Image retrieval method and device based on adaptive threshold, equipment and medium

Abstract

The present application relates to the technical field of data retrieval, and particularly relates to an image retrieval method and device based on an adaptive threshold, equipment and a medium. The method retrieves the first K reference images most similar to the image to be retrieved from a database, calculates a first-order differential curve according to a reference curve formed by the similarity corresponding to the first K reference images, uses a preset window to slide on the first-order differential curve to obtain a curve segment, calculates a standardized value according to the curve segment, takes the curve segment corresponding to the maximum standardized value as a target curve segment, determines the similarity of the reference image corresponding to the center point of the target curve segment as a similarity threshold, uses the similarity threshold to screen the first K reference images, obtains an image retrieval result, normalizes the reference curve to represent the distribution characteristics of the similarity, and adaptively determines the similarity threshold to further screen the first K reference images, reduces the missed detection and false detection, and improves the accuracy of image retrieval.

Description

Technical Field

[0001] This invention relates to the field of data retrieval technology, and in particular to an image retrieval method, apparatus, device, and medium based on adaptive threshold. Background Technology

[0002] Currently, with the development of artificial intelligence technology, image retrieval has been widely applied in many scenarios, such as intelligent security, smart communities, and smart campuses. Existing image retrieval methods often sort the images to be retrieved against all stored images in the retrieval database based on their similarity, returning the K most similar images, or perform a retrieval based on a similarity threshold, returning images whose similarity to the image to be retrieved is greater than a preset similarity threshold.

[0003] However, in practical applications, image retrieval usually requires that the returned search results belong to the same object as the image to be retrieved. Since different objects have different numbers of images stored in the retrieval database, it is difficult to retrieve them by setting a K value. Furthermore, the environmental conditions when the same object was collected and stored are not the same. Retrieving images using a fixed similarity threshold can lead to missed or false detections, resulting in low accuracy of image retrieval. Therefore, how to improve the accuracy of image retrieval has become an urgent problem to be solved. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide an image retrieval method, apparatus, device, and medium based on adaptive thresholds to solve the problem of low accuracy in image retrieval.

[0005] In a first aspect, embodiments of the present invention provide an image retrieval method based on an adaptive threshold, the image retrieval method comprising:

[0006] Obtain the image to be retrieved, and based on the similarity between the images, retrieve the top K reference images that are most similar to the image to be retrieved from a preset image database, where K is an integer greater than zero;

[0007] The similarity scores corresponding to the first K reference images are arranged in descending order to form a reference curve. The first-order difference is calculated for each similarity score in the reference curve to obtain the first-order difference curve.

[0008] Using a preset window, slide it on the first-order difference curve at a preset step size to obtain the curve segment within the preset window after each slide stops. Then, perform a standardized calculation on the curve segment to obtain the standardized value of the corresponding curve segment.

[0009] The largest standardized value among all standardized values ​​is taken as the target value. The curve segment used in the standardization calculation of the target value is determined as the target curve segment, and the center point of the target curve segment is determined.

[0010] Obtain the reference image corresponding to the center point, use the similarity between the reference image corresponding to the center point and the image to be retrieved as a similarity threshold, and use the similarity threshold to filter the first K reference images to obtain the image retrieval results.

[0011] Secondly, embodiments of the present invention provide an image retrieval device based on an adaptive threshold, the image retrieval device comprising:

[0012] The preliminary retrieval module is used to acquire the image to be retrieved and, based on the similarity between the images, retrieve the top K reference images that are most similar to the image to be retrieved from a preset image database, where K is an integer greater than zero;

[0013] The difference calculation module is used to form a reference curve by arranging the similarities corresponding to the first K reference images in descending order, and to perform first-order difference calculation on each similarity in the reference curve to obtain a first-order difference curve.

[0014] The curve standardization module is used to slide a preset window on the first-order difference curve at a preset step size to obtain the curve segment within the preset window after each sliding stops, and to perform standardization calculation on the curve segment to obtain the standardized value of the corresponding curve segment.

[0015] The center point determination module is used to take the largest standardized value among all standardized values ​​as the target value, determine the curve segment used when performing standardization calculation on the target value as the target curve segment, and determine the center point of the target curve segment.

[0016] The image filtering module is used to obtain the reference image corresponding to the center point, use the similarity between the reference image corresponding to the center point and the image to be retrieved as a similarity threshold, and use the similarity threshold to filter the first K reference images to obtain the image retrieval results.

[0017] Thirdly, embodiments of the present invention provide a computer device, the computer device including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the image retrieval method as described in the first aspect.

[0018] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the image retrieval method as described in the first aspect.

[0019] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0020] The process involves acquiring the image to be retrieved, retrieving the top K most similar reference images from a pre-defined image database based on their similarity, and forming reference curves by arranging the similarity scores of these top K reference images in descending order. First-order differencing is then performed on each similarity score within these reference curves to obtain a first-order differencing curve. A pre-defined window is then slid along the first-order differencing curve at a pre-defined step size to obtain the curve segment within the window after each slid. Standardization calculations are performed on these curve segments to obtain their standardized values. The largest standardized value among all standardized values ​​is selected as the target value. The standardization calculation is then performed on the target value. The curve segment used is the target curve segment. The center point of the target curve segment is determined, and the reference image corresponding to the center point is obtained. The similarity between the reference image corresponding to the center point and the image to be retrieved is used as the similarity threshold. The first K reference images are filtered using the similarity threshold to obtain the image retrieval results. The reference curve composed of the similarity corresponding to the first K reference images retrieved in the initial step is standardized to characterize the distribution characteristics of the similarity. The similarity threshold is adaptively determined according to the distribution characteristics of the similarity, and the first K reference images are further filtered to obtain the retrieval images belonging to the same object, thereby reducing the situation of missed detections and false detections and improving the accuracy of image retrieval. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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.

