Image query method, device and equipment

Abstract

The present application provides an image query method, apparatus, and device, the method comprising: acquiring an initial image, the initial image comprising a first region of interest and a first noise region; generating a candidate image based on the initial image, the candidate image comprising a second region of interest and a second noise region; the image features of the second region of interest matching the image features of the first region of interest, and the image features of the second noise region not matching the image features of the first noise region; based on the initial image and the candidate image, querying a target image corresponding to the initial image from an image database; the image database being used to record images with identity information. Through the technical solution of the present application, an accurate target image can be queried based on the features of multiple frames of images to query a target image that matches the initial image. Even if only one frame of the initial image is provided, the target image can be queried based on multiple frames of images, thereby improving the performance of image queries.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an image query method, device and equipment. Background Art

[0002] Image query, also known as image search, refers to the process of searching for a target image that matches an initial image from an image database. The image database is used to record images with identity information. For example, after obtaining the initial image, the features of the initial image (such as color, shape, texture, etc.) are obtained, and the similarity between the features of the initial image and the features of each frame in the image database is calculated. The image with the greatest similarity is then used as the target image that matches the initial image. Since the target image has identity information, the identity information corresponding to the target image is used as the identity information corresponding to the initial image, that is, the initial image is also an image belonging to this identity information, thereby realizing the image query process.

[0003] However, based on the similarity between the features of the initial image and the features of each frame in the image database, when searching for a target image that matches the initial image, the correct target image may not be found. In other words, the wrong target image may be found, resulting in an incorrect image query result. For example, the initial image should match image a1 in the image database, but the query result shows that the initial image matches image a2 in the image database. Summary of the Invention

[0004] The present application provides an image query method, the method comprising:

[0005] Acquire an initial image, wherein the initial image includes a first region of interest and a first noise region;

[0006] generating a candidate image based on the initial image, the candidate image including a second region of interest and a second noise region; image features of the second region of interest match image features of the first region of interest, and image features of the second noise region do not match image features of the first noise region;

[0007] Based on the initial image and the candidate images, a target image corresponding to the initial image is searched from an image database; wherein the image database is used to record images with identity information.

[0008] The present application provides an image query device, comprising:

[0009] An acquisition module is used to acquire an initial image, wherein the initial image includes a first region of interest and a first noise region; a generation module is used to generate a candidate image based on the initial image, wherein the candidate image includes a second region of interest and a second noise region; wherein the image features of the second region of interest match the image features of the first region of interest, and the image features of the second noise region do not match the image features of the first noise region; a query module is used to query an image database for a target image corresponding to the initial image based on the initial image and the candidate image; wherein the image database is used to record images with identity information.

[0010] The present application provides an image query device, comprising: a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions that can be executed by the processor; the processor is used to execute the machine-executable instructions to implement the image query method disclosed in the above example of the present application.

[0011] It can be seen from the above technical solutions that in the embodiment of the present application, a candidate image is generated based on the initial image, and a target image corresponding to the initial image is queried from the image database based on the initial image and the candidate image, that is, based on the features of the multi-frame image (the features of the initial image and the features of the candidate image), the target image matching the initial image is queried, and the accurate target image can be queried, that is, the accurate target image is queried, and the image query result is correct. When using the image query function, even if multiple frames of initial images cannot be provided and only one frame of initial image can be provided, at least one frame of candidate image can be generated based on one frame of initial image, and then the target image can be queried based on the multiple frames of image, which significantly improves the performance of the image query function, enables the query function based on multiple frames of image to play a role, and realizes the multi-image query function. It is possible to generate richer multi-frame candidate images based on one frame of initial image, generate diverse candidate images, enrich the expression form of the query target, realize image query based on multiple frames of image, make the query result more reliable and accurate, and improve the query performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments of the present application or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings of the embodiments of the present application.

[0013] Figure 1 is a flowchart of an image query method in one embodiment of the present application;

[0014] Figure 2is a flowchart of an image query method in one embodiment of the present application;

[0015] Figure 3 is a flowchart of an image query method in one embodiment of the present application;

[0016] Figure 4 This is a schematic diagram of the structure of an image query device in one embodiment of the present application;

[0017] Figure 5 This is a hardware structure diagram of an image query device in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The terms used in the embodiments of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms "a," "the," and "the" used in this application and claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to any or all possible combinations of one or more associated listed items.

[0019] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" used may also be interpreted as "at the time of" or "when" or "in response to determining".

[0020] In the embodiment of the present application, an image query method is proposed, which can also be called an image retrieval method. Figure 1 FIG. 1 is a flow chart of the image query method, which may include:

[0021] Step 101: Acquire an initial image, where the initial image includes a first region of interest and a first noise region.

[0022] Step 102: Generate a candidate image based on the initial image, where the candidate image may include a second region of interest and a second noise region, wherein image features of the second region of interest match image features of the first region of interest, and image features of the second noise region do not match image features of the first noise region.

[0023] Exemplarily, generating a candidate image based on the initial image may include, but is not limited to: determining a first noise region in the initial image based on the acquired generation conditions; adjusting the first noise region in the initial image to obtain a second noise region, and replacing the first noise region in the initial image with the second noise region to obtain a candidate image, i.e., the candidate image includes the second noise region.

[0024] Exemplarily, adjusting the first noise region in the initial image to obtain the second noise region may include, but is not limited to: if the generation condition is a posture generation condition and the first noise region includes the initial posture corresponding to the target object, adjusting the initial posture corresponding to the target object to obtain the target posture corresponding to the target object, and generating the second noise region based on the target posture. If the generation condition is an angle generation condition and the first noise region includes the initial angle corresponding to the target object, adjusting the initial angle corresponding to the target object to obtain the target angle corresponding to the target object, and generating the second noise region based on the target angle. If the generation condition is an illumination generation condition and the first noise region includes the initial illumination, adjusting the initial illumination to obtain the target illumination, and generating the second noise region based on the target illumination. If the generation condition is a background generation condition and the first noise region includes the initial background, adjusting the initial background to obtain the target background, and generating the second noise region based on the target background.

