Encryption and decryption system and method

Abstract

The present disclosure provides an encryption and decryption system and method. The encryption and decryption system includes a transmitting device and a receiving device. The transmitting device is configured to store a plurality of original images. The receiving device is connected to the transmitting device and configured to store a corresponding table. The transmitting device generates an encrypted string, selects a representative transformed image from the plurality of transformed images according to the encrypted string, transmits the representative transformed image to the receiving device, and does not transmit the encrypted string to the receiving device. The receiving device identifies a first original image number and a second original image number from the representative transformed image, and looks up the corresponding table according to the first original image number and the second original image number to generate the encrypted string. Thus, various network attacks can be prevented, resource and time consumption can be greatly reduced, the problem of excessive distortion or low capacity of embedded data can be solved, and the information recovery capability can be greatly improved.

Description

Technical Field

[0001] This disclosure relates to an encryption / decryption system and method. Background Technology

[0002] In the field of network encryption, complex encryption or hash algorithms are often required to encrypt data. Common algorithms include symmetric and asymmetric encryption algorithms. These algorithms often require enormous resources and time, or the data embedding capacity is too small. Furthermore, these algorithms often cannot guarantee encryption recovery capabilities after an attack. Summary of the Invention

[0003] This disclosure proposes an encryption / decryption system including a transmitting device and a receiving device. The transmitting device includes a first memory, a first processor, and a first transceiver circuit. The first memory stores multiple original images. The first processor is used to perform image warping processing on the multiple original images in pairs to generate multiple warped images, and records the correspondence between the numbers of the multiple original images and the numbers of the multiple warped images in a correspondence table stored in the first memory. The first processor is also used to generate an encrypted string and select a representative warped image from the multiple warped images based on the encrypted string. The first transceiver circuit is used to transmit the representative warped image but does not transmit the encrypted string. The receiving device is connected to the transmitting device and includes a second memory, a second transceiver circuit, and a second processor. The second memory stores the correspondence table. The second transceiver circuit is used to receive the representative warped image from the first transceiver circuit but does not receive the encrypted string. The second processor identifies the first original image number and the second original image number from the representative warped image and looks up the corresponding number in the correspondence table to generate the encrypted string.

[0004] In one embodiment, the first memory and the second memory share a key. The first processor uses the key to perform an encryption algorithm on the data to generate an encrypted string. When the encrypted string is generated by looking up a corresponding table based on the first original image number and the second original image number, the second processor uses the key to perform a decryption algorithm on the encrypted string to generate the data.

[0005] In one embodiment, the number representing the deformed image is the same as the encrypted string.

[0006] In one embodiment, a first processor selects a first image and a second image from multiple original images. The first processor performs feature point recognition processing on the first image and the second image to generate multiple first target points in the first image and multiple second target points in the second image, and establishes a target point correspondence relationship between the multiple first target points and the multiple second target points. The first processor generates the coordinates of multiple deformation points from the coordinates of the multiple first target points and the multiple second target points according to the adjustment parameters and the target point correspondence relationship. The first processor performs alignment processing on the first image according to the coordinates of the multiple first target points and the multiple deformation points to generate a first distorted image, and performs alignment processing on the second image according to the coordinates of the multiple second target points and the multiple deformation points to generate a second distorted image. Finally, the first processor converts the first distorted image and the second distorted image into corresponding deformed images according to the adjustment parameters.

[0007] In one embodiment, the first processor records the correspondence between the first image and the second image and the corresponding deformed image in a correspondence table.

[0008] In one embodiment, the second processor performs image deformation processing on multiple original images in pairs according to multiple adjustment parameters to generate multiple corresponding deformed images, and the second processor updates the recognition model using the multiple corresponding deformed images and multiple original images, so as to use the recognition model to identify the first original image number and the second original image number from the representative deformed images.

[0009] In one embodiment, the second processor selects a first image and a second image from multiple original images, and performs image deformation processing on the first image and the second image according to adjustment parameters to generate corresponding deformed images. The second processor uses the corresponding deformed images as training samples, and uses the first image and the second image as a first training label and a second training label. The second processor inputs the training samples into the recognition model to generate a first predicted label and a second predicted label, and uses the first predicted label, the second predicted label, the first training label, and the second training label to calculate additive angular interval loss to generate a loss. The second processor uses the loss to perform backpropagation processing on the recognition model to update the recognition model.