[0022] Figure 1 This is a schematic diagram of an application environment for an image retrieval method based on adaptive threshold provided in Embodiment 1 of the present invention;

[0023] Figure 2 This is a flowchart illustrating an image retrieval method based on adaptive threshold provided in Embodiment 1 of the present invention;

[0024] Figure 3 This is a flowchart illustrating an image retrieval method based on adaptive threshold provided in Embodiment 2 of the present invention;

[0025] Figure 4 This is a schematic diagram of the structure of an image retrieval device based on adaptive threshold provided in Embodiment 3 of the present invention;

[0026] Figure 5 This is a schematic diagram of the structure of a computer device provided in Embodiment 4 of the present invention. Detailed Implementation

[0027] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0028] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0029] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0030] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0031] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0032] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0033] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0034] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0035] The first embodiment of this invention provides an image retrieval method based on adaptive thresholds, which can be applied to applications such as... Figure 1 In this application environment, the client communicates with the server. The client includes, but is not limited to, handheld computers, desktop computers, laptops, ultra-mobile personal computers (UMPCs), netbooks, cloud terminal devices, and personal digital assistants (PDAs). The server can be a standalone server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0036] The server can communicate with at least one image acquisition device to obtain the image to be retrieved from the device. Image acquisition devices include, but are not limited to, cameras, video recorders, and handheld video cameras. These devices can be deployed in various application scenarios, such as smart security, smart communities, and smart campuses. The client can search for the image to be retrieved in a pre-defined image database to find images similar to the image to be retrieved. For example, in a smart community scenario, image retrieval can be used to monitor the flow of people within the community in real time to identify abnormal individuals. Yes, in this case, image retrieval primarily targets facial images. Clients deployed within the community store facial images of residents. Image acquisition devices deployed within the community obtain real-time images of residents and transmit them to the server. Since the object of image retrieval is facial images, the server needs to preprocess the real-time images to obtain images of facial regions within the real-time images. If a person in the real-time image is not a community resident, the processed facial image will be dissimilar to all facial images of community residents, thus enabling the identification of non-residents for entry and exit records. It is understood that in the specific embodiments of this application, user identity information, address information, facial image information, and other related privacy data are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of related privacy data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0037] See Figure 2 This is a flowchart illustrating an image retrieval method based on adaptive thresholding provided in Embodiment 1 of the present invention. The aforementioned image retrieval method based on adaptive thresholding can be applied to... Figure 1 The client-side computer connects to the server to retrieve the images to be searched. The client has storage capabilities; the computer device storing the client contains an image database, which can be used for image retrieval of the images to be searched. For example... Figure 2 As shown, the image retrieval method based on adaptive thresholds may include the following steps:

[0038] Step S201: Obtain the image to be retrieved, and retrieve the top K reference images that are most similar to the image to be retrieved from the preset image database based on the similarity between the images.

[0039] Here, the image to be retrieved can refer to the image that needs to be retrieved in the image database. Similarity can be used to characterize the difference between two images. The image database can store several images and can be used to provide retrieval data for image retrieval. K is an integer greater than zero. The reference image can refer to the image in the image database that ranks higher in similarity to the image to be retrieved.

[0040] Specifically, in this embodiment, both the image to be retrieved and the images stored in the image database can be face images. That is, the image retrieval task in this embodiment is face image retrieval. By default, after the image acquisition device acquires the real-time image, it has already preprocessed the real-time image to obtain the image to be retrieved that only contains face region image information. The preprocessing process can use object detection models, semantic segmentation models, etc. In this embodiment, object detection models can be used for preprocessing. The detection object is the human face. The real-time image is input into the object detection model, and the output is the detection bounding box of the human face. The real-time image is cropped according to the detection bounding box, and the size of the cropping result is normalized to obtain the image to be retrieved.

[0041] In one embodiment, a semantic segmentation model can also be used for preprocessing. The segmentation categories of pixels can include human face categories and other categories. The real-time image is input into the target detection model, and the output is a semantic segmentation map. The minimum bounding rectangle of all pixels belonging to the human face category in the semantic segmentation map is obtained. The real-time image is cropped using the minimum bounding rectangle, and the size of the cropping result is normalized to obtain the image to be retrieved.

[0042] The similarity between each stored image in the preset image database and the image to be retrieved is obtained. The similarity can be calculated using a Siamese network. Specifically, the input to the Siamese network for each calculation is any stored image and the image to be retrieved from the preset image database. The stored image and the image to be retrieved are respectively processed by the encoder in the Siamese network to extract features, resulting in a first feature vector for the corresponding stored image and a second feature vector for the corresponding image to be retrieved. Euclidean distance is calculated between the first feature vector and the second feature vector. The calculated Euclidean distance is used to represent the similarity between the stored image and the image to be retrieved. The larger the Euclidean distance, the smaller the similarity between the stored image and the image to be retrieved. The closer the Euclidean distance is to zero, the greater the similarity between the stored image and the image to be retrieved.

[0043] The Siamese network used for similarity calculation is a pre-trained Siamese network, meaning the model parameters of the Siamese network are fixed. Therefore, the first feature vector output by each stored image after inputting into the pre-trained Siamese network is also fixed, and the output depends only on the input. Thus, the first feature vector corresponding to each stored image can be pre-stored. After obtaining the image to be retrieved, it is input into the pre-trained Siamese network to obtain the second feature vector. The Euclidean distance is then calculated directly between the second feature vector and the first feature vector corresponding to each stored image. Therefore, the entire similarity calculation process only requires one feature extraction of the image to be retrieved through the pre-trained Siamese network to achieve similarity calculation at the feature level between the image to be retrieved and each stored image in the image database.

