in the target human image set, and calculating the similarity of facial features between any two different face images in the target human image set, includes:

[0010] Obtain the target human body image from the target human body image set and the target face image from the target face image set, wherein the target human body image is the human body image with the top N image quality in the target human body image set, and the target face image is the face image with the top N image quality in the target face image set;

[0011] Calculate the similarity of human features between any two different target human body images and the similarity of facial features between any two different target face images in the target human body image.

[0012] Optionally, calculating the trajectory score and dwell time of the target file respectively includes:

[0013] Based on the spatiotemporal information of the target human images in the target human image set and the target face images in the target face image set, the target face images in the target file and the target human images in the target human image set are sorted to obtain the image sequence of the target file.

[0014] Obtain the number of valid images and the total number of images in the image sequence of the target file, and calculate the trajectory score of the target file based on the number of valid images and the total number of images;

[0015] The image time of the first image and the image time of the last image in the image sequence of the target file are obtained, and the dwell time of the target file is calculated based on the image time of the first image and the image time of the last image.

[0016] Optionally, obtaining the number of valid images in the image sequence of the target file includes:

[0017] Determine whether the next image in the image sequence of the target file is within the second-order adjacent range of the previous image;

[0018] If the next image in the image sequence of the target file is within the second-order adjacent range of the previous image, then the number of valid images is incremented by 1.

[0019] Optionally, determining the file quality level of the target file based on the trajectory score and dwell time of the target file includes:

[0020] If the trajectory score is lower than the first trajectory score threshold and / or the dwell time is lower than the first dwell time threshold, then the file quality level of the target file is determined to be the second file quality level.

[0021] If the trajectory score is higher than the first trajectory score threshold and the dwell time is higher than the first dwell time threshold, then the file quality level of the target file is determined based on the trajectory score, the second trajectory score threshold, the dwell time, and the second dwell time range.

[0022] Optionally, determining the file quality level of the target file based on the trajectory score, the second trajectory score threshold, the dwell time, and the second dwell time range includes:

[0023] If the trajectory score is lower than the second trajectory score threshold, the archive quality level of the target archive is determined to be the third archive quality level;

[0024] If the trajectory score is higher than the second trajectory score threshold, then it is determined whether the dwell time is within the range of the second dwell time.

[0025] If the trajectory score is not within the second dwell time range, then the file quality level of the target file is determined to be the third file quality level;

[0026] If the trajectory score falls within the second dwell time range, then the target file's file quality level is determined to be the fourth file quality level.

[0027] Optionally, calculating the similarity of human features between any two different human images in the target human image set further includes:

[0028] Count the number of target human body images in the target file;

[0029] If there are multiple sets of target human images in the target file, then the similarity of human features for each set of target human images is calculated separately.

[0030] Secondly, embodiments of the present invention also provide an archival quality assessment device, the archival quality assessment device comprising:

[0031] The acquisition module is used to acquire a target file, which includes a set of target face images and a set of target human body images associated with the set of target face images;

[0032] The calculation module is used to calculate the similarity of human features between any two different human images in the target human image set, and to calculate the similarity of facial features between any two different face images in the target human image set.

[0033] The first determining module is used to determine the file quality level of the target file as the first file quality level if the average similarity of the human body feature similarity and the average similarity of the face feature similarity both meet the preset similarity threshold condition.

[0034] The second determining module is used to calculate the trajectory score and dwell time of the target file respectively if the average similarity of the human body feature similarity and the average similarity of the face feature similarity do not meet the preset similarity threshold condition, and determine the file quality level of the target file based on the trajectory score and dwell time of the target file.

[0035] Thirdly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the archive quality assessment method provided in embodiments of the present invention.

[0036] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the archival quality assessment method provided in the embodiments of the present invention.