[0025] Exemplarily, before determining the first noise region in the initial image based on the acquired generation conditions, a configured generation condition may also be acquired; alternatively, the initial image may be analyzed, and the generation condition may be determined based on the analysis results. Exemplarily, determining the generation condition based on the analysis results may include, but is not limited to: if the analysis result indicates a need to enhance the expressiveness of the target object's posture features, determining the generation condition to be a posture generation condition; if the analysis result indicates a need to enhance the expressiveness of the target object's angle features, determining the generation condition to be an angle generation condition; if the analysis result indicates a need to enhance the expressiveness of the initial image's illumination features, determining the generation condition to be an illumination generation condition; if the analysis result indicates a need to enhance the expressiveness of the initial image's background features, determining the generation condition to be a background generation condition.

[0026] Step 103: Based on the initial image and the candidate image, query an image database for a target image corresponding to the initial image; wherein the image database is used to record images with identity information.

[0027] In a possible embodiment, based on the initial image and the candidate image, querying the target image corresponding to the initial image from the image database may include but is not limited to: generating a fusion feature based on the features corresponding to the initial image and the features corresponding to the candidate image; querying the target image corresponding to the initial image from the image database based on the similarity between the fusion feature and each image in the image database; or, for each image in the image database, determining the fusion similarity corresponding to the image based on the similarity between the image and the features corresponding to the initial image and the similarity between the image and the features corresponding to the candidate image; querying the target image corresponding to the initial image from the image database based on the fusion similarity corresponding to each image in the image database.

[0028] Exemplarily, the features corresponding to the initial image include first eigenvalues ​​of D feature dimensions, the features corresponding to the candidate image include second eigenvalues ​​of D feature dimensions, and the fused feature includes fused feature values ​​of D feature dimensions. Based on this, generating a fused feature based on the features corresponding to the initial image and the features corresponding to the candidate image includes, but is not limited to: for each feature dimension, determining the average of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension as the fused feature value of the feature dimension; generating the fused feature based on the fused feature values ​​of all feature dimensions; or, for each feature dimension, determining the maximum of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension as the fused feature value of the feature dimension; generating the fused feature based on the fused feature values ​​of all feature dimensions.

[0029] Exemplarily, based on the initial image and the candidate image, querying the target image corresponding to the initial image from the image database may include but is not limited to: determining the target type of the target object in the initial image; based on the initial image and the candidate image, querying the target image corresponding to the initial image from the image database corresponding to the target type; wherein the target type may be a vehicle type, or the target type may be a person type, or the target type may be a commodity type.

[0030] It can be seen from the above technical solutions that in the embodiment of the present application, a candidate image is generated based on the initial image, and a target image corresponding to the initial image is queried from the image database based on the initial image and the candidate image, that is, based on the features of the multi-frame image (the features of the initial image and the features of the candidate image), the target image matching the initial image is queried, and the accurate target image can be queried, that is, the accurate target image is queried, and the image query result is correct. When using the image query function, even if multiple frames of initial images cannot be provided and only one frame of initial image can be provided, at least one frame of candidate image can be generated based on one frame of initial image, and then the target image can be queried based on the multiple frames of image, which significantly improves the performance of the image query function, enables the query function based on multiple frames of image to play a role, and realizes the multi-image query function. It is possible to generate richer multi-frame candidate images based on one frame of initial image, generate diverse candidate images, enrich the expression form of the query target, realize image query based on multiple frames of image, make the query result more reliable and accurate, and improve the query performance.

[0031] The technical solutions of the embodiments of the present application are described below in conjunction with specific embodiments.

[0032] Image query systems, also known as image retrieval systems, are used to implement image query functions. For ease of description, we'll use an image query system as an example. Image query systems typically require maintaining an image database, which records images with identity information. In other words, each image in the database has identity information that represents the relevant attributes of the image.

[0033] For example, if the image database is a vehicle-type image database, then every image in the image database is a vehicle image, and the identity information corresponding to the vehicle image may include, but is not limited to, license plate identification, vehicle color, owner information, vehicle model, and vehicle appearance, without limitation. If the image database is a person-type image database, then every image in the image database is a person image, and the identity information corresponding to the person image may include, but is not limited to, facial features, ID number, mobile phone number, home address, gender, etc., without limitation. If the image database is a commodity-type image database, then every image in the image database is a commodity image, and the identity information corresponding to the commodity image may include, but is not limited to, commodity identification, commodity price, place of origin, commodity quantity, etc., without limitation.

[0034] Exemplarily, in an image query system, image query refers to a process of searching an image database for a target image that matches an initial image (ie, an image input to the image query system by a user).

[0035] For example, a user inputs an initial image into the image query system. After obtaining the initial image, the image query system extracts its features (such as color, shape, and texture), calculates the similarity between the initial image's features and the features of each image in the image database, and selects the image with the greatest similarity as the target image that matches the initial image. Because the target image has identity information, the image query system uses the identity information corresponding to the target image as the identity information corresponding to the initial image, meaning that the initial image also belongs to this identity information, thus completing the image query process.

[0036] However, the image query system may not be able to find the correct target image when searching for a target image that matches the initial image based on the similarity between the features of the initial image and the features of each frame of the image in the image database. That is, the image query system queries for the wrong target image and the image query result is wrong.

[0037] To improve the accuracy of image queries, image queries can be performed based on multiple initial images, significantly improving the performance of the image query system. However, when using the image query system, users are typically unable to provide multiple initial images. In most cases, only a single initial image is provided. This makes it impossible to perform image queries based on multiple initial images, rendering the image query function based on multiple initial images ineffective. This significantly limits the application scope of image queries based on multiple initial images.

[0038] In response to the above findings, an image query method is proposed in an embodiment of the present application. It can generate candidate images based on an initial image and implement image query based on the initial image and candidate images, that is, implement image query based on multiple frames of images. The image query function based on multiple frames of images can play a role, improve the accuracy of image query, significantly improve the performance of the image query system, and can query the accurate target image. When using the image query function, even if multiple frames of initial images cannot be provided and only one frame of initial image can be provided, at least one frame of candidate image can be generated based on the one frame of initial image, and then the target image can be queried based on the multiple frames of images.

[0039] In the above application scenario, an image query method is proposed in the embodiment of the present application. The method can be applied to an image query system, which can also be called an image retrieval system. Figure 2 FIG. 1 is a flow chart of the image query method, which may include the following steps:

[0040] Step 201: Acquire an initial image, where the initial image includes a first region of interest and a first noise region.