[0044] It should be noted that the trained Siamese network mentioned above may only include a single trained encoder, that is, only feature extraction is required. During the training process of the Siamese network, the training architecture may include two encoders and comparators with shared parameters. The implementer can obtain the training set from the actual application scenario in this embodiment to train the Siamese network so that the trained encoder can more accurately extract the feature information in the application scenario. For example, in the smart community scenario, it is necessary to calculate the similarity of face images. During training, each training sample contains two images. The two images can be face images of the same person or face images of different people. The labels of the training samples can be two categories, one for the same person category and the other for different people categories. The loss function during training can be a contrastive loss function.

[0045] The steps described above—obtaining the image to be retrieved, retrieving the top K reference images most similar to the image to be retrieved from a preset image database based on the similarity between images—perform a preliminary search of the image to be retrieved from the preset image database, excluding stored images in the image database that are irrelevant to the image to be retrieved. This also facilitates subsequent optimization of the preliminary search results to obtain more accurate search results.

[0046] Step S202: The similarity scores corresponding to the first K reference images are arranged in descending order to form reference curves. First-order difference calculation is performed on each similarity score in the reference curves to obtain the first-order difference curve.

[0047] The reference curve can be used to characterize the trend of similarity changes corresponding to the first K reference images. The coordinate system in which the reference curve is located can be the first coordinate system. The horizontal axis of the first coordinate system can be the sequence identifier axis, and the vertical axis of the first coordinate system can be the similarity axis. The first-order difference can refer to the similarity difference value corresponding to adjacent sequence identifiers in the reference curve. The first-order difference curve can be used to characterize the trend of first-order difference value changes.

[0048] Specifically, the similarity scores of the first K reference images are arranged in descending order. Based on the arrangement, a sequence identifier is determined for each reference image. The sequence identifier can be an integer within the range [1, K]. For example, the sequence identifier for the reference image with the highest similarity is 1, and the sequence identifier for the reference image with the lowest similarity is K. For any reference image, using its sequence identifier as the x-axis and its corresponding similarity score as the y-axis, a coordinate point can be obtained on a preset first coordinate system based on the sequence identifier and similarity score. By traversing each reference image, K coordinate points on the first coordinate system are obtained. Curve fitting is then performed on the K coordinate points on the first coordinate system to obtain a reference curve.

[0049] Optionally, the reference curve contains K similarities;

[0050] Performing first-order difference calculations on each similarity in the reference curve yields the following first-order difference curves:

[0051] For any similarity in the reference curve, determine the associated similarity adjacent to that similarity in the reference curve;

[0052] Calculate the difference between similarity and association similarity, and use the difference as the first-order difference value of similarity in the first-order difference curve;

[0053] By iterating through all similarities, we obtain K first-order difference values ​​corresponding to the similarities, and then determine the K first-order difference values ​​to form a first-order difference curve.

[0054] Among them, the sequence identifier corresponding to the association similarity can be adjacent to the sequence identifier corresponding to the target similarity, and the first-order difference value can be used to characterize the difference between the similarity between the two reference images and the image to be retrieved.

[0055] Specifically, for any similarity in the reference curve, the corresponding sequence identifier is obtained. That is, in the first coordinate system, for any coordinate point's ordinate, the abscissa of that coordinate point is obtained. Since the value range of the sequence identifier in this embodiment is an integer within [1, K], and the difference in abscissa between adjacent coordinate points is 1, let the sequence identifier corresponding to this similarity be j. The similarity with sequence identifier j+1 is determined as the association similarity of this similarity. The first-order difference value corresponding to the similarity with sequence identifier j can be expressed as:

[0056] Δp j =p j -p j+1

[0057] Where, p j It can refer to the similarity of sequence identifier j, p j+1It can refer to the similarity of the sequence number j+1, Δp j It can refer to the first-order difference value corresponding to the similarity of the sequence identifier j, where the value of j ranges from [1, K-1].

[0058] Construct a second coordinate system based on the sequence identifier and the first-order difference value. For any reference image, use its sequence identifier as the x-axis and its corresponding first-order difference value as the y-axis to obtain a coordinate point in the second coordinate system. Traverse each reference image to obtain K coordinate points in the second coordinate system. Perform curve fitting on the K coordinate points in the second coordinate system to obtain the first-order difference curve.

[0059] It should be noted that, since the first K reference images retrieved initially need to be filtered again in this embodiment, K can be set to a large value to avoid omissions due to the different number of face images of each person stored in the image database. For example, in this embodiment, K can be set to 250. Under normal circumstances, the number of face images of the same person in the image database will not exceed this value of K. Therefore, when sorting similarity, the similarity of the last one can be defaulted to 0. When calculating the first-order difference value, the first-order difference value of the reference image with the order identifier K can be directly set to 0.

[0060] 0. This embodiment uses sequential identifiers to determine the association similarity, avoiding situations where there are identical similarities.

[0061] The calculation of the first-order difference values ​​between similarity values ​​is chaotic, which makes it impossible to provide a reliable first-order difference curve for subsequent standardization processing, thereby improving the accuracy of the overall image retrieval process.

[0062] The above-mentioned similarity scores of the first K reference images are arranged in descending order to form a reference curve.

[0063] The five steps involved in calculating the first-order difference curve for each similarity level in the reference curve highlight the differences between various reference images.

[0064] This allows for the determination of similarity thresholds, leading to more accurate image retrieval.

