Image deduplication method and device, equipment, medium and product

Abstract

The embodiment of the invention provides an image deduplication method and device, equipment, a medium and a product, and the method specifically comprises the steps: obtaining at least two first images and the image information of each first image, the image information comprising at least one of image feature information and image semantic information; for every two first images in the at least two first images, calculating the image similarity of the two first images based on the image information of the two first images; and under the condition that the image similarity of the two first images is greater than a preset similarity threshold value, performing de-duplication processing on the two first images so as to acquire at least one different second image from the at least two first images. According to the embodiment of the invention, the accuracy of image deduplication is improved, so that the storage space of a cloud disk can be saved.

Description

Technical Field

[0001] The present application belongs to the technical field of image deduplication, and in particular, relates to an image deduplication method, device, equipment, medium and product. Background Art

[0002] With the popularization of Internet technology, users have higher and higher demands for image management. Therefore, in real life, users can store images in the cloud through cloud disks so that they can access them anytime and anywhere, but the storage space of cloud disks is often limited.

[0003] Based on this, in order to save storage space of the cloud disk, the prior art generally performs deduplication processing based on the image name. However, since different images have the same name, or the same image has different names, the accuracy of image deduplication is poor. Summary of the invention

[0004] The embodiments of the present application provide an image deduplication method, apparatus, device, medium and product, which improve the accuracy of image deduplication and thus save the storage space of a cloud disk.

[0005] In a first aspect, an embodiment of the present application provides an image deduplication method, which is applied to a base station, and the method includes:

[0006] Acquire at least two first images and image information of each first image, where the image information includes at least one of image feature information and image semantic information;

[0007] For each two first images of the at least two first images, based on image information of the two first images, calculating image similarity between the two first images;

[0008] When the image similarity between the two first images is greater than a preset similarity threshold, deduplication processing is performed on the two first images to obtain at least one different second image from the at least two first images.

[0009] In a second aspect, an embodiment of the present application provides an image deduplication device, the device comprising:

[0010] An acquisition module, used to acquire at least two first images and image information of each first image, where the image information includes at least one of image feature information and image semantic information;

[0011] a calculation module, configured to calculate, for each two first images of the at least two first images, an image similarity of the two first images based on image information of the two first images;

[0012] The deduplication module is used to perform deduplication processing on the two first images when the image similarity of the two first images is greater than a preset similarity threshold, so as to obtain at least one different second image from the at least two first images.

[0013] In a third aspect, an electronic device is provided, comprising: a memory for storing computer program instructions; and a processor for reading and running the computer program instructions stored in the memory to execute the image deduplication method provided by any optional implementation in the first aspect.

[0014] In a fourth aspect, a computer storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the image deduplication method provided by any optional implementation of the first aspect is implemented.

[0015] In a fifth aspect, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, the image deduplication method provided by any optional implementation in the first aspect is implemented.

[0016] In an embodiment of the present application, by acquiring at least two first images and image information of each first image, the image information may include at least one of image feature information and image semantic information, and then for each two first images of the at least two first images, based on the image information of the first image, the image similarity of the two first images can be calculated, and when the image similarity is greater than a preset similarity threshold, that is, it indicates that the two first images are the same image, the two first images are deduplicated, so as to obtain at least one different second image from the at least two first images. In this way, compared with the prior art method of deduplicating based only on the image name, the present application can accurately and effectively determine whether two images are the same image based on the features of the image at the physical level and the semantics at the visual level, and then can accurately and effectively perform deduplication, thereby improving the accuracy of image deduplication, thereby saving storage space of the cloud disk. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 This is one of the flowcharts of an image deduplication method provided in an embodiment of the present application;

[0019] Figure 2 This is a second flow chart of an image deduplication method provided in an embodiment of the present application;

[0020] Figure 3 This is a third flow chart of an image deduplication method provided in an embodiment of the present application;

[0021] Figure 4 is a structural schematic diagram of an image denoising model provided in an embodiment of the present application;

[0022] Figure 5 is a structural schematic diagram of an image deduplication device provided in an embodiment of the present application;

[0023] Figure 6 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0025] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0026] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0027] As described in the background technology, in order to solve the problem of poor accuracy of image deduplication in the prior art, the embodiments of the present application provide an image deduplication method, device, equipment, medium and product, which can obtain at least two first images and image information of each first image, and the image information can include at least one of image feature information and image semantic information, and then for each of the at least two first images, based on the image information of the first image, the image similarity of the two first images can be calculated, and when the image similarity is greater than a preset similarity threshold, it indicates that the two first images are the same image, and the two first images are deduplicated, so as to achieve at least one different second image from the at least two first images. In this way, compared with the method of deduplication based on image name only in the prior art, the present application can accurately and effectively determine whether two images are the same image based on the features of the image at the physical level and the semantics at the visual level, and then can accurately and effectively perform deduplication, improve the accuracy of image deduplication, and thus save the storage space of the cloud disk.

[0028] It should be noted that the execution subject of the image deduplication method provided in the embodiment of the present application can be an image deduplication device, or a control module in the image deduplication device for executing the image deduplication method. In the embodiment of the present application, the image deduplication method provided in the embodiment of the present application is explained by taking the image deduplication device executing the image deduplication method as an example.

[0029] Figure 1 This is one of the flow charts of an image deduplication method provided in an embodiment of the present application.

[0030] like Figure 1 As shown, the execution subject of the method may be an image deduplication device, and the method may specifically include the following steps:

[0031] S110: Acquire at least two first images and image information of each first image.

[0032] The at least two first images mentioned above may be related images for uploading and storing in a cloud disk, which is not specifically limited here.

[0033] In addition, the image information of each first image may include at least one of image feature information and image semantic information. The image feature information may be a feature of the image itself that is related to physical properties and representation. For example, the image feature information may include information such as color, texture, contour (or shape), size, grayscale value, etc., which is not specifically limited here. The above-mentioned image semantic information may refer to the meaning of the image content at the visual level.