[0037] In this embodiment of the invention, a target file is acquired, comprising a set of target face images and a set of target human body images associated with the target face image set. The similarity of human body features between any two different human body images in the target human body image set is calculated, as is the similarity of facial features between any two different face images in the target human body image set. If both the average similarity of the human body features and the average similarity of the facial features satisfy a preset similarity threshold, the file quality level of the target file is determined to be a first file quality level. If neither the average similarity of the human body features nor the average similarity of the facial features satisfies the preset similarity threshold, the trajectory score and dwell time of the target file are calculated respectively, and the file quality level of the target file is determined based on the trajectory score and dwell time. Thus, after determining the file quality level of the target file based on the average facial similarity and the average human body similarity, the file quality level is then re-determined based on the trajectory score and dwell time, achieving a dual determination of the file quality level and thus accurately rating the file quality level of the target file. This provides direction for archive optimization, allowing for targeted optimization of low-rated archives. It also enables the evaluation of overall archiving results, facilitating the measurement of the effectiveness of algorithm iterations and their improvement. This further enhances the accuracy of archive quality assessment. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of an archival quality assessment method provided in an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the structure of an archive quality assessment device provided in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] like Figure 1 As shown, Figure 1 This is a flowchart of a method for assessing the quality of archives provided in an embodiment of the present invention. The method for assessing the quality of archives includes the following steps:

[0044] Step 101: Obtain the target file, which includes a set of target face images and a set of target human body images associated with the set of target face images.

[0045] The target archives mentioned above are those that require archival quality assessment. There can be multiple target archives. These target archives can exist in an archival aggregation system or be stored in a pre-configured database for archival storage.

[0046] The aforementioned target face image set is a collection of face images of the corresponding individuals in the target file. The target face image set includes multiple face images, where each face image can correspond to a trajectory, which can be formed by the location of the face image capturing device.

[0047] The aforementioned target human image set is a collection of human images of the corresponding individuals in the target file. The target human image set includes multiple human images, and each human image can also correspond to a trajectory. This trajectory can be formed by the position of the human image capturing device. It should be noted that when an image includes both a face and a body, the face image and the body image correspond to the same trajectory. The target human image set can be one or multiple.

[0048] Specifically, the target file can be obtained from the archive aggregation system or database.

[0049] Step 102: Calculate the similarity of human features between any two different human images in the target human image set, and calculate the similarity of facial features between any two different face images in the target human image set.

[0050] The aforementioned human body feature similarity refers to the similarity of human body features between different human body images in the target human body set. The aforementioned facial feature similarity refers to the similarity of facial features between different facial images in the target facial face set.

[0051] In this embodiment of the invention, step 102 includes:

[0052] Obtain target human images from the target human image set and target face images from the target face image set. The target human images are the top N human images in the target human image set, and the target face images are the top N face images in the target face image set. Calculate the similarity of human features between any two different target human images and the similarity of face features between any two different target face images.

[0053] More specifically, to improve the efficiency of similarity calculation and reduce the computational load, the top N human images of quality can be selected from the target human image set to obtain the Top N quality human images. Similarly, the top N face images of quality can be selected from the target face image set to obtain the Top N quality face images. Then, the human feature similarity between the Top N quality human images is calculated, followed by the average human similarity of the Top N images. Finally, the average face similarity of the Top N face images is calculated. The accuracy is measured by the average similarity of the Top N quality face or human images, which can identify multiple individuals within the same category with a high probability. This improves the computational efficiency of average human and face similarity and reduces the computational load, while also improving the accuracy of average human similarity. Here, N is a positive integer greater than or equal to 2. N can be set according to the actual similarity accuracy requirements; a larger N corresponds to higher similarity accuracy.

[0054] In this embodiment of the invention, step 102, calculating the similarity of human features between any two different human images in the target human image set, further includes:

[0055] Count the number of target human image sets in the target file. If there are multiple target human image sets in the target file, calculate the human feature similarity for each target human image set separately.

[0056] Specifically, once the target file is identified, it's necessary to count the number of target human image sets corresponding to the target face image sets within that file. If a target face image set corresponds to multiple target human image sets, it indicates that the target file contains multiple target human image sets, and in this case, the average human body similarity for each target human image set needs to be calculated separately. Of course, if the target file contains only one target human image set, only the human feature similarity for that target human image set needs to be calculated.

[0057] Of course, when a target file contains multiple sets of target human images, and the similarity of human features in each set meets the human similarity threshold, it indicates that these sets of target human images belong to the same person. This means that the target human image set in the target file corresponds to one set of target human images, i.e., the target file is for one person per file. Conversely, if the similarity of human features in any one of the multiple sets of target human images in the target file does not meet the human similarity threshold, it indicates that the target human image set in the target file corresponds to multiple sets of human images of different people, further suggesting that the target file contains multiple people per file.