[0065] Step S203: Use a preset window to slide on the first-order difference curve with a preset step size to obtain the curve segment within the preset window after each slide stops. Perform standardized calculation on the curve segment to obtain the standardized value of the corresponding curve segment.

[0066] 0. Here, the preset window can refer to a sliding window in the coordinate system where the first-order difference curve is located, with a preset step size.

[0067] This can refer to the distance a preset window moves each time; sliding can refer to the preset window moving in a preset direction; a curve segment can refer to a part of a first-order difference curve; standardized calculation can be used to convert multiple sets of data into unitless scores, making the data standards uniform, improving data comparability, and weakening...

[0068] Data interpretability: Standardized values ​​can refer to the result of standardization processing, i.e., the unitless scores mentioned above. Specifically, the preset window's size can be determined by its width and height to ensure that the preset window can cover...

[0069] For all points on the first-order difference curve, the height h of the preset window should satisfy h ≥ max(Δp). j The value of j is in the range of [1, K-1]. The bottom border of the preset window moves along the horizontal axis of the coordinate system containing the first-order difference curve, with the direction of movement being the positive direction of the horizontal axis. Initially, the left border of the preset window coincides with the vertical axis of the coordinate system containing the first-order difference curve.

[0070] In this embodiment, the width w of the preset window is set to 10 and the step size is set to 1. Then, each time the window is slid, a curve segment containing 10 first-order difference values ​​can be extracted. The size of the preset window and the preset step size can be adjusted by the implementer according to the actual situation.

[0071] Optionally, the curve segment includes the curve segment inside the window and the curve segment at the tail of the window;

[0072] Using a preset window, slide it along the first-order difference curve at a preset step size. The curve segment within the preset window after each slide stops includes:

[0073] Use a preset window to slide on the first-order difference curve at a preset step size. After each slide stops, obtain the start and end positions of the preset window.

[0074] The first-order difference curve is truncated by the start and end positions to obtain the curve segment inside the window;

[0075] By using the starting position and the ending position of the first-order difference curve, the first-order difference curve is truncated to obtain the curve segment at the tail of the window.

[0076] Accordingly, the curve segments are standardized to obtain standardized values ​​for the corresponding curve segments, including:

[0077] Based on the curve segments inside and at the end of the window, the curve segments inside the window are standardized to obtain the standardized values ​​of the corresponding curve segments inside the window.

[0078] The internal curve segment of the window can refer to the curve segment extracted within the preset window, while the tail curve segment of the window can refer to the curve segment extracted within the preset window and to the right of the preset window. The starting position can refer to the x-coordinate of the left border of the preset window when it stops scrolling, and the ending position can refer to the x-coordinate of the right border of the preset window when it stops scrolling. The ending position of the first-order difference curve can refer to the x-coordinate of the rightmost point of the first-order difference curve.

[0079] Specifically, after obtaining the start and end positions of the preset window, the portion of the first-order difference curve whose x-coordinate is less than the start position or greater than the end position of the preset window is discarded, and the remaining portion is the curve segment inside the window. Similarly, the portion of the first-order difference curve whose x-coordinate is less than the start position of the preset window is discarded, and the remaining portion is the curve segment at the end of the window.

[0080] If the starting position of the preset window is set to i, then the ending position can be represented as i+w, and the ending position of the first-order difference curve can be represented as K. That is, the curve segment inside the window is extracted through the interval [i, i+w], and the curve segment at the end of the window is extracted through the interval [i, K].

[0081] In this embodiment, the curve segments inside the window and the curve segments at the end of the window are extracted by pre-setting the window. This allows the information of the curve segments at the end of the window to be used as a reference during the standardization process of the curve segments inside the window, effectively uncovering the correlation between data and thus improving the effect of standardization.

[0082] Optionally, based on the curve segments inside the window and the curve segments at the end of the window, the curve segments inside the window are standardized to obtain the standardized values ​​of the corresponding curve segments inside the window, including:

[0083] Calculate the first mean of all first-order difference values ​​in the curve segment inside the window, and calculate the second mean and standard deviation of all first-order difference values ​​in the curve segment at the tail of the window.

[0084] Based on the first mean, the second mean, and the standard deviation, the curve segments inside the window are standardized to obtain the standardized values ​​of the corresponding curve segments inside the window.

[0085] The first mean can refer to the mean of all first-order difference values ​​in the curve segment inside the window, the second mean can refer to the mean of all first-order difference values ​​in the curve segment at the tail of the window, and the standard deviation can be used to characterize the numerical stability of all first-order difference values ​​in the curve segment at the tail of the window.

[0086] Specifically, the first mean μ1 can be expressed as Where w is the width of the preset window, which is the difference between the end position and the start position of the preset window, and Δp nIt can refer to the first-order difference value corresponding to the nth sequential identifier, where n ranges from [i, i+w]. The second mean μ2 can be expressed as... Δp m It can refer to the first-order difference value corresponding to the m-th sequential identifier, where m ranges from [i, K]. The standard deviation σ can be expressed as...

[0087] This embodiment uses the mean and standard deviation for standardization calculation, which is convenient and quick, and can effectively improve the efficiency of curve segment standardization processing, thereby improving the overall efficiency of image retrieval.

[0088] Optionally, based on the first mean, the second mean, and the standard deviation, the curve segments within the window are standardized to obtain the standardized values ​​of the corresponding curve segments within the window, including:

[0089] Calculate the difference between the first mean and the second mean;

[0090] The absolute value of the calculation result is compared with the standard deviation, and the resulting ratio is used as the standardized value of the curve segment inside the corresponding window.