[0041] For example, a user may input an image into the image query system. For ease of distinction, the image input by the user is referred to as an initial image, i.e., the image query system may obtain the initial image. Of course, the image query system may also use other methods to obtain the initial image, and the method for obtaining the initial image is not limited.

[0042] For example, the image query system can obtain one initial image or at least two initial images. There is no limit on the number of initial images, and the following uses one initial image as an example. When there are at least two initial images, each initial image is processed in the same manner and will not be further described in this embodiment.

[0043] Exemplarily, for an initial image, the initial image can be divided into a region of interest and a noise region. For ease of distinction, the region of interest in the initial image is referred to as a first region of interest, and the noise region in the initial image is referred to as a first noise region. The first region of interest is an image region that the user is interested in. When generating a candidate image based on the initial image, the first region of interest cannot be adjusted, that is, the first region of interest is an area that will not change. If the first region of interest changes, it will cause an anomaly in the query of the initial image, and the accurate target image cannot be queried. The first noise region is an image region that the user is not interested in. When generating a candidate image based on the initial image, the first noise region can be adjusted, that is, the first noise region is an area that may change. Even if the first noise region changes, it will not cause an anomaly in the query of the initial image, and the accurate target image can still be queried.

[0044] Regarding the first region of interest and the first noise area, the position information of the first region of interest can be configured (for example, when the first region of interest is a rectangular area, the position information may include the coordinates of the upper left corner, the coordinates of the upper right corner, the coordinates of the lower right corner and the coordinates of the lower left corner, or the position information may include the coordinates of any corner point (such as the coordinates of the upper left corner), and the width and height of the rectangular area. Of course, the above is just an example, and there is no limitation on this position information, as long as the first region of interest can be determined based on the position information). Based on the position information of the first region of interest, the first region of interest can be obtained from the initial image, and the remaining area in the initial image except the first region of interest can be used as the first noise area.

[0045] Regarding the first region of interest and the first noise region, the image query system can analyze the initial image to obtain the first region of interest and the first noise region. For example, for an initial image of a person, the face region is the image region of interest to the user and cannot be modified (if the face region is modified, it will become another face, causing an abnormality in the image query). Therefore, the face region can be analyzed from the initial image, and the face region in the initial image is used as the first region of interest, and the remaining region in the initial image other than the first region of interest is used as the first noise region. For another example, for an initial image of a vehicle, the license plate region is the image region of interest to the user and cannot be modified (if the license plate region is modified, it will become another license plate, causing an abnormality in the image query). Therefore, the license plate region can be analyzed from the initial image, and the license plate region in the initial image is used as the first region of interest, and the remaining region in the initial image other than the first region of interest is used as the first noise region.

[0046] Of course, the above is only an example of dividing the first region of interest and the first noise region, and there is no limitation thereto, as long as the initial image can be divided into the first region of interest and the first noise region.

[0047] After the initial image is divided into the first region of interest and the first noise region, when generating a candidate image based on the initial image, the first region of interest cannot be adjusted, that is, the first region of interest is a region that will not change, and the first noise region can be adjusted, that is, the first noise region is a region that may change. Even if the first noise region changes, it will not cause an abnormality in the image query.

[0048] Step 202: Generate a candidate image based on the initial image, the candidate image may include a second region of interest and a second noise region, wherein image features of the second region of interest match image features of the first region of interest, and image features of the second noise region do not match image features of the first noise region.

[0049] For example, the second ROI in the candidate image may be the same as the first ROI in the initial image, and thus the image features of the second ROI may match the image features of the first ROI. A second noise region in the candidate image may be different from the first noise region in the initial image, and thus the image features of the second noise region may not match the image features of the first noise region.

[0050] The initial image includes a first region of interest and a first noise region. The first region of interest is an area that cannot be adjusted, that is, the first region of interest will not change, and the first noise region is an area that can be adjusted, that is, the first noise region can change. On this basis, with respect to step 202, when generating a candidate image based on the initial image, the first region of interest in the initial image is retained unchanged, that is, the first region of interest is not adjusted, and the first noise region in the initial image is adjusted to obtain an adjusted second noise region. The second noise region is used to replace the first noise region in the initial image to obtain a candidate image. Obviously, the candidate image can include a region of interest and a noise region. The region of interest in the candidate image is referred to as a second region of interest, and the noise region in the candidate image is referred to as a second noise region.

[0051] When generating a candidate image based on the initial image, the first region of interest in the initial image is kept unchanged, and the first noise region in the initial image is adjusted. Then, the second region of interest in the candidate image is the same as the first region of interest in the initial image, that is, the image features of the second region of interest match the image features of the first region of interest, and the second noise region in the candidate image is different from the first noise region in the initial image, that is, the image features of the second noise region may not match the image features of the first noise region.

[0052] In one possible embodiment, the first noise region may include multiple regions to be adjusted. For example, for an initial image of a person, the posture of a target object (e.g., a target user) in the first noise region may be adjusted, i.e., the posture of the target object serves as the first noise region to be adjusted. Alternatively, the angle of the target object in the first noise region may be adjusted, i.e., the angle of the target object serves as the first noise region to be adjusted. Alternatively, the lighting in the first noise region may be adjusted, i.e., the lighting serves as the first noise region to be adjusted. Alternatively, the background in the first noise region may be adjusted, i.e., the background serves as the first noise region to be adjusted. For an initial image of a vehicle, the lighting in the first noise region may be adjusted, i.e., the lighting serves as the first noise region to be adjusted. Alternatively, the background in the first noise region may be adjusted, i.e., the background serves as the first noise region to be adjusted. For an initial image of a product, the lighting in the first noise region may be adjusted, i.e., the lighting serves as the first noise region to be adjusted. Alternatively, the background in the first noise region may be adjusted, i.e., the background serves as the first noise region to be adjusted.

[0053] Of course, the above are just a few examples of the first noise area to be adjusted, and there is no limitation thereto.

[0054] In order to determine which type of first noise region to adjust, in this embodiment, generation conditions can be obtained. Based on the obtained generation conditions, a first noise region in the initial image can be determined, the first noise region in the initial image is adjusted to obtain a second noise region, and the first noise region in the initial image is replaced by the second noise region to obtain a candidate image. When generating the candidate image based on the initial image, the first region of interest in the initial image is retained unchanged, and the first noise region in the initial image is adjusted so that the second region of interest in the candidate image is the same as the first region of interest in the initial image, and the second noise region in the candidate image is different from the first noise region in the initial image.