[0010] In one embodiment, the recognition model is a version 2 deformable face recognition network, which includes a feature extraction layer and a fully connected layer. The input of the fully connected layer is connected to the output of the feature extraction layer. The feature extraction layer is a deep residual neural network, and the fully connected layer is a fully connected L2 normalization layer.

[0011] In one embodiment, the number of original images is N, and the number of deformed images is And the number of bits for the numbering of multiple deformed images is

[0012] In one embodiment, the original images are numbered in decimal and the deformed images are numbered in binary.

[0013] This disclosure proposes an encryption / decryption method, comprising: generating an encrypted string; performing image deformation processing on multiple original images in pairs to generate multiple deformed images; recording the correspondence between the numbers of the multiple original images and the numbers of the multiple deformed images in a correspondence table; selecting a representative deformed image from the multiple deformed images based on the encrypted string; transmitting the representative deformed image without transmitting the encrypted string; when receiving the representative deformed image, identifying a first original image number and a second original image number from the representative deformed image; and looking up the correspondence table based on the first original image number and the second original image number to generate the encrypted string.

[0014] In one embodiment, the encryption / decryption method further includes: using a shared key to perform an encryption algorithm on the data to generate an encrypted string; and when an encrypted string is generated by looking up a corresponding table based on a first original image number and a second original image number, using the key to perform a decryption algorithm on the encrypted string to generate the data.

[0015] In one embodiment, the number representing the deformed image is the same as the encrypted string.

[0016] In one embodiment, the encryption / decryption method further includes: selecting a first image and a second image from a plurality of original images; performing feature point recognition processing on the first image and the second image to generate a plurality of first target points in the first image and a plurality of second target points in the second image, and establishing a target point correspondence relationship between the plurality of first target points and the plurality of second target points; generating the coordinates of a plurality of deformable points from the coordinates of the plurality of first target points and the coordinates of the plurality of second target points according to adjustment parameters and the target point correspondence relationship; aligning the first image according to the coordinates of the plurality of first target points and the coordinates of the plurality of deformable points to generate a first distorted image, and aligning the second image according to the coordinates of the plurality of second target points and the coordinates of the plurality of deformable points to generate a second distorted image; and converting the first distorted image and the second distorted image into corresponding deformable images according to adjustment parameters.

[0017] In one embodiment, the encryption / decryption method further includes: recording the correspondence between the first image and the second image and the corresponding deformed image in a correspondence table.

[0018] In one embodiment, the encryption / decryption method further includes: performing image deformation processing on multiple original images in pairs according to multiple adjustment parameters to generate multiple corresponding deformed images, and updating the recognition model using the multiple corresponding deformed images and multiple original images, so as to use the recognition model to identify the first original image number and the second original image number from the representative deformed images.

[0019] In one embodiment, the encryption / decryption method further includes: selecting a first image and a second image from multiple original images, performing image deformation processing on the first image and the second image according to adjustment parameters to generate corresponding deformed images, using the corresponding deformed images as training samples, and using the first image and the second image as a first training label and a second training label, inputting the training samples into a recognition model to generate a first predicted label and a second predicted label, calculating an additive angular interval loss using the first predicted label, the second predicted label, the first training label, and the second training label to generate a loss, and performing backpropagation processing on the recognition model using the loss to update the recognition model.

[0020] In one embodiment, the recognition model is a version 2 deformable face recognition network, which includes a feature extraction layer and a fully connected layer. The input of the fully connected layer is connected to the output of the feature extraction layer. The feature extraction layer is a deep residual neural network, and the fully connected layer is a fully connected L2 normalization layer.

[0021] In one embodiment, the number of original images is N, and the number of deformed images is And the number of bits for the numbering of multiple deformed images is

[0022] In one embodiment, the original images are numbered in decimal and the deformed images are numbered in binary. Attached Figure Description

[0023] Figure 1 This is a block diagram of the encryption / decryption system disclosed herein.