[0091] The difference between the first mean and the second mean can be used to characterize the difference between the overall similarity change of the curve segments inside the window and the overall similarity change of the curve segments at the end of the window. The ratio of the absolute value of the calculation result to the standard deviation can be used to characterize the standardized score.

[0092] Specifically, the formula for calculating the standardized value z can be expressed as:

[0093]

[0094] This embodiment uses the z-score method to standardize the curve segments inside the window, which can obtain standardized values ​​that are easy to compare. These values ​​can then be compared in the subsequent determination of target values ​​to more accurately identify the target curve segments, thereby improving the accuracy of image retrieval.

[0095] The above steps involve using a preset window to slide along a first-order difference curve at a preset step size, obtaining the curve segment within the preset window after each slide stops, and then performing standardized calculations on the curve segment to obtain the standardized value of the corresponding curve segment. By using the curve segment as the object of standardization, the information of multiple reference images can be measured simultaneously when determining the target curve segment, thereby improving the accuracy of target curve segment determination and thus improving the accuracy of image retrieval.

[0096] Step S204: Take the largest standardized value among all standardized values ​​as the target value, determine the curve segment used for standardization calculation of the target value as the target curve segment, and determine the center point of the target curve segment.

[0097] The standardized value can be used to characterize the degree of similarity change, while the target value can be used to identify the curve segment with the greatest degree of similarity change, i.e., the target curve segment. The center point can be determined by the average of the starting and ending positions of the target curve segment.

[0098] Specifically, in this embodiment, the target curve segment is determined based on the standardized value. The standardized value corresponding to each curve segment can characterize the degree of drastic change in similarity within that curve segment. The larger the standardized value, the greater the magnitude and the faster the frequency of change in similarity within that curve segment. Similarly, the smaller the standardized value, the smaller the magnitude and the slower the frequency of change in similarity within that curve segment.

[0099] When the similarity within a curve segment changes significantly, it can be considered that the segment contains a turning point in image retrieval. For example, in a face image retrieval task, this turning point might refer to a change in the retrieved images from belonging to the same person as the image to belonging to a different person. At this point, the search results can be further filtered based on this turning point, retaining only images belonging to the same person as the image to be retrieved. These images are the target of face image retrieval. However, using the top K most similar images as the search results can be problematic because the number of face images stored for each person in the image database is not large. If the same applies to the retrieval of facial images of different individuals, the number of facial images stored may exceed the K value, leading to omissions in the image retrieval process. Alternatively, the number of facial images stored may be less than the K value, resulting in the retrieval results including facial images of other individuals, i.e., false detections, leading to low accuracy in facial image retrieval. If a similarity threshold is used, due to the occlusion of the facial region or the influence of external environmental factors, two facial images of different individuals may also be similar. When using a fixed similarity threshold, it is difficult to determine a similarity threshold applicable to all individuals. Therefore, determining the inflection point becomes particularly important.

[0100] In this embodiment, the center point of the target curve segment is used as the turning point. The center point can be determined based on the average of the starting and ending positions of the target curve segment. It should be noted that the result of the average calculation needs to be moduloed, and the result of the modulo calculation is used as the center point.

[0101] In one embodiment, the starting position of the target curve segment, the width of the curve segment, and a preset coefficient can be used to determine the turning point. That is, the curve segment width and the preset coefficient are multiplied together, and the result of the multiplication is added to the starting position of the target curve segment. The result of the addition is used as the turning point. Since the center point of the target curve segment needs to be obtained in this embodiment, the preset coefficient can be set to 0.5. The result of the addition also needs to be calculated by taking the modulus, and the result of the modulus calculation is used as the center point.

[0102] It should be noted that implementers can adjust the values ​​of the above preset coefficients according to the actual situation to adjust the position of the inflection point in the target curve segment. For example, the preset coefficient can be set to 1 / 3, and the value range of the preset coefficient is [0, 1].

[0103] The above steps involve taking the largest standardized value among all standardized values ​​as the target value, determining the curve segment used for standardization calculation of the target value as the target curve segment, and determining the center point of the target curve segment to determine the inflection point. This allows for further filtering of the preliminary search results based on the inflection point, thereby improving the accuracy of image retrieval results.

[0104] Step S205: Obtain the reference image corresponding to the center point, use the similarity between the reference image corresponding to the center point and the image to be retrieved as the similarity threshold, and use the similarity threshold to filter the first K reference images to obtain the image retrieval results.

[0105] The similarity threshold can be used to determine whether a reference image is sufficiently similar to the image to be retrieved. The image retrieval result can refer to a reference image that is sufficiently similar to the image to be retrieved.

[0106] Specifically, implementers can use common face image datasets to evaluate and verify the method of this embodiment, such as the CASIA Webface dataset. Precision and recall can be defined as evaluation criteria. Precision can be represented as the ratio of the number of images belonging to the retrieved person in the image retrieval results to the total number of images in the image retrieval results. Recall can be represented as the ratio of the number of images belonging to the retrieved person in the image retrieval results to the total number of images belonging to the retrieved person in the image database. When both precision and recall meet preset thresholds, the image retrieval method of this embodiment can be applied to specific application scenarios.

[0107] Optionally, a similarity threshold can be used to filter the top K reference images, resulting in image retrieval results including:

[0108] The similarity between each of the first K reference images and the image to be retrieved is compared with a similarity threshold;

[0109] The reference images with similarity scores greater than the similarity threshold are retained. All retained reference images are then sorted in descending order of similarity to obtain the image retrieval results.

[0110] In this embodiment, reference images with similarity scores greater than the similarity threshold can be considered as images sufficiently similar to the image to be retrieved, while reference images with similarity scores less than or equal to the similarity threshold can be considered as images not sufficiently similar to the image to be retrieved.