[0055] Exemplarily, in order to obtain the generation condition, the configured generation condition can be obtained. For example, if the pre-configured generation condition is a posture generation condition, the configured posture generation condition is obtained, and the posture generation condition indicates that the posture of the target object in the first noise area is adjusted, that is, the posture of the target object is used as the first noise area to be adjusted. Alternatively, if the pre-configured generation condition is an angle generation condition, the configured angle generation condition is obtained, and the angle generation condition indicates that the angle of the target object in the first noise area is adjusted, that is, the angle of the target object is used as the first noise area to be adjusted. Alternatively, if the pre-configured generation condition is a lighting generation condition, the configured lighting generation condition is obtained, and the lighting generation condition indicates that the lighting in the first noise area is adjusted, that is, the lighting is used as the first noise area to be adjusted. Alternatively, if the pre-configured generation condition is a background generation condition, the configured background generation condition is obtained, and the background generation condition indicates that the background in the first noise area is adjusted, that is, the background is used as the first noise area to be adjusted.

[0056] Exemplarily, to obtain the generation condition, the initial image can be analyzed, and the generation condition corresponding to the initial image can be determined based on the analysis result. For example, if the analysis result indicates a need to enhance the expressiveness of the target object's posture characteristics (e.g., the posture of the target object in the initial image does not meet a preset posture requirement, and no restrictions are placed on this preset posture requirement), the generation condition is determined to be a posture generation condition, and the posture generation condition indicates adjusting the posture of the target object in the first noise region, i.e., the posture of the target object serves as the first noise region to be adjusted. Alternatively, if the analysis result indicates a need to enhance the expressiveness of the target object's angle characteristics (e.g., the angle of the target object in the initial image does not meet a preset angle requirement, and no restrictions are placed on this preset angle requirement), the generation condition is determined to be an angle generation condition, and the angle generation condition indicates adjusting the angle of the target object in the first noise region, i.e., the angle of the target object serves as the first noise region to be adjusted. Alternatively, if the analysis result indicates a need to enhance the expressiveness of the initial image's illumination characteristics (e.g., the illumination in the initial image does not meet a preset illumination requirement, and no restrictions are placed on this preset illumination requirement), the generation condition is determined to be an illumination generation condition, and the illumination generation condition indicates adjusting the illumination in the first noise region, i.e., the illumination serves as the first noise region to be adjusted. Alternatively, if the analysis result indicates that the background expression ability of the initial image needs to be enhanced (for example, the background in the initial image does not meet the preset background requirements, and no restrictions are placed on the preset background requirements), the generation condition is determined to be a background generation condition, and the background generation condition indicates that the background in the first noise area is adjusted, that is, the background is used as the first noise area to be adjusted.

[0057] Of course, the posture generation condition, angle generation condition, illumination generation condition, and background generation condition are merely examples of generation conditions and are not intended to be limiting. Other types of generation conditions may also be used. The above-mentioned method for obtaining the generation conditions is also merely an example and is not intended to be limiting. Exemplarily, the generation condition may be one of the posture generation condition, angle generation condition, illumination generation condition, and background generation condition, or may be at least two of the posture generation condition, angle generation condition, illumination generation condition, and background generation condition. For example, the configured generation conditions may be the posture generation condition and the illumination generation condition, and the generation conditions determined based on the analysis results may be the posture generation condition, the illumination generation condition, and the background generation condition, and so on.

[0058] Exemplarily, based on the acquired generation condition, a first noise region corresponding to the generation condition can be determined in the initial image, and the first noise region in the initial image is adjusted to obtain a second noise region. For example, if the generation condition is a posture generation condition, the posture of the target object in the first noise region (for convenience of distinction, this posture is referred to as the initial posture) needs to be adjusted. Therefore, the initial posture corresponding to the target object is determined from the first noise region, that is, the first noise region includes the initial posture corresponding to the target object, the initial posture corresponding to the target object is adjusted (there is no restriction on the adjustment method, as long as the initial posture is changed), and the target posture corresponding to the target object (that is, the adjusted posture corresponding to the initial posture) is obtained. The second noise region is generated based on the target posture, that is, the initial posture in the first noise region is replaced by the target posture, and the replaced first noise region is used as the second noise region.

[0059] For another example, if the generation condition is an angle generation condition, it is necessary to adjust the angle of the target object in the first noise area (for convenience of distinction, this angle is referred to as the initial angle). Therefore, the initial angle corresponding to the target object is determined from the first noise area, that is, the first noise area includes the initial angle corresponding to the target object, and the initial angle corresponding to the target object is adjusted (there is no restriction on the adjustment method, as long as the initial angle is changed) to obtain the target angle corresponding to the target object (that is, the adjusted angle corresponding to the initial angle), and the second noise area is generated based on the target angle, that is, the initial angle in the first noise area is replaced by the target angle, and the replaced first noise area is used as the second noise area.

[0060] For another example, if the generation condition is an illumination generation condition, the illumination in the first noise area (for the sake of convenience, the illumination is referred to as initial illumination) needs to be adjusted. Therefore, the initial illumination is determined from the first noise area, that is, the first noise area includes the initial illumination, and the initial illumination is adjusted (there is no restriction on the adjustment method, as long as the initial illumination is changed) to obtain the target illumination (that is, the adjusted illumination corresponding to the initial illumination), and the second noise area is generated based on the target illumination, that is, the initial illumination in the first noise area is replaced by the target illumination, and the replaced first noise area is used as the second noise area.

[0061] For another example, if the generation condition is a background generation condition, the background in the first noise area (for convenience of distinction, the background is referred to as the initial background) needs to be adjusted. Therefore, the initial background is determined from the first noise area, that is, the first noise area includes the initial background, and the initial background is adjusted (there is no restriction on the adjustment method, as long as the initial background is changed) to obtain the target background (that is, the adjusted background corresponding to the initial background), and the second noise area is generated based on the target background, that is, the initial background in the first noise area is replaced by the target background, and the replaced first noise area is used as the second noise area.