[0024] Figure 2 This is a flowchart of the encryption and decryption method disclosed herein.

[0025] Figure 3 This is a schematic diagram of an encryption / decryption system implemented according to some embodiments of this disclosure.

[0026] Figure 4 This is a schematic diagram illustrating the generation of a correspondence table according to some embodiments of this disclosure.

[0027] Figure 5 According to some embodiments of this disclosure Figure 3 A detailed flowchart of the first step.

[0028] Figure 6 This is a schematic diagram of target points according to some embodiments of this disclosure.

[0029] Figure 7 This is a schematic diagram illustrating the effect of adjustment parameters on deformed images according to some embodiments of this disclosure.

[0030] Figure 8 This is a schematic diagram of an updated identification model according to some embodiments of this disclosure. Detailed Implementation

[0031] Current encryption technologies often require complex data processing steps to encrypt and decrypt data, resulting in wasted resources and time. Furthermore, previous image encryption technologies frequently suffer from excessive image distortion or insufficient embedded data capacity. They may even lack the ability to recover information after an attack on the encrypted data. In view of this, this disclosure proposes an encryption / decryption system and method. This system and method first generates a large number of deformed images from a small number of original images, and then selects representative deformed images (i.e., the corresponding deformed images) based on the encryption string to transmit from the transmitting device to the receiving device. This allows for the identification of pairs of original image numbers from the representative deformed images, and the generation of the encryption string is performed based on these pairs of original image numbers. This solves the problems of excessive distortion or insufficient embedded data capacity and significantly improves the ability to recover information.

[0032] See Figure 1 , Figure 1 This is a block diagram of the encryption / decryption system 100 disclosed herein. (See diagram below.) Figure 1 As shown, the encryption / decryption system 100 includes a transmitting device 110 and a receiving device 120. The transmitting device 110 and the receiving device 120 are connected. In some embodiments, the transmitting device 110 and the receiving device 120 can be connected to each other wirelessly or via a wired connection.

[0033] In some embodiments, the transmitting device 110 and the receiving device 120 may be any electronic device with communication and processing functions (e.g., a mobile phone, a laptop computer, or a tablet computer). In some embodiments, the transmitting device 110 may include a transceiver circuit 111, a memory 112, and a processor 113, and the receiving device 120 may include a transceiver circuit 121, a memory 122, and a processor 123.

[0034] In this embodiment, memory 112 pre-stores multiple original images. Processor 113 performs image deformation processing on the multiple original images in pairs to generate multiple morphed images, and records the correspondence between the numbers of the multiple original images and the numbers of the multiple morphed images in a correspondence table and stores it in memory 112. Additionally, memory 122 also stores this correspondence table. It is worth noting that the generation of the morphed images and the correspondence table will be further explained in subsequent paragraphs.

[0035] In this embodiment, transceiver circuit 111 transmits the distorted image rm1 to transceiver circuit 121.

[0036] In some embodiments, transceiver circuits 111 and 121 may be implemented using a communication interface (e.g., a Wi-Fi communication interface) for communicating with other devices or systems. In some embodiments, transceiver circuit 111 may transmit a representative morphing image ROM to transceiver circuit 121. It is worth noting that the selection of the representative morphing image ROM will be further explained in subsequent paragraphs.

[0037] In some embodiments, memory 112 and memory 122 may be implemented by storage units, flash memory, read-only memory, hard disk, or any equivalent storage component. In some embodiments, memory 112 may store an original image database (not shown), a deformed image database (not shown), and a generated correspondence table, wherein the original image database stores the original images and the deformed image database stores the generated deformed images. In some embodiments, memory 122 may also store the original image database and the correspondence table. In some embodiments, the deformed images in the deformed image database are used to update (i.e., train) the recognition model IM executed by processor 123.

[0038] In other words, the original image and its corresponding table are pre-stored in both memory 112 and memory 122. This allows for data encryption and decryption in subsequent steps.