[0111] This embodiment retains reference images with similarity scores greater than a similarity threshold. By dynamically determining the similarity threshold, it effectively integrates sorting and threshold retrieval methods, thereby improving the accuracy of image retrieval.

[0112] The steps described above—obtaining the reference image corresponding to the center point, using the similarity between the reference image corresponding to the center point and the image to be retrieved as a similarity threshold, and using the similarity threshold to filter the first K reference images to obtain the image retrieval results—further filter the preliminary retrieval results using the similarity threshold, thereby retaining the retrieved images that belong to the same object as the image to be retrieved, thus improving the accuracy of image retrieval.

[0113] In this embodiment, the reference curve composed of the similarity of the first K reference images retrieved is standardized to characterize the distribution characteristics of the similarity. Based on the distribution characteristics of the similarity, the similarity threshold is adaptively determined, and the first K reference images are further filtered to obtain the retrieved images belonging to the same object, thereby reducing the cases of missed detections and false detections and improving the accuracy of image retrieval.

[0114] See Figure 3 This is a flowchart illustrating an image retrieval method based on adaptive threshold provided in Embodiment 2 of the present invention. In this image retrieval method based on adaptive threshold, when retrieving the top K reference images most similar to the image to be retrieved from a preset image database, it is necessary to calculate the similarity between the images. The similarity between the images can be calculated using a model or by direct calculation.

[0115] When using a model for calculation, the similarity between images can be compared at the feature level. It is not easily affected by factors such as image noise, and the similarity calculation is highly robust. For the specific process of using the model for similarity calculation, please refer to Example 1, which will not be repeated here.

[0116] When directly calculating the similarity between images, the similarity calculation process includes the following steps:

[0117] Step S301: Calculate the similarity between each stored image in the image database and the image to be retrieved;

[0118] Step S302: Determine the top K similarities from all similarities, and determine the reference images corresponding to the top K similarities as the top K reference images most similar to the image to be retrieved.

[0119] The similarity can be calculated using metrics such as cosine similarity, Euclidean distance, and Manhattan distance. Similarity can be used to characterize the differences between the stored image and the image to be retrieved.

[0120] Specifically, in this embodiment, cosine similarity is used to calculate the similarity. The value range of cosine similarity is [0, 1]. The closer the cosine similarity is to 0, the less similar the stored image used for cosine similarity calculation is to the image to be retrieved. The closer the cosine similarity is to 1, the more similar the stored image used for cosine similarity calculation is to the image to be retrieved.

[0121] Using a model for similarity calculation requires additional deployment of the model within the client, which is not suitable for scenarios with limited client storage space. Moreover, if the calculation is accelerated by storing feature vectors as in Example 1, additional storage space is required. Therefore, using a model for similarity calculation is difficult to apply well to scenarios with low deployment costs. For example, in the case of smart security in residential buildings, since security work in residential buildings is usually handled by property management companies rather than dedicated security companies, the deployment cost is low and computing devices with large storage space cannot be provided. In this case, the method of directly performing similarity calculation can calculate the cosine similarity between the image to be retrieved and the stored images after the image to be retrieved is collected, thereby obtaining the top K reference images in the image database that are most similar to the image to be retrieved without increasing the storage space load.

[0122] In this embodiment, a method for directly calculating similarity is provided for scenarios with low deployment costs, which reduces the application difficulty of image retrieval methods and improves the breadth of application scenarios for image retrieval methods.

[0123] Corresponding to the image retrieval method based on adaptive thresholds in the above embodiments, Figure 4 A structural block diagram of an image retrieval device based on adaptive thresholding according to Embodiment 3 of the present invention is shown. This device is applied to a client, whose corresponding computer device is connected to a server to obtain the image to be retrieved. The client has a storage function, and the computer device storing the client contains an image database, which can be used to perform image retrieval on the image to be retrieved. For ease of explanation, only the parts relevant to the embodiments of the present invention are shown.

[0124] See Figure 4 The image retrieval device based on adaptive thresholds includes:

[0125] The preliminary retrieval module 41 is used to acquire the image to be retrieved and, based on the similarity between the images, retrieve the top K reference images that are most similar to the image to be retrieved from the preset image database, where K is an integer greater than zero;

[0126] The difference calculation module 42 is used to form a reference curve by arranging the similarities corresponding to the first K reference images in descending order, and to perform first-order difference calculation on each similarity in the reference curve to obtain the first-order difference curve.

[0127] The curve standardization module 43 is used to slide a preset window on a first-order difference curve at a preset step size to obtain the curve segment within the preset window after each slide stops, and to perform standardization calculation on the curve segment to obtain the standardized value of the corresponding curve segment.

[0128] The center point determination module 44 is used to take the largest standardized value among all standardized values ​​as the target value, determine the curve segment used when performing standardization calculation on the target value as the target curve segment, and determine the center point of the target curve segment.

[0129] The image filtering module 45 is used to obtain the reference image corresponding to the center point, and use the similarity between the reference image corresponding to the center point and the image to be retrieved as the similarity threshold. The first K reference images are filtered using the similarity threshold to obtain the image retrieval results.

[0130] Optionally, the aforementioned preliminary retrieval module 41 includes:

[0131] The similarity calculation submodule is used to calculate the similarity between each stored image in the image database and the image to be retrieved;

[0132] The image determination submodule is used to determine the top K similarities from all similarities, and to determine the reference images corresponding to the top K similarities as the top K reference images most similar to the image to be retrieved.

[0133] Optionally, the reference curve contains K similarities;

[0134] The above-mentioned difference calculation module 42 includes:

[0135] The association determination submodule is used to determine the association similarity adjacent to any similarity in the reference curve.