[0034] S120: For every two first images of the at least two first images, calculate the image similarity between the two first images based on image information of the two first images.

[0035] Specifically, after obtaining at least two first images and the image information of each first image, the image processing device can calculate the image similarity of the two first images for each two first images in the at least two first images based on the image information of the two first images.

[0036] In an example, if the image information may include image feature information and image semantic information, then the above S120 satisfies the following formula (1):

[0037]

[0038] in, is the image semantic information of the i-th first image, is the image feature information of the i-th first image, correspondingly, The image semantic information of the jth first image, is the image feature information of the j-th first image.

[0039] s represents the similarity between the i-th first image and the j-th first image. The value of s can take any value between -1 and 1. The higher the s value is, the greater the similarity between the i-th first image and the i-th first image is, and the more likely it is that the i-th first image and the i-th first image are the same image.

[0040] In addition, in the embodiment of the present application, the sigmoid activation function shown in the following formula (2) can also be used to map the s value to between [0, 1]:

[0041] f s =sigmoid(s) (2)

[0042] Among them, f s Represents the similarity after mapping by the sigmoid activation function.

[0043] In addition, in order to ensure the accuracy of image deduplication, the image deduplication method provided in the embodiment of the present application can also be further processed by the following formula (3):

[0044]

[0045] in, The loss function representing the cross entropy of the image feature information and image semantic information representation of the first image; θ sem The model parameters representing the image semantic information of the first image v that has not been updated; θ imageModel parameters representing image feature information of the first image v that has not been updated; Model parameters representing the updated image semantic information of the first image v; represents the model parameters of the updated image feature information of the first image v.

[0046] If the f calculated according to the above method s If it does not match the actual similarity between different images, the high-level semantic representation sem of the first image v can be updated according to the cross entropy loss function. v The weights of , and update the s of the first image v image .

[0047] S130: When the image similarity between the two first images is greater than a preset similarity threshold, perform deduplication processing on the two first images to obtain at least one different second image from the at least two first images.

[0048] The preset similarity threshold may be a similarity threshold pre-set based on actual experience or circumstances, and is not specifically limited here. In addition, M is a positive integer greater than or equal to 1.

[0049] Specifically, after obtaining the image similarity of the two first images, the image processing device can first determine whether the image similarity of the two first images is greater than a preset similarity threshold. If so, it indicates that the two first images may be the same image. In this way, the image processing device can deduplicate the two first images, and then obtain at least one different second image from at least two first images.

[0050] It should be noted that, in the process of performing deduplication processing on the above-mentioned two first images, the image processing device may select any one of the two first images for deduplication. In addition, the image processing device may also obtain image quality information of the two first images, and based on the image quality information of the two first images, select the first image with poor image quality from the two first images for deduplication.

[0051] In an embodiment of the present application, by acquiring at least two first images and image information of each first image, the image information may include at least one of image feature information and image semantic information, and then for each two first images of the at least two first images, based on the image information of the first image, the image similarity of the two first images can be calculated, and when the image similarity is greater than a preset similarity threshold, that is, it indicates that the two first images are the same image, the two first images are deduplicated, so as to obtain at least one different second image from the at least two first images. In this way, compared with the prior art method of deduplicating based only on the image name, the present application can accurately and effectively determine whether two images are the same image based on the features of the image at the physical level and the semantics at the visual level, and then can accurately and effectively perform deduplication, thereby improving the accuracy of image deduplication, thereby saving storage space of the cloud disk.

[0052] In order to more accurately and in detail describe an image deduplication method provided by an embodiment of the present application, in one embodiment, if the above-mentioned image information includes image semantic information, the step of obtaining the image information of each first image may specifically include the following steps:

[0053] Processing the first image using a preset semantic recognition model to obtain initial semantic information of the first image, where the initial semantic information includes a plurality of semantic feature vectors;

[0054] Obtaining the semantic weight of each semantic feature vector, where the semantic weight represents the importance of the semantic feature;

[0055] Based on the semantic weight of each semantic feature vector, a weighted sum is performed on the multiple semantic feature vectors to obtain image semantic information of the first image.

[0056] The above-mentioned preset semantic recognition model may be pre-set based on actual experience or circumstances. The model is used to recognize semantic information in an image and is not specifically limited here.

[0057] In addition, the initial semantic information may include multiple semantic feature vectors. Based on this, the semantic weight of each semantic feature vector may be used to represent the importance of the semantic feature vector, that is, the greater the semantic weight, the more important the corresponding semantic feature vector, which is not specifically limited here.

[0058] Specifically, after acquiring at least two first images, the image processing device can process each first image using a preset semantic recognition model to obtain initial semantic information of the first image. The initial semantic information can include multiple semantic feature vectors, and then the semantic weight of each semantic feature vector in the multiple semantic feature vectors can be obtained. Then, the multiple semantic feature vectors can be weighted summed up according to the semantic weight of each semantic feature vector to obtain the image semantic information of the first image.

[0059] In one embodiment, the preset semantic recognition model may be a pre-trained EfficientNetB7 model, which may be constructed using a 1024-dimensional fully connected layer, and is not specifically limited here. Based on this, the first image v may be processed by the following formula to obtain the initial semantic information S of the first image: v , as shown in the following formula (4):

[0060] S v =EfficientNetB7(v) (4)

[0061] Based on this, the initial semantic information S of the first image is obtained. v Afterwards, since each initial semantic information S v It may include multiple semantic feature vectors, for example, the multiple semantic feature vectors may be v1, v2, ..., v i , ……v T , thus, the image semantic information sem of each first image can be calculated by formula (5):

[0062]

[0063] Among them, α i is the semantic weight of the i-th semantic feature vector.