[0039] In some embodiments, processors 113 and 123 may be implemented by a processing unit, a central processing unit, or a computing unit, etc. In some embodiments, processor 123 may run the identification model IM based on corresponding software or firmware instructions. It is worth noting that the architecture and updates of the identification model IM will be further described in subsequent paragraphs.

[0040] Refer to together Figure 2 as well as Figure 3 , Figure 2 This is a flowchart of the encryption / decryption method 200 disclosed herein. Figure 3This is a schematic diagram of an encryption / decryption system 100 implemented according to some embodiments of this disclosure. Figure 1 The components in the encryption / decryption system 100 are used to execute steps S210 to S250 of the encryption / decryption method 200. For example... Figure 2 as well as Figure 3 As shown, firstly, in step S210, the processor 113 generates the encrypted string se. i In some embodiments, memory 112 and memory 122 may share key K. h (That is, all store the same key K) h In some embodiments, processor 113 may utilize key K. h For data s i Perform an encryption algorithm to generate the encrypted string se i In some embodiments, the encryption algorithm may be a commonly used encryption and decryption algorithm such as the Advanced Encryption Standard (AES).

[0041] In an optional embodiment, step S210' can be performed after step S210. In step S210', the processor 113 performs image deformation processing on multiple original images to generate multiple deformed images.

[0042] In some embodiments, the processor 113 may obtain multiple original images from the original image database OD, and perform image deformation processing on the multiple original images in pairs to generate multiple deformed images. Then, the processor 113 may store the multiple deformed images in the deformed image database MD, and record the correspondence between the numbers of the multiple original images and the numbers of the multiple deformed images in a correspondence table. It is worth noting that "in pairs" means that before generating a deformed image, two random selections are made from these original images, wherein the same original image may be selected twice.

[0043] In some embodiments, the number of original images is N, and the number of deformed images is The original images are numbered in decimal, the deformed images are numbered in binary, and the number of bits in the deformed image numbers is... N can be any positive integer.

[0044] The following examples illustrate the generation of distorted images and their corresponding tables. (See also...) Figure 4 , Figure 4 This is a schematic diagram illustrating the generation of the correspondence table ct according to some embodiments of this disclosure. For example... Figure 4As shown, assume the original image database OD includes three original images oi1 to oi3, each with a SID of 1 to 3, and that original images oi1 to oi3 can be selected repeatedly, up to two times. In this case, there are six possible number combinations (e.g., the first selected SID is 1 and the second selected SID is 2), resulting in six deformable images, and the binary deformable image ID (MID) has three bits. Therefore, binary numbers (e.g., 001 to 110) can be assigned to each number combination of the original images in ascending order. This allows the correspondence between the number combinations of the original images and the deformable image IDs to be stored in a correspondence table ct.

[0045] Furthermore, according to the correspondence table ct, when both the first SID and the second SID are 1 (i.e., the original image oi1 with SID 1 is deformed with itself), the corresponding deformed image has the number MID 001. In other words, deforming the original image oi1 with itself produces a deformed image with the number MID 001. Similarly, when both the first SID and the second SID are 1 (i.e., the original image oi1 with SID 1 is deformed with the original image oi2 with SID 2), the corresponding deformed image has the number MID 010. In other words, deforming the original image oi1 with itself produces a deformed image with the number MID 010. This pattern continues, and the correspondence table ct records the relationship between all combinations of SIDs (first and second SIDs) and the MIDs of the deformed images.

[0046] The following provides a further detailed explanation of the steps in step S210'. Please refer to [link / reference]. Figure 5 , Figure 5 According to some embodiments of this disclosure Figure 3 A flowchart detailing the steps of step S210'. (See attached flowchart.) Figure 5 As shown, step S210' includes steps S211' to S215'.

[0047] First, in step S211', the processor 113 selects a first image and a second image from multiple original images. In other words, the processor 113 selects two images from the original image database OD. It is worth noting that although this example uses different first and second images, in actual applications, the first and second images may be the same image.

[0048] In step S212', the processor 113 performs feature point recognition processing on the first image and the second image to generate multiple first target points (lamdmarks) in the first image, multiple second target points in the second image, and the correspondence between the multiple first target points and the multiple second target points (i.e., which second target point one of the first target points corresponds to).