[0136] The difference calculation submodule is used to calculate the difference between similarity and association similarity, and the difference is used as the first difference value of similarity in the first difference curve;

[0137] The curve composition submodule is used to traverse all similarities, obtain K first-order difference values ​​corresponding to the similarity, and determine the K first-order difference values ​​to form a first-order difference curve.

[0138] Optionally, the curve segment includes the curve segment inside the window and the curve segment at the tail of the window;

[0139] The curve standardization module 43 mentioned above includes:

[0140] The position acquisition submodule is used to slide a preset window on a first-order difference curve at a preset step size, and acquire the start and end positions of the preset window after each slide stops.

[0141] The first extraction submodule is used to extract the first-order difference curve by means of the start position and the end position, so as to obtain the curve segment inside the window;

[0142] The second truncation submodule is used to truncate the first-order difference curve by using the starting position and the ending position of the first-order difference curve to obtain the curve segment at the end of the window.

[0143] Accordingly, the curve standardization module 43 mentioned above includes:

[0144] The standardization calculation submodule is used to perform standardization calculations on the curve segments inside the window based on the curve segments inside the window and the curve segments at the end of the window, so as to obtain the standardized value of the corresponding curve segments inside the window.

[0145] Optionally, the above-mentioned standardized computing submodule includes:

[0146] The first calculation unit is used to calculate the first mean of all first-order difference values ​​in the curve segment inside the window, and to calculate the second mean and standard deviation of all first-order difference values ​​in the curve segment at the tail of the window.

[0147] The second calculation unit is used to perform standardized calculations on the five curve segments inside the window based on the first mean, the second mean, and the standard deviation, to obtain the standardized values ​​of the corresponding curve segments inside the window.

[0148] Optionally, the second computing unit mentioned above includes:

[0149] The difference calculation subunit is used to calculate the difference between the first mean and the second mean;

[0150] The ratio calculation sub-unit is used to take the absolute value of the calculation result and compare it with the standard deviation, and use the resulting ratio as the standardized value of the curve segment inside the corresponding window.

[0151] Optionally, the image filtering module 45 mentioned above includes:

[0152] The threshold comparison submodule is used to compare the similarity between each of the first K reference images and the image to be retrieved with a similarity threshold.

[0153] The result determination submodule is used to retain reference images with similarity greater than the similarity threshold, and sort all the retained reference images in descending order of similarity to obtain the image retrieval results.

[0154] 5. It should be noted that the information interaction and execution between the above modules, sub-modules, units, and sub-units...

[0155] The process and other details are based on the same concept as the method embodiments of the present invention. For specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0156] Figure 5 This is a schematic diagram of the structure of a computer device provided in Embodiment 4 of the present invention. Figure 5 As shown, the computer device of this embodiment includes: at least one processor ( Figure 5 Only one is shown in the image), memory.

[0157] And a computer program stored in memory and executable on at least one processor, wherein the processor executes the computer program to implement the steps in any of the above embodiments of the adaptive threshold-based image retrieval method.

[0158] The computer device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that... Figure 5 The examples of computer devices are merely examples and do not constitute a limitation on computer devices. Computer devices may include more or fewer components than shown in the illustration, or combinations of certain components, or different components, such as network interfaces, displays, and input devices.

[0159] The processor referred to can be a CPU, but it can also be other general-purpose processors or digital signal processors.

[0160] Processors (Digital Signal Processors, DSPs), Application-Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0161] The memory includes readable storage media, internal memory, etc., wherein the internal memory can be the RAM of a computer device, and the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the readable storage media. The readable storage media can be the hard drive of the computer device, and in other embodiments it can also be the external storage device of the computer device, such as a plug-in hard drive, a smart media card (SMC), a secure digital card (SD card), or flash memory equipped on the computer device.

[0162] Memory includes devices such as flash cards. Furthermore, memory can include both internal storage units and external storage devices within a computer device. Memory is used to store operating systems, applications, bootloaders, data, and other programs, such as the program code of computer programs. Memory can also be used to temporarily store data that has been output or will be output.

[0163] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, only the above...

[0164] The division of the functional units and modules described above is illustrated with examples. In practical applications, the functions described above can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. In the embodiments, the functional units and modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware.

[0165] The invention can be implemented in the form of software functional units. Furthermore, the specific names of each functional unit and module are merely for easy differentiation and are not intended to limit the scope of protection of this invention.

[0166] The specific working process of the centering unit and module can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, this…

[0167] The invention implements all or part of the processes in the above embodiments by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it implements the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may at least include: a medium capable of carrying a computer program.

[0168] Any entity or device containing code, recording media, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical discs. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0169] The present invention can implement all or part of the processes in the methods of the above embodiments, or it can be accomplished by a computer program product. When the computer program product is run on a computer device, the computer device executes the steps in the above method embodiments.

[0170] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0171] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0172] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0173] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0174] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An adaptive threshold-based image retrieval method, characterized by, The image retrieval method comprises: The method comprises the following steps: acquiring a to-be-retrieved image, retrieving a predetermined number of reference images most similar to the to-be-retrieved image from a predetermined image database according to the similarity between the images, The predetermined number of reference images are used as a reference for image retrieval. The predetermined number is an integer greater than zero. The former The similarity scores corresponding to each of the reference images are arranged in descending order to form a similarity reference curve. The first-order difference is calculated for each similarity score in the similarity reference curve to obtain a first-order difference curve. The horizontal axis of the similarity reference curve is the sequential identifier of the K reference images, and the vertical axis is the similarity score. sliding on the first-order difference curve using a preset window according to a preset step length, obtaining a curve segment in the preset window after each sliding stop, and performing standardization calculation on the curve segment to obtain a standardization value corresponding to the curve segment; determining a maximum standardization value among all standardization values as a target value, determining a curve segment used for standardization calculation of the target value as a target curve segment, and determining a center point of the target curve segment, wherein the center point is a sequential identifier in the similarity reference curve by multiplying a curve segment width and a preset coefficient, adding a multiplication result and a starting position of the target curve segment, and performing modulo calculation on an addition result; Obtain the reference image corresponding to the sequence identifier represented by the center point, and use the similarity between the reference image corresponding to the center point and the image to be retrieved as a similarity threshold. Use the similarity threshold to evaluate the previous... The image retrieval results are obtained by filtering through reference images.