[0064] In this embodiment, the first image can be processed by a preset semantic recognition model to obtain the initial semantic information of the first image, and the semantic weight of each semantic feature vector in the initial semantic information of the first image can be obtained, and then based on the semantic weight of each semantic feature vector, a weighted sum of multiple semantic feature vectors can be performed to obtain the image semantic information of the first image. In this way, the obtained initial semantic information can be processed to obtain image semantic information that can more accurately express the image semantics, so that the image can be deduplicated more accurately based on the image semantic information.

[0065] In the above embodiment, it is necessary to obtain the semantic weight of each semantic feature vector so that the initial semantic information can be processed later to obtain more accurate image semantic information. Based on this, in one embodiment, the step of obtaining the semantic weight of each semantic feature vector may specifically include the following steps:

[0066] Determining a weight matrix, a bias term, and a transposed weight vector for a first image;

[0067] Constructing a semantic weight function corresponding to the first image based on the weight matrix, the bias term, and the transposed weight vector of the first image;

[0068] For each semantic feature vector in the plurality of semantic feature vectors in the first image, the semantic feature vector is processed using a semantic weight function corresponding to the first image to obtain a semantic weight of the semantic feature vector.

[0069] In some embodiments, the weight matrix, bias term and transposed weight vector of the first image can be determined based on a deep learning model, which will not be described in detail here. Among them, the weight matrix refers to a trainable parameter matrix for learning how to map input features to weight scores, the bias term refers to a trainable parameter vector for mapping features that have undergone nonlinear changes to weight scores, and the transposed weight vector refers to a trainable bias vector, which can be determined based on the weight matrix, and is not specifically limited here.

[0070] Specifically, the image processing device can determine the weight matrix, bias term and transposed weight vector of each first image, and then construct a semantic weight function corresponding to the first image based on the weight matrix, bias term and transposed weight vector of the first image. In this way, for each semantic feature vector among the multiple semantic feature vectors in the first image, the semantic feature vector can be processed using the semantic weight function corresponding to the first image to obtain the semantic weight of the semantic feature vector.

[0071] In this embodiment, the weight matrix, bias term and transposed weight vector of each first image can be determined first, and a semantic weight function corresponding to each first image can be constructed based on the above. Then, each semantic feature vector corresponding to the first image can be processed based on the semantic weight function, so as to accurately obtain the semantic weight of each semantic feature vector.

[0072] Based on this, in one embodiment, the above-mentioned semantic weight function may include a semantic scoring function and a scoring conversion function, wherein the scoring conversion function is used to convert the semantic score of each semantic feature vector into a semantic weight, so that the scoring conversion function can characterize the mapping relationship between the semantic score and the semantic weight.

[0073] Thus, the above-mentioned step of processing the semantic feature vector using the semantic weight function corresponding to the first image to obtain the semantic weight of the semantic feature vector may specifically include the following steps:

[0074] The semantic feature vector is processed by using a semantic scoring function to obtain a semantic score of the semantic feature vector, and the semantic score represents the importance of the semantic feature vector;

[0075] Based on the mapping relationship between the semantic score and the semantic weight, the semantic score of each semantic feature vector is matched to obtain the semantic weight of each semantic feature vector.

[0076] The semantic score can represent the importance of the corresponding semantic feature vector. Since the semantic weight can also represent the importance of the corresponding semantic feature vector, it should be noted that the semantic score and the semantic weight are positively correlated, and no specific limitation is made here.

[0077] Specifically, since the above-mentioned semantic weight function can include a semantic scoring function and a scoring conversion function, and the scoring conversion function can characterize the mapping relationship between the semantic score and the semantic weight. In this way, the image deduplication device can process the semantic feature vector using the semantic scoring function to obtain the semantic score of the semantic feature vector, and then can match the semantic score of each semantic feature vector based on the mapping relationship between the semantic score and the semantic weight to obtain the semantic weight of each semantic feature vector.

[0078] In one example, the semantic scoring function k v It can be expressed as the following formula (6):

[0079]

[0080] Among them, W v is the weight matrix of the first image; b v is the deviation term of the first image; is the transposed weight vector of the first image. S v Represents image semantic information of the first image.

[0081] Based on this, the above score conversion function α v It can be expressed as the following formula (7):

[0082]

[0083] In this embodiment, the semantic feature vector can be processed using a semantic scoring function to obtain a semantic score of the semantic feature vector, and based on the mapping relationship between the semantic score and the semantic weight, the semantic score of each semantic feature vector is matched to obtain a semantic weight of each semantic feature vector. In this way, the semantic weight of each semantic feature vector can be accurately obtained.

[0084] In addition, in the prior art, after deduplication processing is usually performed on the image, the image is directly stored in the cloud disk, but the security of the image transmission process is not taken into consideration. In order to avoid the accuracy of the image during the transmission process, in one embodiment, Figure 2 As shown, after calculating the image similarity of the two first images based on the image information of the two first images for each of the at least two first images, the image deduplication method provided in the embodiment of the present application may specifically include the following steps:

[0085] S210: construct a pixel matrix corresponding to each second image.

[0086] S220, obtaining an encryption matrix and a control sequence for encrypting the pixel matrix.

[0087] S230, generating a pixel coding matrix based on the pixel matrix, and generating an encryption coding matrix based on the encryption matrix.

[0088] S240, encrypting the second image based on the pixel coding matrix, the encryption coding matrix and the control sequence to obtain an encrypted second image.

[0089] Specifically, after obtaining at least one second image, the image deduplication device can, for each second image, construct a pixel matrix corresponding to each second image, and obtain an encryption matrix and a control sequence for encrypting the pixel matrix, and then generate a pixel coding matrix based on the pixel matrix, and generate an encryption coding matrix based on the encryption matrix, and then encrypt the second image based on the pixel coding matrix, the encryption coding matrix and the control sequence to obtain an encrypted second image.