[0058] Step 103: If the average similarity of human body features and the average similarity of facial features both meet the preset similarity threshold conditions, then the file quality level of the target file is determined to be the first file quality level.

[0059] The first file quality level is low-quality - suspected multiple files per file. Preset similarity threshold conditions include preset human body similarity threshold conditions and preset face similarity threshold conditions. The preset human body similarity threshold is pre-set and is used as one of the conditions to determine whether the target file's file quality level is at the first file quality level. The preset face similarity threshold is also pre-set and is used together with the preset human body similarity threshold to determine whether the target file's file quality level is at the first file quality level. The preset human body similarity threshold and the preset face similarity threshold are different values. The average similarity of human body features can be called the average human body similarity. The average similarity of face features can be called the average face similarity.

[0060] Specifically, when the average similarity of human body features and the average similarity of facial features are both lower than the preset similarity threshold, the file quality level of the target file is determined to be the first file quality level.

[0061] More specifically, firstly, it is determined whether the average similarity of human body features is lower than a preset human body similarity threshold and whether the average similarity of facial features is lower than a preset facial similarity threshold. If the average similarity of human body features and the average similarity of facial features are both lower than the preset human body similarity thresholds, then the file quality level of the target file is determined to be the first file quality level. If the average similarity of human body features and the average similarity of facial features are both higher than the preset human body similarity thresholds, then the file quality level of the target file is determined to be not the first file quality level.

[0062] More specifically, after calculating the average similarity of the human body images and the average similarity of the faces in the target file's set of human images, these values ​​can be compared with the corresponding preset human body similarity thresholds and preset face similarity thresholds, respectively. If the average human body similarity is lower than the preset human body similarity threshold and the average face similarity is lower than the preset face similarity threshold, then the file quality level of the target file is considered to be at the first file quality level. Once the file quality level of the target file is determined to be at the first file quality level, the file quality level of the target file can be updated accordingly to the first file quality level, that is, the file quality level of the target file can be updated to low quality - suspected multiple people in one file.

[0063] In this embodiment, the accuracy is measured by the average similarity of the top N quality face images or human body images, which can screen out multiple people in one file with a high probability. In addition, the judgment of the human body set can avoid missed judgments. When most of the images in the multiple people file belong to the same person, the selection of top N quality images is not global. If the top N quality images are all of the same person, the global average similarity will be high, which will not be effective in filtering. Adding the human body image set makes the selection of top N quality images more dispersed, which can improve the accuracy of multiple people in one file judgment.

[0064] Step 104: If the average similarity of human body features and the average similarity of facial features do not meet the preset similarity threshold, then calculate the trajectory score and dwell time of the target file respectively, and determine the file quality level of the target file based on the trajectory score and dwell time of the target file.

[0065] Specifically, in step 104, the trajectory score and dwell time of the target file are calculated, including:

[0066] Based on the spatiotemporal information of the target human images in the target human image set and the target face images in the target face image set, the target face images in the target file and the target human images in the target human image set are sorted to obtain the image sequence of the target file; the number of valid images and the total number of images in the image sequence of the target file are obtained, and the trajectory score of the target file is calculated based on the number of valid images and the total number of images; the image time of the first image and the image time of the last image in the image sequence of the target file are obtained, and the dwell time of the target file is calculated based on the image time of the first image and the image time of the last image.

[0067] Furthermore, the number of valid images in the image sequence of the target file is obtained, including:

[0068] Determine whether the next image in the image sequence of the target file is within the second-order adjacent range of the previous image; if the next image in the image sequence of the target file is within the second-order adjacent range of the previous image, increment the number of valid images by 1.

[0069] More specifically, if the average similarity of human bodies is higher than the preset human body similarity threshold and the average similarity of faces is higher than the preset face similarity threshold, it means that the file quality level of the target file is not the first file quality level. In this case, it is necessary to determine the file quality level of the target file by calculating the trajectory score and dwell time of the target file.