[0062] Exemplarily, if the generation conditions are at least two of a posture generation condition, an angle generation condition, an illumination generation condition, and a background generation condition, then at least two adjustments can be made to the first noise region in the initial image to obtain an adjusted second noise region. For example, if the generation conditions are a posture generation condition and an illumination generation condition, then the initial posture corresponding to the target object is determined from the first noise region, and the initial posture corresponding to the target object is adjusted to obtain a target posture corresponding to the target object. Furthermore, the initial illumination is determined from the first noise region, and the initial illumination is adjusted to obtain a target illumination. Based on this, the initial posture in the first noise region is replaced by the target posture, and the initial illumination in the first noise region is replaced by the target illumination, and the replaced first noise region serves as the second noise region.

[0063] In summary, the first noise region in the initial image can be adjusted to obtain a second noise region. Based on this, the first noise region in the initial image can be replaced by the second noise region to obtain a candidate image. Clearly, the second ROI in the candidate image is identical to the first ROI in the initial image, and the second noise region in the candidate image is different from the first noise region in the initial image.

[0064] For example, when generating candidate images based on the initial image, at least one frame of candidate images may be generated, such as two frames of candidate images, three frames of candidate images, four frames of candidate images, five frames of candidate images, etc. There is no limit on the number of candidate images. For example, if three frames of candidate images need to be generated, then:

[0065] If the generation condition is a posture generation condition, the initial posture corresponding to the target object is adjusted to target posture 1, the initial posture in the first noise area is replaced by the target posture 1, and the second noise area 1 is obtained, and the first noise area in the initial image is replaced by the second noise area 1 to obtain candidate image 1; the initial posture corresponding to the target object is adjusted to target posture 2, the initial posture in the first noise area is replaced by the target posture 2, and the second noise area 2 is used to replace the first noise area in the initial image to obtain candidate image 2; the initial posture corresponding to the target object is adjusted to target posture 3, the initial posture in the first noise area is replaced by the target posture 3, and the second noise area 3 is obtained, and the first noise area in the initial image is replaced by the second noise area 3 to obtain candidate image 3.

[0066] By analogy, if the generation condition is a background generation condition, the initial background can be adjusted to target background 1, and the initial background in the first noise area can be replaced by target background 1 to obtain second noise area 1, and the first noise area in the initial image can be replaced by the second noise area 1 to obtain candidate image 1; the initial background can be adjusted to target background 2 (different from target background 1), and the initial background in the first noise area can be replaced by target background 2 to obtain second noise area 2, and the first noise area in the initial image can be replaced by the second noise area 2 to obtain candidate image 2; the initial background can be adjusted to target background 3, and the initial background in the first noise area can be replaced by target background 3 to obtain second noise area 3, and the first noise area in the initial image can be replaced by the second noise area 3 to obtain candidate image 3.

[0067] To summarize, in step 202, at least one candidate image can be generated based on the initial image, that is, candidate images of the target object in the initial image under different generation conditions are generated, such as candidate images with different postures, different angles, different lighting or different backgrounds. The candidate image does not change the identity information of the target object in the initial image (the identity information is ensured to remain unchanged by the second region of interest being the same as the first region of interest). The candidate image can enhance the feature expression ability of the initial image (the feature expression ability is enhanced by the difference between the second noise region and the first noise region), enhance the feature robustness, and obtain better query results.

[0068] For example, given an initial image with background A, dim lighting, and facing forward, we can generate candidate image 1 with background B (i.e., the initial background is adjusted to the target background), normal lighting (i.e., the initial lighting is adjusted to the target lighting), and facing forward. We can also generate candidate image 2 with background B, normal lighting, and facing sideways (i.e., the initial angle is adjusted to the target angle). We can also generate candidate image 3 with background C (i.e., the initial background is adjusted to the target background), normal lighting, and facing forward. We can also generate candidate image 4 with background C, normal lighting, and facing sideways. We can also generate candidate image 5 with background A, normal lighting, and facing forward. In summary, based on one frame of the original image, we can obtain five frames of candidate images. Of course, the above five frames of candidate images are just examples and are not limiting.

[0069] For example, if the user is concerned about whether the target object appears in a specific background (i.e., a specific scene) and the user can only provide an initial image with a non-specific background, then the non-specific background in the initial image can be replaced with a specific background to obtain candidate images with the specific background, that is, generate candidate images for image query. By generating candidate images with a specific background for query, the interference of background information can be reduced, so that when calculating the similarity of the target object, more attention is paid to the foreground target, that is, the target object itself.

[0070] Step 203: Generate fusion features based on the features corresponding to the initial image and the features corresponding to the candidate images.

[0071] For example, features corresponding to the initial image and features corresponding to at least one candidate image are fused into a fused feature for querying. For example, a mean fusion method or a maximum fusion method is used to fuse the features corresponding to the initial image and features corresponding to at least one candidate image into a fused feature.

[0072] Exemplarily, the features corresponding to the initial image include first eigenvalues ​​of D feature dimensions, the features corresponding to the candidate image include second eigenvalues ​​of D feature dimensions, and the fused feature includes fused feature values ​​of D feature dimensions. Based on this, for each feature dimension, the average of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension can be determined as the fused feature value of the feature dimension; and the fused feature is generated based on the fused feature values ​​of all feature dimensions. Alternatively, for each feature dimension, the maximum value of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension can be determined as the fused feature value of the feature dimension; and the fused feature is generated based on the fused feature values ​​of all feature dimensions.

[0073] For example, an example of the calculation formula for fusion features can be seen in the following formula:

[0074]

[0075] In the above formula, N represents the total number of images (i.e., the total number of initial images and candidate images), n ranges from 1 to N, representing the nth frame image (which can be the initial image or the candidate image), D represents the total number of feature dimensions, and i ranges from 1 to D, representing the i-th feature dimension. i Q represents the fusion feature value of the i-th feature dimension, f i n Represents the eigenvalue of the i-th feature dimension of the n-th frame image (i.e., the first eigenvalue in the initial image, or the second eigenvalue in the candidate image). Based on the above formula, for feature dimension i, the average of the eigenvalues ​​of feature dimension i corresponding to all N frames of image can be used as the fused feature value of feature dimension i, and the fused feature values ​​of all D feature dimensions constitute the fused feature.

[0076] Of course, the above formula takes the mean fusion method as an example. When the maximum fusion method is adopted, the fusion feature value of each feature dimension can be taken as the maximum value of the feature value of the feature dimension corresponding to N frames of images.

[0077] Step 204: Based on the similarity between the fusion feature and each image in the image database (used to record images with identity information), query the image database for a target image corresponding to the initial image.