[0090] In the embodiment of the present application, for each second image, a corresponding pixel matrix can be constructed, and an encryption matrix and a control sequence for encrypting the pixel matrix can be used to generate a pixel coding matrix based on the pixel matrix, and an encryption coding matrix can be generated based on the encryption matrix, and then the second image can be encrypted based on the pixel encryption matrix, the encryption coding matrix and the control sequence to obtain an encrypted second image. In this way, compared with the direct plaintext storage method in the prior art, the embodiment of the present application can encrypt the image to be uploaded to avoid the image being tampered with during the transmission process, thereby improving the security of the image during the image transmission process.

[0091] In order to accurately describe the image deduplication method provided by the embodiment of the present application, in one embodiment, the above S210 may specifically include the following steps:

[0092] Determine image color information of each second image, the second image comprising a plurality of pixels, the image color information comprising a pixel RGB value of each pixel in the plurality of pixels;

[0093] A pixel matrix is ​​constructed based on the pixel RGB value of each pixel in each second image.

[0094] In some embodiments, each of the second images may include a plurality of pixels, and correspondingly, the image color information may include a pixel RGB value of each of the plurality of pixels.

[0095] Specifically, the image deduplication device can first determine the image color information of each second image for each second image. Since the second image may include multiple pixels, the image color information may include the pixel RGB value of each pixel in the multiple pixels. In this way, a pixel matrix is ​​constructed based on the pixel RGB value of each pixel in each second image.

[0096] In this embodiment, the image color information of each second image can be determined, and then the corresponding pixel matrix can be constructed based on the color information of each second image. In this way, the pixel matrix corresponding to each image can be accurately obtained, which is convenient for ensuring the accuracy of the subsequently encrypted image.

[0097] Based on this, in one embodiment, the above S230 may specifically include the following steps:

[0098] A pixel matrix is ​​obtained by converting the pixel matrix into a binary matrix, and a pixel coding matrix is ​​generated based on the pixel matrix according to a first preset rule;

[0099] An encryption matrix is ​​obtained by converting the encryption matrix into a binary matrix, and an encryption coding matrix is ​​generated based on the encryption matrix according to a second preset rule.

[0100] Among them, the first preset rule and the second preset rule can be preset based on actual experience or circumstances, and are not specifically limited here.

[0101] Specifically, after acquiring the pixel matrix and the corresponding encryption matrix of each second image, the image deduplication device can convert the pixel matrix into a binary matrix to obtain a pixel base matrix, and generate a pixel coding matrix based on the pixel base matrix according to a first preset rule, and convert the encryption matrix into a binary matrix to obtain an encryption base matrix, and generate an encryption coding matrix based on the encryption base matrix according to a second preset rule.

[0102] In this embodiment, the pixel matrix and the encryption matrix used to encrypt the pixel matrix can be simultaneously converted into a binary matrix, and then converted into a coding matrix to obtain a pixel coding matrix and an encryption coding matrix, so as to facilitate the subsequent better encryption of the image and optimize the image encryption effect.

[0103] In order to accurately describe the image encryption process involved in the embodiments of the present application, in some embodiments, the pixel RGB value of each pixel may include the R value, G value and B value of the pixel, and accordingly, the above-mentioned pixel matrix may include a first sub-matrix constructed based on the R value of each pixel in the second image, a second sub-matrix constructed based on the G value of each pixel in the second image, and a third sub-matrix constructed based on the B value of each pixel in the second image;

[0104] Correspondingly, the pixel base matrix includes a first sub-pixel matrix corresponding to the first sub-matrix, a second sub-pixel matrix corresponding to the second sub-matrix, and a third sub-pixel matrix corresponding to the third sub-matrix, and the pixel coding matrix includes a first sub-pixel coding matrix, a second sub-pixel coding matrix, and a third sub-pixel coding matrix;

[0105] In addition, the encryption matrix may include a first sub-encryption matrix corresponding to the first sub-matrix, a second sub-encryption matrix corresponding to the second sub-matrix, and a third sub-encryption matrix corresponding to the third sub-matrix;

[0106] Correspondingly, the encryption matrix includes a first sub-encryption matrix, a second sub-encryption matrix and a second sub-encryption matrix, and the encryption coding matrix includes a first sub-encryption coding matrix, a second sub-encryption coding matrix and a second sub-encryption coding matrix.

[0107] Based on this, in a specific embodiment, for each second image, the image deduplication device can construct the first sub-matrix, the second sub-matrix and the third sub-matrix according to the following formula based on the R value, G value and B value in the pixel RGB value of each pixel point in the second image, as shown below:

[0108] I RC ={RC1,RC2,…,RC M×N} (8)

[0109] I GC = {GC1,GC2,…,GC M×N} (9)

[0110] I BC ={BC1,BC2,…,BC M×N} (10)

[0111] Among them, the above IRC It can be the first sub-matrix constructed based on the R value of each pixel. GC It can be a second sub-matrix constructed based on the G value of each pixel. BC It can be a third sub-matrix constructed by B for each pixel. M and N represent the image size of the second image in rows and columns, respectively.

[0112] The sub-encryption matrices corresponding to the first sub-matrix, the second sub-matrix, and the third sub-matrix are as follows:

[0113] S x ={x1,x2,…,x M×N} (11)

[0114] S y ={y1,y2,…,y M×N} (12)

[0115] S z ={y1,y2,…,y M×N} (13)

[0116] Among them, S x is the first sub-encryption matrix, S y is the second sub-encryption matrix, S z It is the third sub-encryption matrix.

[0117] in addition,

[0118]

[0119] Based on this, the above parameter values ​​can be As a security key corresponding to the second image.

[0120] In this way, we can RC ,I GC ,I BC and S x ,S y ,S z After obtaining the six binary matrices, the six binary matrices are converted into six encoding matrices using the preset rules, which are recorded as and

[0121] In this embodiment, a pixel matrix corresponding to different color values ​​can be constructed and a corresponding encryption matrix can be obtained, so as to facilitate subsequent pixel matrices corresponding to different color values ​​and ensure the accuracy of the subsequently encrypted image.