[0070] Furthermore, the spatiotemporal information includes the image acquisition time and the location information of the image acquisition device. After sorting the images in the target human body image set and target face image set in the target file (sorted chronologically) and obtaining the image sequence of the target file, the number of valid images is calculated by determining whether the next trajectory in the target file is within the second-order adjacent range of the previous trajectory. If the next image in the image sequence of the target file is within the second-order adjacent range of the previous image, the number of valid images is incremented by 1. That is, traversing the image sequence of the target file chronologically, starting from the second image, if the device to which the current image belongs is within the second-order adjacent range of the device to which the previous image belongs, the number of valid images is incremented by 1. Here, the second-order adjacent range refers to the ability to associate two devices through one device. The above trajectory score is equal to the ratio of the number of valid images to the total number of images, that is, trajectory score = number of valid images / total number of images. The trajectory score is a value less than 1. The closer the trajectory score is to 1, the more continuous and uniform all the images in the file are. Otherwise, the smaller the trajectory score, the more severe the image disconnection or the image trajectory is too short.

[0071] The total number of images is calculated by counting all images in the image sequence of the target file, that is, counting all face images and human body images in the target file. When a face image and a human body image are of the same type, they are counted as the same image.

[0072] The aforementioned dwell time is the image time of the last image in the image sequence of the target file minus the image time of the first image. Specifically, all images in the target file can be sorted chronologically, and the dwell time of the target file can be obtained by subtracting the image time of the first image from the image time of the last image.

[0073] In step 104, determining the file quality level of the target file based on the target file's trajectory score and dwell time includes:

[0074] If the trajectory score is lower than the first trajectory score threshold and / or the dwell time is lower than the first dwell time threshold, the archive quality level of the target archive is determined to be the second archive quality level; if the trajectory score is higher than the first trajectory score threshold and the dwell time is higher than the first dwell time threshold, the archive quality level of the target archive is determined based on the trajectory score, the second trajectory score threshold, the dwell time, and the second dwell time range.

[0075] Furthermore, based on trajectory scoring, second trajectory scoring threshold, dwell time, and the range of second dwell time, the archive quality level of the target archive is determined, including:

[0076] If the trajectory score is lower than the second trajectory score threshold, the target file's file quality level is determined to be the third file quality level; if the trajectory score is higher than the second trajectory score threshold, it is determined whether the dwell time is within the second dwell time range; if the trajectory score is not within the second dwell time range, the target file's file quality level is determined to be the third file quality level; if the trajectory score is within the second dwell time range, the target file's file quality level is determined to be the fourth file quality level.

[0077] In this embodiment, the system determines the target file's quality level as a second quality level by judging whether the trajectory score is lower than a first trajectory score threshold and / or whether the dwell time is lower than a first dwell time threshold. If the trajectory score is lower than the first trajectory score threshold or the dwell time is lower than the first dwell time threshold, the system determines whether the trajectory score is lower than a second trajectory score threshold. If the trajectory score is lower than the second trajectory score threshold, the system determines whether the target file's quality level is a third quality level. If the trajectory score is not lower than the second trajectory score threshold, the system determines whether the dwell time is within a second dwell time range. If the trajectory score is not within the second dwell time range, the system determines whether the target file's quality level is a third quality level. If the trajectory score is within the second dwell time range, the system determines whether the target file's quality level is a fourth quality level.

[0078] The second file quality level is low-quality – suspected of having multiple files belonging to one person; the third file quality level is medium-quality; and the fourth file quality level is high-quality. The first trajectory scoring threshold is a pre-set condition used to determine whether a target file belongs to the second file quality level, i.e., to determine whether the target file is low-quality – multiple files belonging to one person. The aforementioned first dwell time threshold is also a pre-set condition used to determine whether a target file belongs to the second file quality level, i.e., to determine whether the target file is low-quality – multiple files belonging to one person. The first trajectory scoring threshold and the first dwell time threshold are different values.

[0079] Specifically, after calculating the trajectory score and dwell time of the target file, they are compared with the first trajectory score threshold and the first dwell time threshold. If the trajectory score is lower than the first trajectory score threshold, or the dwell time is lower than the first dwell time threshold, it indicates that the trajectory of the target file is severely disconnected or the trajectory is too short. In this case, the target file is considered to be a small file split from the main file in the case of multiple files for one person. At this time, the file quality level of the target file is determined to be the second file quality level, i.e., low quality - suspected multiple files for one person.

[0080] The second trajectory scoring threshold is also pre-set and is used to determine whether the target file belongs to the third file quality level. The third file quality level is medium quality. The second trajectory scoring threshold is a value greater than the first trajectory scoring threshold.