[0078] For example, the similarity between the fused feature and the features of each image in the image database is calculated, and there is no restriction on the method for calculating this similarity. Based on the similarity between the fused feature and the features of each image, the maximum similarity can be selected from all similarities, and the image corresponding to the maximum similarity (i.e., the image in the image database) is used as the target image corresponding to the initial image.

[0079] For another example, the similarity between the fused feature and the feature of each image in the image database is calculated, and based on the similarity between the fused feature and the feature of each image, the images in the image database are sorted in descending order of similarity, and based on the sorting result, the top K (K is a positive integer) images (i.e., the images in the image database) are selected as target images corresponding to the initial image.

[0080] To sum up, the target image corresponding to the initial image can be obtained. Since the target image has identity information, the image query system uses the identity information corresponding to the target image as the identity information corresponding to the initial image, that is, the initial image is also an image belonging to this identity information, thereby realizing the image query process.

[0081] In one possible implementation, the target type of the target object in the initial image may be determined, and a target image corresponding to the initial image may be retrieved from an image database corresponding to the target type. For example, based on the similarity between the fused feature and each image in the image database corresponding to the target type, a target image corresponding to the initial image may be retrieved from the image database corresponding to the target type.

[0082] The target type may be a vehicle type, a person type, or a commodity type. Of course, the above are just a few examples of target types and are not limiting. For example, if the target type is a vehicle type, then the image database corresponding to the target type is an image database for vehicle types, and each image in the image database is a vehicle image. If the target type is a person type, then the image database corresponding to the target type is an image database for person types, and each image in the image database is a person image. If the target type is a commodity type, then the image database corresponding to the target type is an image database for commodity types, and each image in the image database is a commodity image.

[0083] In the embodiment of the present application, an image query method is proposed, which can be applied to an image query system. Figure 3 FIG. 1 is a flow chart of the image query method, which may include the following steps:

[0084] Step 301: Acquire an initial image, where the initial image includes a first region of interest and a first noise region.

[0085] Step 302: Generate a candidate image based on the initial image, where the candidate image may include a second region of interest and a second noise region, wherein image features of the second region of interest match image features of the first region of interest, and image features of the second noise region do not match image features of the first noise region.

[0086] For example, steps 301 to 302 may refer to steps 201 to 202 and are not described in detail here.

[0087] Step 303: For each image in the image database (the image database is used to record images with identity information), determine the fusion similarity corresponding to the image based on the similarity between the features corresponding to the image and the initial image and the similarity between the features corresponding to the image and the candidate image.

[0088] For example, the features corresponding to the initial image and the features corresponding to at least one candidate image are combined into a feature set. This feature set includes the features corresponding to the initial image and the features corresponding to each candidate image. For each image in the image database, the similarity between the image's features and each feature in the feature set is calculated. Then, the average of these similarities is calculated as the fused similarity corresponding to the image.

[0089] For example, an example of the calculation formula for fusion similarity can be seen in the following formula:

[0090]

[0091] In the above formula, N represents the total number of images (i.e. the total number of initial images and candidate images), and the value range of n is 1-N, which represents the nth frame image (which can be the initial image or the candidate image). It represents the fusion similarity between the m-th image in the image database and all the features in the feature set, that is, the fusion similarity corresponding to the m-th image in the image database. It represents the similarity between the n-th frame image (i.e., the n-th feature in the feature set) and the m-th image in the image database.

[0092] Step 304: Based on the fusion similarity corresponding to each image in the image database, a target image corresponding to the initial image is searched from the image database. For example, the image with the greatest fusion similarity (i.e., the image in the image database) is used as the target image corresponding to the initial image. In another example, the images in the image database are sorted in descending order of fusion similarity, and the top K images (i.e., the images in the image database) are selected based on the sorting results as the target images corresponding to the initial image.

[0093] To sum up, the target image corresponding to the initial image can be obtained. Since the target image has identity information, the image query system uses the identity information corresponding to the target image as the identity information corresponding to the initial image, that is, the initial image is also an image belonging to this identity information, thereby realizing the image query process.

[0094] In a possible implementation, the target type of the target object in the initial image can also be determined, and the target image corresponding to the initial image can be queried from the image database corresponding to the target type. For example, for each image in the image database corresponding to the target type, the fusion similarity corresponding to the image is determined based on the similarity between the features corresponding to the image and the initial image, and the similarity between the features corresponding to the image and the candidate image. Based on the fusion similarity corresponding to each image in the image database corresponding to the target type, the target image corresponding to the initial image is queried from the image database corresponding to the target type. The target type can be a vehicle type, or a person type, or a commodity type. Of course, the above are just examples of target types and there is no limitation to this.

[0095] It can be seen from the above technical solutions that in the embodiment of the present application, the target image that matches the initial image can be queried based on the features of the multi-frame images, and the accurate target image can be queried, and the query result of the image is correct. When using the image query function, even if it is impossible to provide multiple frames of initial images and only one frame of initial image can be provided, at least one frame of candidate image can be generated based on the one frame of initial image, and then the target image can be queried based on the multiple frames of image, which significantly improves the performance of the image query function, enables the query function based on the multiple frames of image to play a role, and realizes the multi-image query function. It can generate richer multi-frame candidate images based on one frame of initial image, generate diverse candidate images, enrich the expression form of the query target, realize image query based on multi-frame images, make the query result more reliable and accurate, and the retrieval result of the query feature set formed is more robust, which expands the application scope of the technology based on multi-image query and improves the query performance of the image query system.

[0096] Based on the same application concept as the above method, an image query device is proposed in the embodiment of the present application. Figure 4 FIG. 1 is a schematic diagram of the structure of the image query device, which may include:

[0097] An acquisition module 41 is used to acquire an initial image, wherein the initial image includes a first region of interest and a first noise region; a generation module 42 is used to generate a candidate image based on the initial image, wherein the candidate image includes a second region of interest and a second noise region; wherein the image features of the second region of interest match the image features of the first region of interest, and the image features of the second noise region do not match the image features of the first noise region; a query module 43 is used to query an image database for a target image corresponding to the initial image based on the initial image and the candidate image; wherein the image database is used to record images with identity information.