[0122] Based on this, in one embodiment, the above S240 may specifically include the following steps:

[0123] Based on the control sequence, the pixel coding matrix and the encryption coding matrix are XOR-calculated respectively to obtain the target matrix;

[0124] The second image is encrypted based on the target matrix to obtain an encrypted second image.

[0125] Specifically, the image deduplication device can perform XOR calculations on the pixel coding matrix and the encryption coding matrix respectively based on the control sequence to obtain a target matrix, and encrypt the second image based on the target matrix to obtain an encrypted second image.

[0126] In addition, continuing with the above embodiment, if the above control sequence can be obtained by the following formula:

[0127]

[0128] a=10etb=1 (17)

[0129] Among them, a, e, t, b, and m are preset control sequence generation parameters. In addition, according to actual experience, the value of a can be set to 10, the value of b can be set to 1, the value of m can be set to -1, and the value of k can be set to 1.

[0130] Then, four control sequences (X, Y, Z and W) are obtained by solving according to the above method.

[0131] Based on this, after obtaining 4 control sequences and 6 encoding matrices, the XOR calculation is performed as follows:

[0132]

[0133] In this way, the three mixed matrices can be combined to obtain an encrypted image.

[0134] In this embodiment, based on the control sequence, the pixel coding matrix and the encryption coding matrix can be XORed to obtain the target matrix, and the second image can be encrypted based on the target matrix to obtain the encrypted second image. In this way, compared with the direct plaintext storage method in the prior art, the image to be uploaded can be encrypted to avoid the image being tampered with during the transmission process, thereby improving the security of the image during the image transmission process.

[0135] In addition, considering that in the process of transmitting images to the cloud disk, noise may be generated due to external factors such as impurity signals of the transmission equipment, resulting in the image actually uploaded being degraded and inconsistent with the image to be uploaded. In order to avoid this problem, the image deduplication method provided in the embodiment of the present application can perform denoising on the image to be uploaded before uploading the image. The purpose of denoising is to remove noise points, restore the original image, and optimize the retention of initial details.

[0136] Based on this, in one embodiment, Figure 3 As shown, before obtaining the image information of each first image, the image deduplication method provided in the embodiment of the present application may specifically include the following steps:

[0137] S310: extract features from each first image using an image sparse model to obtain first feature data.

[0138] S320: Perform enhancement processing on the first feature data using an image enhancement model to obtain second feature data.

[0139] S330, extracting features from each first image using a feature extraction model to obtain third feature data.

[0140] S340: Use an image fusion model to fuse the second feature data and the third feature data to obtain fourth feature data.

[0141] S350: Process the fourth feature data using the attention mechanism model to obtain a denoised first image.

[0142] Specifically, the image deduplication device can use the image sparse model to extract features from each first image to obtain first feature data, and then use the image enhancement model to enhance the first feature data to obtain second feature data, and can also use the feature extraction model to extract features from each first image to obtain third feature data, and then use the image fusion model to fuse the second feature data and the third feature data to obtain fourth feature data. The fourth feature data is processed using the attention mechanism model to obtain the first image after denoising.

[0143] In a specific example, Figure 4 As shown, the present application embodiment provides a structural schematic diagram of an image denoising model, which may include five sub-models, namely, an image sparse model (i.e. Figure 1 The green frame part), image enhancement model Figure 1 The yellow frame in the figure), feature extraction model (i.e. Figure 1 The red frame part), image fusion model is Figure 1The blue box in the middle) and the attention mechanism model (i.e. Figure 1 The purple framed part).

[0144] Among them, the above-mentioned image sparse model mainly includes layers 1 to 12 in the image denoising model structure. In addition, in layers 2, 5, 9 and 12, the embodiment of the present application uses dilated convolution (Dilated_Conv) instead of ordinary convolution, and each layer includes dilated convolution + BN + ReLU operations. Other layers (that is, layers other than layers 2, 5, 9 and 12 in layers 1 to 12) follow Conv+BN+ReLU. The size of the convolution kernel is 3×3, the depth of the convolution layer is 64, and the dilation factor of the dilated convolution is 2. In addition, the padding of the dilated convolution layer is set to 2, and the padding of other convolution layers is set to 1. The image sparse model with such a setting can provide a receptive field that is twice the size of the receptive field of the traditional convolution kernel.

[0145] After the image to be uploaded (the first image) is input into the image denoising model, it is processed by the image sparse model according to the following calculation method:

[0146]

[0147] in, represents the output of the (l-1)th layer in the image sparse model; ψ [l] represents the activation function of the lth layer; γ [l] represents the batch normalization process; ) represents the convolution kernel of the lth layer. Where f and c represent the size and depth of the convolution kernel respectively; D k represents the dilation factor of the dilated convolution. If D k =1, it means using ordinary convolution.

[0148] Combined with the above model parameters set for the image sparse model in the embodiment of the present application, assuming that the spatial resolution of the image to be uploaded is 256×256, the output result of the image sparse model can be further obtained as follows:

[0149]

[0150] Among them, O SB(256,256,64) It indicates the output result of the image to be uploaded with a spatial resolution of 256×256 under the image sparse model with a convolution layer depth of 64. The above spatial resolution of 256×256 is only an example and can also be other parameters.

[0151] This image enhancement model consists of four layers, the first three layers have Conv+BN+ReLU, and the fourth layer has only one Conv operation. The image enhancement model mainly achieves feature enhancement through dimensionality reduction, which can perform compression on the features extracted by the image sparse model. Here, the convolution kernel size is 3×3, and the output is used as its input. When passing through the image enhancement model, the output depth is compressed from 64 to a depth of 1. In addition, the compression achieved by dimensionality reduction can retain the uniqueness of the features obtained from the image sparse model.