[0049] In some embodiments, the feature point recognition process includes a supervised descent method (SDM) algorithm and an edge detection algorithm.

[0050] Refer to together Figure 6 , Figure 6 This is a schematic diagram of target points LM(1) to LM(M) according to some embodiments of this disclosure. Figure 6 As shown, multiple target points LM(1) to LM(M) can be generated from the original image oi2, where M is the number of target points LM(1) to LM(M) and is a positive integer. Target points LM(1) to LM(M) mark all feature points on the original image oi2 (e.g., facial features, hair, and facial edges). It is worth noting that the number of target points LM(1) to LM(M) generated on different original images is fixed (e.g., 12 target points will be generated on the eyes). Therefore, there is a correspondence between the target points LM(1) to LM(M) generated on different original images (e.g., the target points of the eyes on different original images will correspond to each other).

[0051] like Figure 5 As shown, in step S213', the processor 113 generates the coordinates of multiple deformation points from the coordinates of the multiple first target points and the multiple second target points according to the adjustment parameters and the correspondence between the multiple first target points and the multiple second target points. In some embodiments, the calculation of the coordinates of the deformation points is as shown in the following formula (1).

[0052] Cw=α×Cs+(1-α)×Ct……Formula (1)

[0053] Where Cw is the coordinate of the deformation point, Cs is the coordinate of the first target point, Ct is the coordinate of the second target point corresponding to Cs, and α is the adjustment parameter, where α is a value greater than or equal to 0 and less than or equal to 1.

[0054] In step S214', the processor 113 performs alignment processing on the first image based on the coordinates of multiple first target points and multiple deformation points to generate a first warped image, and performs alignment processing on the second image based on the coordinates of multiple second target points and multiple deformation points to generate a second warped image.

[0055] In some embodiments, the processor 113 may divide the first image into multiple first regions based on the coordinates of multiple first target points (e.g., dividing a triangular first region using the coordinates of three first target points), and divide the first image into multiple first deformed regions based on the coordinates of multiple deformable points (e.g., dividing a triangular first deformed region using the coordinates of three deformable points). Then, it calculates a first distance difference (e.g., the distance difference between the centroids of the regions) between each of the multiple first regions and its corresponding first deformed region. Next, the processor 113 may adjust the positions of the multiple first regions in the first image based on the multiple first distance differences to generate a first distorted image (i.e., the alignment process described above).

[0056] In some embodiments, the processor 113 may divide the second image into multiple second regions based on the coordinates of multiple second target points (e.g., dividing a triangular second region using the coordinates of three second target points), and divide the second image into multiple second deformed regions based on the coordinates of multiple deformed points (e.g., dividing a triangular second deformed region using the coordinates of three deformed points). Then, it may calculate a second distance difference (e.g., the distance difference between the centroids of the regions) between each of the multiple second regions and its corresponding second deformed region. Next, the processor 113 may adjust the positions of the multiple second regions in the second image based on the multiple second distance differences to generate a second distorted image.

[0057] In step S215', the processor 113 generates one of a plurality of deformed images from the first distorted image and the second distorted image according to the adjustment parameters.

[0058] In some embodiments, the processor 113 may superimpose the first distorted image and the second distorted image together, and adjust the pixel values ​​in the superimposed image using adjustment parameters to generate a deformed image, which is then stored in the deformed image database MD. In some embodiments, the pixel values ​​in the deformed image are as shown in the following formula (2).

[0059] Im(x,y)=α×Iws(x,y)+(1-α)×Iwt(x,y)…Formula (2)

[0060] Where x is the coordinate in the x-direction, y is the coordinate in the y-direction, Im(x,y) is the pixel quality of coordinate (x,y) in the deformed image, Iws(x,y) is the pixel quality of coordinate (x,y) in the first image, Iwt(x,y) is the pixel quality of coordinate (x,y) in the second image, and α is also the adjustment parameter mentioned above.