[0081] Specifically, when the trajectory score exceeds the first trajectory score threshold or the dwell time exceeds the first dwell time threshold, the trajectory score is then compared with the second trajectory score threshold. If the trajectory score is lower than the second trajectory score threshold, the target file is determined to be at the third file quality level, i.e., a medium-quality file. By determining that the trajectory score is lower than the second trajectory score threshold or the dwell time is outside the second dwell time range, it is judged that the target file's trajectory is discontinuous, or the dwell time is too short, indicating it is not a valid target audience, or that the dwell time is too long, suggesting it is likely staff, and the target file level is updated to medium-quality.

[0082] The second dwell time threshold is also pre-configured and is used to determine whether the target archive meets the conditions for the third or fourth archive quality level. The fourth archive quality level is a high-quality archive. The second dwell time threshold is a value greater than the second dwell time threshold. The second dwell time threshold and the second trajectory scoring threshold are different values.

[0083] When the trajectory score is higher than the second trajectory score threshold, the dwell time needs to be compared with the second dwell time. If the trajectory score is not within the second dwell time range, the target file is determined to be of the third file quality level, i.e., a medium-quality file. If the trajectory score is within the second dwell time range, the target file is determined to be of the fourth file quality level, i.e., a high-quality file. The trajectory score threshold C is a value higher than the threshold B. Otherwise, the file level is updated to high quality, indicating a high recall accuracy and targeting the business's target audience.

[0084] Specifically, when a target file's quality level is determined to be Level 2, its quality level is updated accordingly to Level 2, meaning the file is classified as low-quality – suspected of having multiple files for one person. When a target file's quality level is determined to be Level 3, its quality level is updated accordingly to Level 3, meaning it is a medium-quality file. When a target file's quality level is determined to be Level 4, its quality level is updated accordingly to Level 4, meaning it is a high-quality file. This indicates that the target file has a high recall accuracy rate and represents the target audience of business interest.

[0085] In this embodiment of the invention, a target file is acquired, comprising a set of target face images and a set of target human body images associated with the target face image set. The similarity of human features between any two different human body images in the target human body image set is calculated, as is the similarity of facial features between any two different face images in the target human body image set. If both the average similarity of human features and the average similarity of facial features meet a preset similarity threshold, the file quality level of the target file is determined to be the first file quality level. If neither the average similarity of human features nor the average similarity of facial features meets the preset similarity threshold, the trajectory score and dwell time of the target file are calculated respectively, and the file quality level of the target file is determined based on the trajectory score and dwell time. Thus, after determining the file quality level of the target file based on the average facial similarity and the average human body similarity, the file quality level is then re-determined based on the trajectory score and dwell time, achieving a dual determination of the file quality level and thus accurately rating the file quality level of the target file. This provides direction for archive optimization, allowing for targeted optimization of low-rated archives. It also enables the evaluation of overall archiving results, facilitating the measurement of the effectiveness of algorithm iterations and their improvement. This further enhances the accuracy of archive quality assessment.

[0086] The beneficial effects brought about by the embodiments of the present invention are as follows:

[0087] 1. Introducing the concept of archive quality allows for the evaluation of overall archiving results through archive rating, which is beneficial for measuring the effectiveness of algorithm iteration and improvement.

[0088] 2. The accuracy is measured by the average similarity of the top N quality face or human body images. This can filter out multiple people in the same file with a high probability. Including the human body image set in the judgment can avoid missed judgments. When most of the trajectories in the multiple people in the same file belong to the same person, the selection of top N quality images is not global. If all the top N quality images are of the same person, the global average similarity will be high, which will not be effective in filtering. Including the human body set makes the selection of top N quality images more dispersed, which can improve the accuracy of multiple people in the same file.

[0089] 3. By calculating trajectory scores using spatiotemporal information, the consistency and rationality of the trajectory can be initially measured through the scores, which can indirectly reflect the recall and accuracy of the archives.

[0090] 4. Incorporate dwell time into the archive rating based on business needs, linking archives to business operations. The contribution of archives to business can be inferred by the proportion of high-quality archives.

[0091] 5. Applying archival quality to actual business operations, and removing files with multiple individuals and their tracking information from business analysis, can make the analysis results more reliable. Horizontal and vertical comparisons of the proportion of files at different levels can monitor archiving. If the proportion of high-quality files within the same institution suddenly drops, or if the proportion of high-quality files in one institution is significantly lower than that of other institutions, it indicates that there may be an anomaly in deployment or archiving, requiring further analysis.