[0152] Secondly, since the 16th layer does not use batch normalization and activation function, the activation function and batch normalization process are set to 1, that is, ψ

[16] =1,γ

[16] =1, in this case, the final output of the image enhancement model is:

[0153]

[0154] O EB(256,256,1) It represents the output result of the uploaded image with a spatial resolution of 256×256 and a convolutional layer depth of 1 processed by the image enhancement model.

[0155] The feature extraction model is the underlying network of the image denoising model, which can provide complementary functions of the image sparse model and the image enhancement model. From the above structure diagram of the image denoising model, it can be seen that the feature extraction model includes 17 layers, among which the 1st to 16th layers are all Conv+BN+ReLU structures, the kernel size is 64, and the last layer contains only a single convolution.

[0156] Among them, since there is no dilated convolution in the feature extraction model, the image denoising model can extract dense noise features from noisy input images. The output result of each layer of the feature extraction model can be expressed as:

[0157]

[0158] Represents the output of the (l-1)th layer of the feature extraction model; That is, the input of the feature extraction model;

[0159] For example, the output of the 17th layer of the feature extraction model can be expressed as:

[0160]

[0161] Among them, O GFB(256,256,1) It represents the output result of the uploaded image with a spatial resolution of 256×256 and a convolutional layer depth of 1 after being processed by the feature extraction model.

[0162] Considering that in the prior art, although deep networks can provide a larger receptive field and thus extract unique features, increasing the network depth may not achieve a good denoising effect. In order to avoid this problem, in the embodiments of the present application, it is considered to combine the image sparse model of the image denoising model, the image enhancement model, and the features extracted by the feature extraction model through an image fusion model.

[0163] Specifically, the features extracted from the image sparse model and the image enhancement model can be added to the image to be uploaded; then the features obtained from the feature extraction model are subtracted from the input; finally, the fusion results obtained from the above two steps are propagated to the activation layer.

[0164] The image sparse model is used to extract key information from the image to be uploaded, and then add it to the image to be uploaded to enhance the weight of the key information. The image enhancement model is used to reduce noise on the input image to be uploaded.

[0165] The output of the image fusion model can be expressed as follows:

[0166]

[0167] Among them, O FB(256,256,3) represents the output of the image fusion model; ψ represents the activation function Tanh.

[0168] The embodiment of the present application introduces the attention mechanism model to further enhance the quality of features extracted from the image fusion model, so that the image denoising model can focus more on the noise-free area and suppress irrelevant features, that is, prevent features from being extracted from the noise area, thereby obtaining a smooth output.

[0169] The attention mechanism model of the embodiment of the present application consists of a convolution kernel and a multiplier. The output of the attention mechanism model will be used as the final feature extracted by the network, which can be expressed as:

[0170]

[0171] Among them, K (1X1,1,Dk＝1) Represents a convolution kernel of size 1x1.

[0172] Finally, a clean image corresponding to the image to be uploaded is constructed through residual learning, which is calculated as follows:

[0173] X (256,256,1) =Y-NF final (25)

[0174] Among them, X (256,256,1) represents the final denoised image.

[0175] In this embodiment, the image can be denoised by the model before uploading the image. During the denoising process, in order to retain the details of the denoised image to the greatest extent and achieve better global image representation, the actual uploaded image may be avoided from being degraded and inconsistent with the image to be uploaded.

[0176] Based on the same inventive concept, the embodiment of the present application provides an image deduplication device, specifically in combination with the attached Figure 5 The image deduplication device provided in the embodiment of the present application is described in detail.

[0177] Figure 5 It is a structural schematic diagram of an image deduplication device provided in an embodiment of the present application.

[0178] like Figure 5 As shown, the image deduplication device 500 may include:

[0179] An acquisition module 510, configured to acquire at least two first images and image information of each first image, where the image information includes at least one of image feature information and image semantic information;

[0180] A calculation module 520, configured to calculate, for each two first images of the at least two first images, an image similarity of the two first images based on image information of the two first images;

[0181] The deduplication module 530 is configured to perform deduplication processing on the two first images when the image similarity between the two first images is greater than a preset similarity threshold, so as to obtain at least one different second image from the at least two first images.

[0182] In one embodiment, the image information includes image semantic information; the image deduplication device provided in the embodiment of the present application further includes:

[0183] A processing module, used to process the first image using a preset semantic recognition model to obtain initial semantic information of the first image, where the initial semantic information includes a plurality of semantic feature vectors;

[0184] An acquisition module is specifically used to acquire the semantic weight of each semantic feature vector, where the semantic weight represents the importance of the semantic feature;

[0185] The weighted summation module is used to perform weighted summation on multiple semantic feature vectors based on the semantic weight of each semantic feature vector to obtain image semantic information of the first image.

[0186] In one embodiment, the image deduplication device provided in the embodiment of the present application further includes:

[0187] The determining device is further used to determine the weight matrix, the bias term and the transposed weight vector of the first image;

[0188] A construction module, used to construct a semantic weight function corresponding to the first image based on the weight matrix, the bias term and the transposed weight vector of the first image;

[0189] The processing module is used to process each semantic feature vector among the multiple semantic feature vectors in the first image using a semantic weight function corresponding to the first image to obtain a semantic weight of the semantic feature vector.

[0190] In one embodiment, the semantic weight function corresponding to the first image includes a semantic scoring function and a scoring conversion function, and the scoring conversion function represents a mapping relationship between the semantic score and the semantic weight;

[0191] Based on this, the above-mentioned image deduplication device also includes:

[0192] The processing module is specifically used to process the semantic feature vector using the semantic scoring function to obtain the semantic score of the semantic feature vector, where the semantic score represents the importance of the semantic feature vector;

[0193] The matching module is used to match the semantic score of each semantic feature vector based on the mapping relationship between the semantic score and the semantic weight to obtain the semantic weight of each semantic feature vector.