[0098] Exemplarily, when the generation module generates a candidate image based on the initial image, it is specifically used to: determine a first noise area in the initial image based on the acquired generation conditions, and adjust the first noise area in the initial image to obtain a second noise area; and replace the first noise area in the initial image with the second noise area to obtain the candidate image.

[0099] Exemplarily, the generation module 42 adjusts the first noise area in the initial image to obtain the second noise area, specifically for: if the generation condition is a posture generation condition, and the first noise area includes the initial posture corresponding to the target object, then adjusting the initial posture corresponding to the target object to obtain the target posture corresponding to the target object, and generating the second noise area based on the target posture; if the generation condition is an angle generation condition, and the first noise area includes the initial angle corresponding to the target object, then adjusting the initial angle corresponding to the target object to obtain the target angle corresponding to the target object, and generating the second noise area based on the target angle; if the generation condition is an illumination generation condition, and the first noise area includes initial illumination, then adjusting the initial illumination to obtain target illumination, and generating the second noise area based on the target illumination; if the generation condition is a background generation condition, and the first noise area includes an initial background, then adjusting the initial background to obtain a target background, and generating the second noise area based on the target background.

[0100] Exemplarily, the acquisition module 41 is also used to obtain the configured generation conditions; or, to analyze the initial image and determine the generation conditions based on the analysis results; wherein, when the acquisition module determines the generation conditions based on the analysis results, it is specifically used to: if the analysis result indicates that it is necessary to enhance the posture feature expression capability of the target object, then determine that the generation condition is a posture generation condition; or, if the analysis result indicates that it is necessary to enhance the angle feature expression capability of the target object, then determine that the generation condition is an angle generation condition; or, if the analysis result indicates that it is necessary to enhance the lighting feature expression capability of the initial image, then determine that the generation condition is a lighting generation condition; or, if the analysis result indicates that it is necessary to enhance the background expression capability of the initial image, then determine that the generation condition is a background generation condition.

[0101] Exemplarily, when the query module 43 queries the target image corresponding to the initial image from the image database based on the initial image and the candidate image, it is specifically used to: generate a fusion feature based on the features corresponding to the initial image and the features corresponding to the candidate image; based on the similarity between the fusion feature and each image in the image database, query the target image corresponding to the initial image from the image database; or, for each image in the image database, determine the fusion similarity corresponding to the image based on the similarity between the image and the features corresponding to the initial image and the similarity between the image and the features corresponding to the candidate image; based on the fusion similarity corresponding to each image in the image database, query the target image corresponding to the initial image from the image database.

[0102] Exemplarily, the features corresponding to the initial image include first eigenvalues ​​of D feature dimensions, the features corresponding to the candidate image include second eigenvalues ​​of D feature dimensions, and the fused features include fused feature values ​​of D feature dimensions; when the query module 43 generates fused features based on the features corresponding to the initial image and the features corresponding to the candidate image, it is specifically used to: for each feature dimension, determine the average value of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension as the fused feature value of the feature dimension; generate fused features based on the fused feature values ​​of all feature dimensions; or, for each feature dimension, determine the maximum value of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension as the fused feature value of the feature dimension; generate fused features based on the fused feature values ​​of all feature dimensions.

[0103] Exemplarily, when the query module 43 queries the target image corresponding to the initial image from the image database based on the initial image and the candidate image, it is specifically used to: determine the target type of the target object in the initial image; based on the initial image and the candidate image, query the target image corresponding to the initial image from the image database corresponding to the target type; wherein the target type is a vehicle type, or the target type is a person type, or the target type is a commodity type.

[0104] Based on the same application concept as the above method, an image query device is proposed in the embodiment of the present application. Figure 5 As shown, the image query device includes: a processor 51 and a machine-readable storage medium 52, the machine-readable storage medium 52 stores machine-executable instructions that can be executed by the processor 51; the processor 51 is used to execute the machine-executable instructions to implement the image query method disclosed in the above example of this application.

[0105] Based on the same application concept as the above method, an embodiment of the present application further provides a machine-readable storage medium, on which a number of computer instructions are stored. When the computer instructions are executed by a processor, the image query method disclosed in the above example of the present application can be implemented.

[0106] The machine-readable storage medium may be any electronic, magnetic, optical, or other physical storage device that may contain or store information, such as executable instructions, data, and the like. For example, the machine-readable storage medium may be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, a storage drive (such as a hard disk drive), a solid-state drive, any type of storage disk (such as a CD, DVD, etc.), or similar storage media, or a combination thereof.

[0107] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email transceiver, game console, tablet computer, wearable device, or any combination of these devices.

[0108] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0109] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0110] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] Furthermore, these computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0113] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. An image query method, characterized in that: The method comprises: Acquire an initial image, wherein the initial image includes a first region of interest and a first noise region; generating a candidate image based on the initial image, the candidate image including a second region of interest and a second noise region; image features of the second region of interest match image features of the first region of interest, and image features of the second noise region do not match image features of the first noise region; Based on the initial image and the candidate image, querying an image database for a target image corresponding to the initial image; wherein the image database is used to record images with identity information; The step of querying an image database for a target image corresponding to the initial image based on the initial image and the candidate image comprises: generating a fusion feature based on features corresponding to the initial image and features corresponding to the candidate image; and querying the image database for a target image corresponding to the initial image based on similarities between the fusion feature and each image in the image database. Alternatively, for each image in the image database, a fusion similarity corresponding to the image is determined based on similarities between features corresponding to the image and the initial image and similarities between features corresponding to the image and the candidate image; based on the fusion similarities corresponding to each image in the image database, a target image corresponding to the initial image is searched from the image database; The step of generating a candidate image based on the initial image includes: determining a first noise region in the initial image based on the acquired generation condition; The first noise region in the initial image is adjusted to obtain a second noise region, and the first noise region in the initial image is replaced by the second noise region to obtain the candidate image.

2. The method according to claim 1, characterized in that The adjusting the first noise region in the initial image to obtain the second noise region includes: If the generation condition is a posture generation condition, and the first noise region includes an initial posture corresponding to the target object, adjusting the initial posture corresponding to the target object to obtain a target posture corresponding to the target object, and generating a second noise region based on the target posture; If the generation condition is an angle generation condition, and the first noise area includes an initial angle corresponding to the target object, adjusting the initial angle corresponding to the target object to obtain a target angle corresponding to the target object, and generating a second noise area based on the target angle; If the generation condition is a lighting generation condition, and the first noise region includes initial lighting, adjusting the initial lighting to obtain target lighting, and generating a second noise region based on the target lighting; If the generation condition is a background generation condition and the first noise region includes an initial background, the initial background is adjusted to obtain a target background, and the second noise region is generated based on the target background.