[0092] It should be noted that the archival quality assessment method provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that can perform archival quality assessment.

[0093] like Figure 2 As shown, the archival quality assessment device 200 includes:

[0094] The acquisition module 201 is used to acquire the target file, which includes a set of target face images and a set of target human body images associated with the set of target face images;

[0095] The calculation module 202 is used to calculate the similarity of human features between any two different human images in the target human image set, and to calculate the similarity of facial features between any two different face images in the target human image set.

[0096] The first determining module 203 is used to determine the file quality level of the target file as the first file quality level if the average similarity of human body feature similarity and the average similarity of face feature similarity both meet the preset similarity threshold condition.

[0097] The second determining module 204 is used to calculate the trajectory score and dwell time of the target file respectively if the average similarity of human body feature similarity and the average similarity of face feature similarity do not meet the preset similarity threshold conditions, and determine the file quality level of the target file based on the trajectory score and dwell time of the target file.

[0098] Optionally, the computing module 202 includes:

[0099] The first acquisition unit is used to acquire target human images from the target human image set and target face images from the target face image set. The target human images are human images with the top N image quality in the target human image set, and the target face images are face images with the top N image quality in the target face image set.

[0100] The first calculation unit is used to calculate the similarity of human features between any two different target human images and the similarity of facial features between any two different target face images.

[0101] Optionally, the second determining module 204 includes:

[0102] The sorting unit is used to sort the target face images in the target face image set and the target human body images in the target human body image set according to the spatiotemporal information of the target human body images in the target human body image set and the spatiotemporal information of the target face images in the target face image set, so as to obtain the image sequence of the target file.

[0103] The second acquisition unit is used to acquire the number of valid images and the total number of images in the image sequence of the target file, and to calculate the trajectory score of the target file based on the number of valid images and the total number of images;

[0104] Obtain the image time of the first image and the image time of the last image in the image sequence of the target file, and calculate the dwell time of the target file based on the image time of the first image and the image time of the last image.

[0105] Optionally, the second acquisition unit includes:

[0106] The judgment sub-unit is used to determine whether the next image in the image sequence of the target file is within the second-order adjacent range of the previous image;

[0107] The overlay subunit is used to increment the number of valid images by 1 if the next image in the image sequence of the target file is within the second-order adjacent range of the previous image.

[0108] Optionally, the second determining module 204 includes:

[0109] The first determining unit is used to determine the file quality level of the target file as the second file quality level if the trajectory score is lower than the first trajectory score threshold and / or the dwell time is lower than the first dwell time threshold.

[0110] The second determining unit is used to determine the file quality level of the target file based on the trajectory score, the second trajectory score threshold, the dwell time, and the second dwell time range if the trajectory score is higher than the first trajectory score threshold and the dwell time is higher than the first dwell time threshold.

[0111] Optionally, the second determining unit includes:

[0112] The first determining subunit is used to determine the archive quality level of the target archive as the third archive quality level if the trajectory score is lower than the second trajectory score threshold.

[0113] The judgment sub-unit is used to determine whether the dwell time is within the second dwell time range if the trajectory score is higher than the second trajectory score threshold.

[0114] The second determining subunit is used to determine the archive quality level of the target archive as the third archive quality level if the trajectory score is not within the second dwell time range.

[0115] The third determining subunit is used to determine the archive quality level of the target archive as the fourth archive quality level if the trajectory score is within the second dwell time range.

[0116] Optionally, the computing module 202 also includes:

[0117] The statistical unit is used to count the number of target human body images in the target file;

[0118] The second calculation unit is used to calculate the human feature similarity of each set of target human images if there are multiple sets of target human images in the target file.

[0119] It should be noted that the archival quality assessment device 200 provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that can perform archival quality assessments.

[0120] The archival quality assessment device 200 provided in this embodiment of the invention can realize all the processes implemented by the archival quality assessment method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0121] See Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 3 As shown, it includes: a memory 302, a processor 301, and a computer program for a file quality assessment method stored on the memory 302 and executable on the processor 301, wherein:

[0122] The processor 301 is used to call the computer program stored in the memory 302 and perform the following steps:

[0123] Obtain the target file, which includes a set of target face images and a set of target human body images associated with the set of target face images;

[0124] Calculate the similarity of human features between any two different human images in the target human image set, and calculate the similarity of facial features between any two different face images in the target human image set;

[0125] If the average similarity of human body features and the average similarity of facial features both meet the preset similarity threshold, then the archive quality level of the target archive is determined to be the first archive quality level.