[0194] In one embodiment, the image deduplication device provided in the embodiment of the present application may further include:

[0195] The construction module is further used to construct a pixel matrix corresponding to each second image;

[0196] The acquisition module is further used to acquire an encryption matrix and a control sequence for encrypting the pixel matrix;

[0197] A generating module, used for generating a pixel coding matrix based on the pixel matrix, and generating an encryption coding matrix based on the encryption matrix;

[0198] The encryption processing module is used to perform encryption processing on the second image based on the pixel encryption matrix, the encryption coding matrix and the control sequence to obtain the encrypted second image.

[0199] In one embodiment, the image processing device provided by the embodiment of the present application further includes:

[0200] The determination module is further used to determine the image color information of each second image, the second image includes a plurality of pixels, and the image color information includes the pixel RGB value of each pixel in the plurality of pixels;

[0201] The construction module is specifically used to construct a pixel matrix based on the pixel RGB value of each pixel point in each second image.

[0202] In one embodiment, the generating module is specifically used to convert the pixel matrix into a binary matrix to obtain a pixel base matrix, and generate a pixel coding matrix based on the pixel base matrix according to a first preset rule;

[0203] The above-mentioned generation module is specifically used to obtain an encryption matrix by converting the encryption matrix into a binary matrix, and generate an encryption coding matrix based on the encryption matrix according to the second preset rule.

[0204] In some embodiments, the pixel RGB value of each pixel includes an R value, a G value, and a B value of the pixel, and the pixel matrix includes a first submatrix constructed based on the R value of each pixel in the second image, a second submatrix constructed based on the G value of each pixel in the second image, and a third submatrix constructed based on the B value of each pixel in the second image;

[0205] The pixel matrix includes a first sub-pixel matrix, a second sub-pixel matrix and a third sub-pixel matrix, and the pixel coding matrix includes a first sub-pixel coding matrix, a second sub-pixel coding matrix and a third sub-pixel coding matrix;

[0206] The encryption matrix includes a first sub-encryption matrix corresponding to the first sub-matrix, a second sub-encryption matrix corresponding to the second sub-matrix, and a third sub-encryption matrix corresponding to the third sub-matrix;

[0207] The encryption base sequence includes a first sub-encryption base matrix, a second sub-encryption base matrix and a second sub-encryption base matrix, and the encryption coding matrix includes a first sub-encryption coding matrix, a second sub-encryption coding matrix and a second sub-encryption coding matrix.

[0208] In one embodiment, the image deduplication device provided in the embodiment of the present application may further include:

[0209] An XOR calculation module is used to perform XOR calculations on the pixel encryption matrix and the encryption coding matrix respectively based on the control sequence to obtain a target matrix;

[0210] The encryption processing module is specifically used to perform encryption processing on the second image based on the target matrix to obtain an encrypted second image.

[0211] In one embodiment, the image deduplication device provided in the embodiment of the present application may further include a model processing module.

[0212] For each first image, the model processing module is specifically used for:

[0213] Extracting features from each first image using an image sparse model to obtain first feature data;

[0214] Performing feature enhancement processing on the first feature data using an image enhancement model to obtain second feature data;

[0215] Extracting features from each first image using a feature extraction model to obtain third feature data;

[0216] The second feature data and the third feature data are fused using an image fusion model to obtain fourth feature data;

[0217] The fourth feature data is processed using the attention mechanism model to obtain the denoised first image.

[0218] In an embodiment of the present application, by acquiring at least two first images and image information of each first image, the image information may include at least one of image feature information and image semantic information, and then for each two first images of the at least two first images, based on the image information of the first image, the image similarity of the two first images can be calculated, and when the image similarity is greater than a preset similarity threshold, that is, it indicates that the two first images are the same image, the two first images are deduplicated, so as to obtain at least one different second image from the at least two first images. In this way, compared with the prior art method of deduplicating based only on the image name, the present application can accurately and effectively determine whether two images are the same image based on the features of the image at the physical level and the semantics at the visual level, and then can accurately and effectively perform deduplication, thereby improving the accuracy of image deduplication, thereby saving storage space of the cloud disk.

[0219] Each module in the image deduplication device provided in the embodiment of the present application can be implemented Figures 1 to 3 The method steps of any of the embodiments shown in the figure can achieve the corresponding technical effects, which will not be repeated here for the sake of brevity.

[0220] Figure 6 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application is shown.

[0221] The electronic device may include a processor 601 and a memory 602 storing computer program instructions.

[0222] Specifically, the processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0223] The memory 602 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 602 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 602 may be inside or outside the electronic device. In a particular embodiment, the memory 602 is a non-volatile solid-state memory.

[0224] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical or other physical / tangible memory storage devices. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to an aspect of the present disclosure.

[0225] The processor 601 implements any one of the image deduplication methods in the above embodiments by reading and executing computer program instructions stored in the memory 602 .

[0226] In one example, the electronic device may further include a communication interface 603 and a bus 610. Figure 6 As shown, the processor 601, the memory 602, and the communication interface 603 are connected via a bus 610 and communicate with each other.

[0227] The communication interface 603 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0228] Bus 610 includes hardware, software or both, and the parts of electronic equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 610 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0229] In addition, in combination with the image deduplication method in the above embodiment, the embodiment of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, the image deduplication method provided in the embodiment of the present application is implemented.

[0230] An embodiment of the present application also provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the image deduplication method provided in the embodiment of the present application.

[0231] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0232] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0233] The above are only specific implementation methods of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. An image deduplication method, characterized in that: The method comprises: Acquire at least two first images and image information of each of the first images, wherein the image information includes at least one of image feature information and image semantic information; For each two first images of the at least two first images, based on image information of the two first images, calculating image similarity of the two first images; When the image similarity between the two first images is greater than a preset similarity threshold, deduplication processing is performed on the two first images to obtain at least one different second image from the at least two first images.