[0061] The numbering of the first image and the numbering of the second image constitute a numbering combination. New numbering combinations can be selected to generate new deformed images using the same adjustment parameters until no new numbering combinations are generated.

[0062] The image deformation processing described above is merely an exemplary embodiment. However, in practical applications, other known image deformation processing methods (e.g., image fusion algorithms) can also be used without particular limitation. This allows for the generation of a large number of deformed images from a small number of original images.

[0063] The following practical examples illustrate the impact of parameter adjustments on deformed images.

[0064] Refer to together Figure 7 , Figure 7 This is a schematic diagram illustrating the effect of adjustment parameter α on deformed image mi in some embodiments according to this disclosure. For example... Figure 7 As shown, firstly, the original images oi1 to oi2 are selected for image deformation processing to generate deformed images mi, where the original images oi1 to oi2 are numbered 1 and 2 respectively, and the deformed image mi is numbered 010.

[0065] Depend on Figure 7 It can be seen that the deformed images mi are generated by adjusting the parameter α to 0, 0.1, and 0.2 to 1 respectively. Based on this, it can be seen that the adjustment parameter α affects the deformed image mi. When the original images oi1 to oi2 are used as the first and second images mentioned above, respectively, the smaller the adjustment parameter α, the more similar the deformed image mi will be to the original image oi1. Conversely, the larger the adjustment parameter α, the more similar the deformed image mi will be to the original image oi2. Therefore, an adjustment parameter α can be selected to establish the entire deformed image database MD. In other words, the processor 113 will only select one adjustment parameter to generate all deformed images.

[0066] like Figure 2 as well as Figure 3 As shown, in step S220, the processor 113 determines the encrypted string se i A representative deformable image RMI is selected from multiple deformable images (i.e., data hiding). In some embodiments, the number of the representative deformable image RMI is the same as the encrypted string se. i For example, suppose the encrypted string is sei If the value is 010, the processor 113 can select the deformed image numbered 010 from the deformed image database MD as the representative deformed image rmi.

[0067] In step S230, the representative deformed image rmi is transmitted to the transceiver circuit 121 via the transceiver circuit 111, without transmitting the encrypted string se. i To transceiver circuit 121. In other words, transceiver circuit 111 only transmits a representation of the deformed image rmi to transceiver circuit 121, and does not directly transmit the encrypted string se. i Transmitted to transceiver circuit 121.

[0068] In step S240, the processor 123 identifies the first original image number SID1 and the second original image number SID2 from the representative deformed image rmi.

[0069] In some embodiments, the processor 123 may use the recognition model IM to perform image recognition on the representative deformed image rmi to generate a first original image number SID1 and a second original image number SID2. In some embodiments, the recognition model IM may be a version 2 morphed face recognition network (MFR-NET V2), which includes a feature extraction layer and two fully connected layers. The input of the two fully connected layers is connected to the output of the feature extraction layer, which is a deep residual neural network (ResNet). The fully connected layers are fully connected L2 normalization (FC L2 norm) layers (i.e., a fully connected layer plus an L2 normalization layer). Although the version 2 morphed face recognition network is used as an example here, in practical applications, the version 1 morphed face recognition network (MFR-NET V1) may also be used.

[0070] In some embodiments, the processor 123 may perform image deformation processing on multiple original images in pairs according to multiple adjustment parameters to generate multiple corresponding deformed images. Then, the processor 123 may update the recognition model IM using the multiple corresponding deformed images and the multiple original images, so as to use the recognition model IM to identify the first original image number SID1 and the second original image number SID2 from the representative deformed image rmi.

[0071] In some embodiments, the processor 123 may select a first image and a second image from multiple original images, and perform image deformation processing on the first image and the second image according to adjustment parameters to generate corresponding deformed images. Next, the processor 123 may use the corresponding deformed images as training samples, and the first image and the second image as first training labels and second training labels. Then, the receiving device 120 inputs the training samples into the recognition model IM to generate first predicted labels and second predicted labels, and uses the first predicted labels, second predicted labels, first training labels, and second training labels to calculate additive angular spacing loss (ArcFace loss) to generate a loss. Next, the processor 123 may use the loss to perform back propagation processing on the recognition model IM to update the recognition model IM.