[0126] If the average similarity of human body features and the average similarity of facial features do not meet the preset similarity threshold, then the trajectory score and dwell time of the target file are calculated respectively, and the file quality level of the target file is determined based on the trajectory score and dwell time of the target file.

[0127] Optionally, the processor 301 performs the calculation of the similarity of human features between any two different human images in the target human image set, and the calculation of the similarity of facial features between any two different face images in the target human image set, including:

[0128] Obtain the target human body image from the target human body image set and the target face image from the target face image set. The target human body image is the human body image with the top N image quality in the target human body image set, and the target face image is the face image with the top N image quality in the target face image set.

[0129] Calculate the similarity of human features between any two different target human body images and the similarity of facial features between any two different target face images.

[0130] Optionally, the processor 301 performs separate calculations of the trajectory score and dwell time of the target file, including:

[0131] Based on the spatiotemporal information of the target human images in the target human image set and the target face images in the target face image set, the target face images in the target file and the target human images in the target human image set are sorted to obtain the image sequence of the target file.

[0132] Obtain the number of valid images and the total number of images in the image sequence of the target file, and calculate the trajectory score of the target file based on the number of valid images and the total number of images;

[0133] Obtain the image time of the first image and the image time of the last image in the image sequence of the target file, and calculate the dwell time of the target file based on the image time of the first image and the image time of the last image.

[0134] Optionally, the processor 301 performs the acquisition of the number of valid images in the image sequence of the target file, including:

[0135] Determine whether the next image in the image sequence of the target file is within the second-order adjacent range of the previous image;

[0136] If the next image in the image sequence of the target file is within the second-order adjacent range of the previous image, then the number of valid images is incremented by 1.

[0137] Optionally, the processor 301 performs the following: determining the file quality level of the target file based on the target file's trajectory score and dwell time, including:

[0138] If the trajectory score is lower than the first trajectory score threshold and / or the dwell time is lower than the first dwell time threshold, the archive quality level of the target archive is determined to be the second archive quality level.

[0139] If the trajectory score is higher than the first trajectory score threshold and the dwell time is higher than the first dwell time threshold, then the archive quality level of the target archive is determined based on the trajectory score, the second trajectory score threshold, the dwell time, and the second dwell time range.

[0140] Optionally, the processor 301 performs a process based on trajectory scoring, a second trajectory scoring threshold, dwell time, and a second dwell time range to determine the archive quality level of the target archive, including:

[0141] If the trajectory score is lower than the second trajectory score threshold, the archive quality level of the target archive is determined to be the third archive quality level;

[0142] If the trajectory score is higher than the second trajectory score threshold, then it is determined whether the dwell time is within the second dwell time range;

[0143] If the trajectory score is not within the second dwell time range, the archive quality level of the target archive is determined to be the third archive quality level;

[0144] If the trajectory score falls within the second dwell time range, then the target file's file quality level is determined to be the fourth file quality level.

[0145] Optionally, the processor 301's calculation of the similarity of human features between any two different human images in the target human image set further includes:

[0146] Count the number of target human images in the target file;

[0147] If there are multiple sets of target human images in the target file, then the similarity of human features for each set of target human images is calculated separately.

[0148] It should be noted that the electronic device 300 provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that can perform archival quality assessment methods.

[0149] The electronic device 300 provided in this embodiment of the invention can implement all the processes of the archive quality assessment method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0150] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the archival quality assessment method or the application-side archival quality assessment method provided in this invention, and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0151] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0152] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. An archival quality assessment method, characterized by, The method comprises the following steps: obtaining a target archive, the target archive comprising a target face image set and a target body image set associated with the target face image set; calculating body feature similarity corresponding to any two different body images in the target body image set, and calculating face feature similarity corresponding to any two different face images in the target face image set; if the average similarity of the body feature similarity and the average similarity of the face feature similarity both satisfy a preset similarity threshold condition, determining that the archive quality level of the target archive is a first archive quality level; if the average similarity of the body feature similarity and the average similarity of the face feature similarity both do not satisfy the preset similarity threshold condition, respectively calculating a track score and a stay time of the target archive, and determining the archive quality level of the target archive according to the track score and the stay time of the target archive; the respective calculation of the track score and the stay time of the target archive comprises: sorting target face images in the target face image set and target body images in the target body image set in the target archive according to the spatiotemporal information of the target face images in the target face image set and the spatiotemporal information of the target body images in the target body image set, to obtain an image sequence of the target archive; obtaining the number of effective images and the total number of images in the image sequence of the target archive, and calculating the track score of the target archive according to the number of effective images and the total number of images; obtaining the image time of the first image and the image time of the last image in the image sequence of the target archive, and calculating the stay time of the target archive according to the image time of the first image and the image time of the last image.