2. The method according to claim 1, characterized in that The image information includes image semantic information; acquiring the image information of each of the first images includes: Processing the first image using a preset semantic recognition model to obtain initial semantic information of the first image, wherein the initial semantic information includes a plurality of semantic feature vectors; Obtaining a semantic weight of each of the semantic feature vectors, wherein the semantic weight represents the importance of the semantic feature; Based on the semantic weight of each of the semantic feature vectors, the multiple semantic feature vectors are weightedly summed to obtain image semantic information of the first image.

3. The method according to claim 2, characterized in that The obtaining of the semantic weight of each of the semantic feature vectors comprises: Determining a weight matrix, a bias term, and a transposed weight vector of the first image; Constructing a semantic weight function corresponding to the first image based on the weight matrix, the bias term, and the transposed weight vector of the first image; For each semantic feature vector in the plurality of semantic feature vectors in the first image, the semantic feature vector is processed using a semantic weight function corresponding to the first image to obtain a semantic weight of the semantic feature vector.

4. The method according to claim 3, characterized in that The semantic weight function corresponding to the first image includes a semantic scoring function and a scoring conversion function, wherein the scoring conversion function represents a mapping relationship between a semantic score and a semantic weight; The using the semantic weight function corresponding to the first image to process the semantic feature vector to obtain the semantic weight of the semantic feature vector includes: Processing the semantic feature vector using the semantic scoring function to obtain a semantic score of the semantic feature vector, wherein the semantic score represents the importance of the semantic feature vector; Based on the mapping relationship between the semantic score and the semantic weight, the semantic score of each of the semantic feature vectors is matched to obtain the semantic weight of each of the semantic feature vectors.

5. The method according to any one of claims 1 to 4, characterized in that: When the image similarity of the two first images is greater than a preset similarity threshold, after performing deduplication processing on the two first images to obtain at least one different second image from the at least two first images, the method further includes: Constructing a pixel matrix corresponding to each of the second images; Obtaining an encryption matrix and a control sequence for encrypting the pixel matrix; generating a pixel encoding matrix based on the pixel matrix, and generating an encryption encoding matrix based on the encryption matrix; Based on the pixel coding matrix, the encryption coding matrix and the control sequence, the second image is encrypted to obtain an encrypted second image.

6. The method according to claim 5, characterized in that The step of constructing a pixel matrix corresponding to each second image includes: Determine image color information of each of the second images, the second image comprising a plurality of pixels, the image color information comprising a pixel RGB value of each of the plurality of pixels; A pixel matrix is ​​constructed based on the pixel RGB value of each pixel in each of the second images.

7. The method according to claim 6, characterized in that The step of generating a pixel coding matrix based on the pixel matrix and generating an encryption coding matrix based on the encryption matrix comprises: Converting the pixel matrix into a binary matrix to obtain a pixel matrix, and generating a pixel coding matrix based on the pixel matrix according to a first preset rule; An encryption matrix is ​​obtained by converting the encryption matrix into a binary matrix, and an encryption coding matrix is ​​generated based on the encryption matrix according to a second preset rule.

8. The method according to claim 7, characterized in that The pixel RGB value of each pixel includes an R value, a G value, and a B value of the pixel, and the pixel matrix includes a first submatrix constructed based on the R value of each pixel in the second image, a second submatrix constructed based on the G value of each pixel in the second image, and a third submatrix constructed based on the B value of each pixel in the second image; The pixel matrix includes a first sub-pixel matrix, a second sub-pixel matrix and a third sub-pixel matrix, and the pixel coding matrix includes a first sub-pixel coding matrix, a second sub-pixel coding matrix and a third sub-pixel coding matrix; The encryption matrix includes a first sub-encryption matrix corresponding to the first sub-matrix, a second sub-encryption matrix corresponding to the second sub-matrix, and a third sub-encryption matrix corresponding to the third sub-matrix; The encryption matrix includes a first sub-encryption matrix, a second sub-encryption matrix and a second sub-encryption matrix, and the encryption coding matrix includes a first sub-encryption coding matrix, a second sub-encryption coding matrix and a second sub-encryption coding matrix.

9. The method according to claim 8, characterized in that The step of encrypting the second image based on the pixel coding matrix, the encryption coding matrix and the control sequence to obtain the encrypted second image comprises: Based on the control sequence, performing XOR calculation on the pixel coding matrix and the encryption coding matrix respectively to obtain a target matrix; The second image is encrypted based on the target matrix to obtain an encrypted second image.

10. The method according to claim 1, characterized in that Before acquiring the image information of the first image for each of the first images, the method further includes: Extracting features from each of the first images using an image sparse model to obtain first feature data; Performing feature enhancement processing on the first feature data using an image enhancement model to obtain second feature data; Using a feature extraction model to extract features from each of the first images to obtain third feature data; fusing the second feature data and the third feature data using an image fusion model to obtain fourth feature data; The fourth feature data is processed using the attention mechanism model to obtain a denoised first image.

11. An image deduplication device, characterized in that: The device comprises: An acquisition module, used to acquire at least two first images and image information of each of the first images, wherein the image information includes at least one of image feature information and image semantic information; a calculation module, configured to calculate, for each two first images of the at least two first images, an image similarity of the two first images based on image information of the two first images; The deduplication module is used to perform deduplication processing on the two first images when the image similarity between the two first images is greater than a preset similarity threshold, so as to obtain at least one different second image from the at least two first images.

12. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; The processor reads and executes the computer program instructions to implement the image deduplication method as described in any one of claims 1-10.

13. A computer storage medium, characterized in that: The computer storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the image deduplication method according to any one of claims 1 to 10 is implemented.

14. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the image deduplication method according to any one of claims 1 to 10 is implemented.

Images