[0072] Furthermore, the numbering of the first image and the numbering of the second image form a numbering combination, and new numbering combinations can be selected to update the recognition model IM until no new numbering combinations are generated.

[0073] It is worth noting that the image deformation processing here is the same as that described in the previous paragraphs; therefore, it will not be elaborated upon further. Furthermore, during the training phase of the aforementioned recognition model IM, multiple deformable image sets are generated for various adjustment parameters (e.g., 0.1, 0.3, 0.5, 0.7, and 0.9). More specifically, for the first adjustment parameter, a first deformable image set can be generated from all numbered combinations, where the deformable image set includes multiple deformable images corresponding to the first adjustment parameter. Similarly, for the second adjustment parameter, a second deformable image set can be generated from all numbered combinations, where the second deformable image set includes multiple deformable images corresponding to the second adjustment parameter. This process can be repeated to generate other deformable image sets.

[0074] The following example illustrates the training phase of the IM recognition model.

[0075] Refer to together Figure 8 , Figure 8 This is a schematic diagram of the updated identification model IM according to some embodiments of this disclosure. For example... Figure 8As shown, taking the original images oi1-oi2 with SIDs 1-2 and the deformed image mi with MID 010 as an example, with the adjustment parameter α set to 0.1, the deformed image mi can be used as a training sample, and images 1-2 (SIDs) as training labels lbl. Next, the deformed image mi can be input into the feature extraction layer FEL to generate a feature vector FV. The feature vector FV is then input into the fully connected layers FCL1-FCL2 to generate two predicted labels. These two predicted labels are then used to calculate the additive angular interval loss on the two training labels lbl (i.e., 1-2) to generate a loss. Finally, the loss can be used to perform backpropagation processing on the fully connected layers FCL1-FCL2 and the feature extraction layer FEL to update the parameters in these layers.

[0076] like Figure 2 as well as Figure 3 As shown, in step S250, the processor 123 looks up the corresponding table based on the first original image number SID1 and the second original image number SID2 to generate the encrypted string se. i (i.e., data analysis).

[0077] For example, with Figure 4 Taking the corresponding table ct as an example. Assume the first original image number SID1 and the second original image number SID2 are 1 and 2 respectively. By looking up the corresponding table ct, we can find that the corresponding deformed image number is 010. Therefore, we can determine the encrypted string se. i It's 010.

[0078] like Figure 2 as well as Figure 3 As shown, in some embodiments, when the encrypted string se is generated by looking up the corresponding table ct based on the first original image number SID1 and the second original image number SID2, i At that time, processor 123 can utilize key K h For the encrypted string se i Perform a decryption algorithm to generate data s i In some embodiments, the decryption algorithm may also be a commonly used encryption and decryption algorithm such as the Advanced Encryption Standard (AES).

[0079] Unlike traditional encryption and decryption methods, the transmitting end disclosed herein hides the data within the distorted image's identifier and does not transmit the identifier itself. The receiving end will use image recognition and table lookup to find the distorted image's identifier and use it as the data originally intended to be transmitted by the transmitting end.

[0080] In summary, the encryption / decryption system and method disclosed herein significantly increase the number of distorted images by utilizing image warping and pre-establishing a correspondence table between distorted and original images. This allows the transmitting end to hide data within the distorted image's identifier without transmitting the identifier itself. Therefore, network attackers cannot directly determine where the data is hidden. Furthermore, the receiving end can easily decipher the data originally intended by the transmitting end simply by combining image recognition with table lookup. This greatly reduces resource and time consumption, solves the problems of excessive distortion or insufficient embedded data capacity, and significantly improves the ability to recover information.

[0081] While specific embodiments of the present disclosure have been disclosed in relation to the above embodiments, these embodiments are not intended to limit the present disclosure. Various alternatives and modifications can be made by those skilled in the art without departing from the principles and spirit of the present disclosure. Therefore, the scope of protection of the present disclosure is determined by the appended claims.