2. The method of claim 1, wherein, the calculation of the body feature similarity corresponding to any two different body images in the target body image set and the calculation of the face feature similarity corresponding to any two different face images in the target face image set comprise: obtaining target body images in the target body image set and target face images in the target face image set, the target body images being topN body images in the target body image set in terms of image quality, and the target face images being topN face images in the target face image set in terms of image quality; respectively calculating body feature similarity of any two different target body images in the target body images and face feature similarity of any two different target face images in the target face images.

3. The method of claim 1, wherein, the obtaining of the number of effective images in the image sequence of the target archive comprises: judging whether a subsequent image in the image sequence of the target archive is in a two-order adjacent range of a previous image; if the subsequent image in the image sequence of the target archive is in the two-order adjacent range of the previous image, the number of effective images is incremented by 1.

4. The method of claim 1, wherein, the determination of the archive quality level of the target archive according to the track score and the stay time of the target archive comprises: determining that the target archive has an archive quality level of a second archive quality level if the trajectory score is lower than a first trajectory score threshold and / or the dwell time is lower than a first dwell time threshold; determining that the target archive has an archive quality level based on the trajectory score, a second trajectory score threshold, the dwell time and a second dwell time range if the trajectory score is higher than the first trajectory score threshold and the dwell time is higher than the first dwell time threshold.

5. The method of claim 4, wherein, The determining that the target archive has an archive quality level based on the trajectory score, a second trajectory score threshold, the dwell time and a second dwell time range comprises: determining that the target archive has an archive quality level of a third archive quality level if the trajectory score is lower than the second trajectory score threshold; determining whether the trajectory score is within the second dwell time range if the trajectory score is higher than the second trajectory score threshold; determining that the target archive has an archive quality level of the third archive quality level if the trajectory score is not within the second dwell time range; determining that the target archive has an archive quality level of a fourth archive quality level if the trajectory score is within the second dwell time range.

6. The method of any one of claims 1-5, wherein, The calculating the body feature similarity between any two different body images in the target body image set further comprises: counting the number of target body image sets in the target archive; calculating the body feature similarity of each target body image set if the number of target body image sets in the target archive is more than one.

7. An archival quality assessment apparatus characterized by comprising: The archive quality assessment device comprises: an acquisition module configured to acquire a target archive, the target archive comprising a target face image set and a target body image set associated with the target face image set; a calculation module configured to calculate the body feature similarity between any two different body images in the target body image set and the face feature similarity between any two different face images in the target face image set; a first determination module configured to determine that the target archive has an archive quality level of a first archive quality level if the average similarity of the body feature similarity and the average similarity of the face feature similarity both satisfy a preset similarity threshold condition; a second determination module configured to calculate the trajectory score and the dwell time of the target archive respectively if the average similarity of the body feature similarity and the average similarity of the face feature similarity both do not satisfy the preset similarity threshold condition, and determine the archive quality level of the target archive based on the trajectory score and the dwell time of the target archive; The calculating the trajectory score and the dwell time of the target archive comprises: sorting the target face images in the target face image set and the target body images in the target body image set according to the spatio-temporal information of the target face images in the target face image set and the spatio-temporal information of the target body images in the target body image set to obtain an image sequence of the target archive. obtaining the number of valid images and the total number of images in the image sequence of the target archive, and calculating the track score of the target archive according to the number of valid images and the total number of images; obtaining the image time of the first image and the image time of the last image in the image sequence of the target archive, and calculating the residence time of the target archive according to the image time of the first image and the image time of the last image.

8. An electronic device, comprising: comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the archive quality evaluation method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps in the archive quality evaluation method according to any one of claims 1 to 6.

Images