3. The method according to claim 1 or 2, characterized in that Before determining the first noise region in the initial image based on the acquired generation condition, the method further includes: Obtaining configured generation conditions; or analyzing the initial image and determining the generation conditions based on the analysis results; wherein determining the generation conditions based on the analysis results includes: If the analysis result indicates that the posture feature expression capability of the target object needs to be enhanced, the generation condition is determined to be the posture generation condition; if the analysis result indicates that the angle feature expression capability of the target object needs to be enhanced, the generation condition is determined to be the angle generation condition; if the analysis result indicates that the illumination feature expression capability of the initial image needs to be enhanced, the generation condition is determined to be the illumination generation condition; if the analysis result indicates that the background expression capability of the initial image needs to be enhanced, the generation condition is determined to be the background generation condition.

4. The method according to claim 1, wherein The features corresponding to the initial image include first eigenvalues ​​of D feature dimensions, the features corresponding to the candidate image include second eigenvalues ​​of D feature dimensions, and the fused features include fused feature values ​​of D feature dimensions; generating the fused features based on the features corresponding to the initial image and the features corresponding to the candidate image includes: For each feature dimension, the average value of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension is determined as the fused feature value of the feature dimension; the fused feature is generated based on the fused feature values ​​of all feature dimensions; or, for each feature dimension, the maximum value of the first eigenvalue of the feature dimension and the second eigenvalue of the feature dimension is determined as the fused feature value of the feature dimension; the fused feature is generated based on the fused feature values ​​of all feature dimensions.

5. The method according to claim 1, wherein The step of querying an image database for a target image corresponding to the initial image based on the initial image and the candidate image includes: Determining a target type of a target object in an initial image; querying a target image corresponding to the initial image from an image database corresponding to the target type based on the initial image and the candidate image; The target type is a vehicle type, a person type, or a commodity type.

6. An image query device, characterized in that: The device comprises: an acquisition module configured to acquire an initial image, the initial image including a first region of interest and a first noise region; a generation module configured to generate a candidate image based on the initial image, the candidate image including a second region of interest and a second noise region; wherein image features of the second region of interest match image features of the first region of interest, and image features of the second noise region do not match image features of the first noise region; and a query module configured to query an image database for a target image corresponding to the initial image based on the initial image and the candidate image; wherein the image database is configured to record images having identity information; The generating module, when generating the candidate image based on the initial image, is specifically configured to: determine a first noise region in the initial image based on the acquired generating conditions, and adjust the first noise region in the initial image to obtain a second noise region; and replace the first noise region in the initial image with the second noise region to obtain the candidate image; Among them, when the query module queries the target image corresponding to the initial image from the image database based on the initial image and the candidate image, it is specifically used to: generate a fusion feature based on the features corresponding to the initial image and the features corresponding to the candidate image; based on the similarity between the fusion feature and each image in the image database, query the target image corresponding to the initial image from the image database; or, for each image in the image database, determine the fusion similarity corresponding to the image based on the similarity between the image and the features corresponding to the initial image and the similarity between the image and the features corresponding to the candidate image; based on the fusion similarity corresponding to each image in the image database, query the target image corresponding to the initial image from the image database.

7. The device according to claim 6, It is characterized in that Wherein, the generating module adjusts the first noise region in the initial image to obtain the second noise region, specifically for: if the generating condition is a posture generating condition, and the first noise region includes an initial posture corresponding to a target object, adjusting the initial posture corresponding to the target object to obtain a target posture corresponding to the target object, and generating the second noise region based on the target posture; if the generating condition is an angle generating condition, and the first noise region includes an initial angle corresponding to the target object, adjusting the initial angle corresponding to the target object to obtain a target angle corresponding to the target object, and generating the second noise region based on the target angle; if the generating condition is an illumination generating condition, and the first noise region includes initial illumination, adjusting the initial illumination to obtain target illumination, and generating the second noise region based on the target illumination; if the generating condition is a background generating condition, and the first noise region includes an initial background, adjusting the initial background to obtain a target background, and generating the second noise region based on the target background; Wherein, the acquisition module is further used to acquire the configured generation conditions; or, analyze the initial image and determine the generation conditions based on the analysis results; wherein, when determining the generation conditions based on the analysis results, the acquisition module is specifically used to: if the analysis result indicates that the posture feature expression capability of the target object needs to be enhanced, then determine that the generation condition is a posture generation condition; or, if the analysis result indicates that the angle feature expression capability of the target object needs to be enhanced, then determine that the generation condition is an angle generation condition; or, if the analysis result indicates that the illumination feature expression capability of the initial image needs to be enhanced, then determine that the generation condition is an illumination generation condition; or, if the analysis result indicates that the background expression capability of the initial image needs to be enhanced, then determine that the generation condition is a background generation condition; The features corresponding to the initial image include first feature values ​​of D feature dimensions, the features corresponding to the candidate image include second feature values ​​of D feature dimensions, and the fused features include fused feature values ​​of D feature dimensions; when the query module generates the fused features based on the features corresponding to the initial image and the features corresponding to the candidate image, it is specifically used to: for each feature dimension, determine the average value of the first feature value of the feature dimension and the second feature value of the feature dimension as the fused feature value of the feature dimension; generate the fused features based on the fused feature values ​​of all feature dimensions; or, for each feature dimension, determine the maximum value of the first feature value of the feature dimension and the second feature value of the feature dimension as the fused feature value of the feature dimension; generate the fused features based on the fused feature values ​​of all feature dimensions; Among them, the query module is specifically used to query the target image corresponding to the initial image from the image database based on the initial image and the candidate image: determine the target type of the target object in the initial image; based on the initial image and the candidate image, query the target image corresponding to the initial image from the image database corresponding to the target type; wherein the target type is a vehicle type, or the target type is a person type, or the target type is a commodity type.

8. An image query device, characterized in that: include: a processor and a machine-readable storage medium storing machine-executable instructions capable of being executed by the processor; The processor is configured to execute machine-executable instructions to implement the method steps described in any one of claims 1-5.

Images