2. The image retrieval method according to claim 1, characterized in that, retrieving, from a preset image database, a preset number of reference images most similar to the image to be retrieved according to similarity between images include retrieving, from a preset image database, a preset number of reference images most similar to the image to be retrieved according to similarity between images calculating a similarity between each stored image in the image database and the image to be retrieved; determining the maximum one from all similarities determining the maximum one from all similarities the reference images corresponding to the maximum one are the top reference images most similar to the image to be searched.

3. The image retrieval method according to claim 1, characterized by, The similarity reference curve comprises a similarity; the first-order difference calculation of each similarity in the similarity reference curve comprises: for any similarity in the similarity reference curve, determining an associated similarity adjacent to the similarity in the similarity reference curve; calculating a difference value of the similarity and the associated similarity, and taking the difference value as a first-order difference value corresponding to the similarity in the first-order difference curve; Traverse all similarities, get The first-order difference value corresponding to the similarity is determined The first-order difference curve is composed of the first-order difference values.

4. The image retrieval method according to claim 1, characterized by, the curve segment comprises an internal window curve segment and a window tail curve segment; the sliding on the first-order difference curve using a preset window according to a preset step length comprises: sliding on the first-order difference curve using a preset window according to a preset step length, obtaining a curve segment in the preset window after each sliding stop, and performing standardization calculation on the curve segment to obtain a standardization value corresponding to the curve segment; the sliding on the first-order difference curve using a preset window according to a preset step length comprises: sliding on the first-order difference curve using a preset window according to a preset step length, obtaining a curve segment in the preset window after each sliding stop, and performing standardization calculation on the curve segment to obtain a standardization value corresponding to the curve segment; the standardization calculation on the curve segment to obtain a standardization value corresponding to the curve segment comprises: performing standardization calculation on the internal window curve segment according to the internal window curve segment and the window tail curve segment to obtain a standardization value corresponding to the internal window curve segment.

5. The image retrieval method according to claim 4, characterized in that, the standardization calculation on the internal window curve segment according to the internal window curve segment and the window tail curve segment to obtain a standardization value corresponding to the internal window curve segment comprises: calculating a first mean value of all first-order difference values in the internal window curve segment, calculating a second mean value and a standard deviation of all first-order difference values in the window tail curve segment; performing standardization calculation on the internal window curve segment according to the first mean value, the second mean value and the standard deviation to obtain a standardization value corresponding to the internal window curve segment.

6. The image retrieval method according to claim 5, characterized by, the standardization calculation on the internal window curve segment according to the first mean value, the second mean value and the standard deviation to obtain a standardization value corresponding to the internal window curve segment comprises: calculating a difference value of the first mean value and the second mean value; The absolute value of the calculation result is compared with the standard deviation to obtain a ratio value, and the ratio value is used as the normalized value corresponding to the curve segment in the window.

7. The image retrieval method according to any one of claims 1 to 6, characterized in that, The similarity threshold is used to evaluate the previous... After filtering through several reference images, the image retrieval results include: The former The similarity between each reference image and the image to be retrieved in each of the reference images is compared with the similarity threshold; Reference images corresponding to the similarity greater than the similarity threshold value are reserved, and all reserved reference images are arranged in descending order of similarity to obtain the image retrieval result.

8. An adaptive threshold-based image retrieval apparatus, characterized by comprising: The image retrieval device comprises: The preliminary retrieval module is used to acquire the image to be retrieved and, based on the similarity between images, retrieve the images most similar to the image to be retrieved from a preset image database. One reference image, Integers greater than zero; The difference calculation module is used to calculate the previous... The similarity scores corresponding to each of the reference images are arranged in descending order to form a similarity reference curve. The first-order difference is calculated for each similarity score in the similarity reference curve to obtain a first-order difference curve. The horizontal axis of the similarity reference curve is the sequential identifier of the K reference images, and the vertical axis is the similarity score. a curve normalization module configured to use a preset window to slide on the first-order differential curve according to a preset step length, obtain a curve segment in the preset window after each sliding stop, and perform normalization calculation on the curve segment to obtain a normalized value corresponding to the curve segment; a center point determination module configured to determine a maximum normalized value among all normalized values as a target value, determine a target curve segment used when the target value is normalized, and determine a center point of the target curve segment, wherein the center point is a sequential identifier in the similarity reference curve by multiplying a curve segment width and a preset coefficient, adding a multiplication result and a starting position of the target curve segment, and performing a modulo calculation on an addition result to use a modulo calculation result as the center point; The image filtering module is used to obtain reference images corresponding to the sequential identifiers representing the center points, and to use the similarity between the reference images corresponding to the center points and the images to be searched as a similarity threshold. The similarity threshold is then used to filter the preceding images. The image retrieval results are obtained by filtering through reference images.

9. A computer device, comprising: The computer device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, and the processor implements the image retrieval method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the image retrieval method according to any one of claims 1 to 7.

Images