[0082] [Symbol Explanation]

[0083] 100: Encryption / Decryption System

[0084] 200: Encryption / Decryption Method

[0085] 110: Transmission equipment

[0086] 120: Receiving equipment

[0087] 111, 121: Transceiver circuits

[0088] 112, 122: Memory

[0089] 113, 123: Processor

[0090] IM: Recognition Model

[0091] OD: Original Image Database

[0092] MD: Deformable Image Database

[0093] s i :data

[0094] se i : Encrypted string

[0095] K h Key

[0096] SID1: First original image number

[0097] SID2: Second Original Image Number

[0098] SID: Original Image ID

[0099] MID: Deformed Image Number

[0100] rmi: Represents distorted image

[0101] oi1~oi3: Original images

[0102] mi: distorted image

[0103] ct: Correspondence Table

[0104] LM(1)~LM(M): Target point

[0105] M: Number of target points

[0106] α: Adjust parameters

[0107] FEL: Feature Extraction Layer

[0108] FV: Feature Vector

[0109] FCL1, FCL2: Fully Connected Layers

[0110] lbl: training labels

[0111] S210~S250, S210', S211'~S215': Steps.

Claims

1. An encryption / decryption system, characterized in that, include: Transmission equipment, including: The first memory is used to store multiple original images; A first processor is configured to perform image warping processing on the plurality of original images in pairs to generate a plurality of warped images, and to record the correspondence between the numbers of the plurality of original images and the numbers of the plurality of warped images in a correspondence table for storage in the first memory. The first processor is further configured to generate an encrypted string and select a representative warped image from the plurality of warped images based on the encrypted string, wherein the number of the representative warped image is the same as the encrypted string. A first transceiver circuit is configured to transmit the representative distorted image but not the encrypted string; and A receiving device, connected to the transmitting device, includes: The second memory is used to store the correspondence table; A second transceiver circuit is configured to receive the representative distorted image from the first transceiver circuit, but not to receive the encrypted string; and The second processor identifies the first original image number and the second original image number from the representative deformed image recognition, and looks up the corresponding table according to the first original image number and the second original image number to generate the encrypted string.

2. The encryption / decryption system according to claim 1, wherein... The first memory and the second memory share a key. The first processor uses the key to perform an encryption algorithm on the data to generate the encrypted string, and When the encrypted string is generated by looking up the corresponding table based on the first original image number and the second original image number, the second processor uses the key to perform a decryption algorithm on the encrypted string to generate the data.

3. The encryption / decryption system according to claim 1, wherein... The first processor selects a first image and a second image from the plurality of original images. The first processor performs feature point recognition processing on the first image and the second image to generate multiple first target points in the first image and multiple second target points in the second image, and establishes a target point correspondence relationship between the multiple first target points and the multiple second target points. The first processor generates the coordinates of multiple deformation points from the coordinates of the multiple first target points and the coordinates of the multiple second target points based on the adjustment parameters and the correspondence of the target points. The first processor aligns the first image based on the coordinates of the plurality of first target points and the coordinates of the plurality of deformation points to generate a first distorted image, and aligns the second image based on the coordinates of the plurality of second target points and the coordinates of the plurality of deformation points to generate a second distorted image. The first processor converts the first distorted image and the second distorted image into corresponding deformed images according to the adjustment parameters.

4. The encryption / decryption system according to claim 3, wherein The first processor records the correspondence between the first image and the second image and the corresponding deformed image in the correspondence table.

5. The encryption / decryption system according to claim 1, wherein... The second processor performs image deformation processing on the multiple original images in pairs according to multiple adjustment parameters to generate multiple corresponding deformed images, and The second processor updates the recognition model using the plurality of corresponding deformed images and the plurality of original images, so as to use the recognition model to identify the first original image number and the second original image number from the representative deformed image.

6. The encryption / decryption system according to claim 1, wherein... The second processor selects a first image and a second image from the plurality of original images, and performs image deformation processing on the first image and the second image according to adjustment parameters to generate a corresponding deformed image. The second processor uses the corresponding deformed image as a training sample, and uses the first image and the second image as the first training label and the second training label, respectively. The second processor inputs the training samples into the recognition model to generate a first predicted label and a second predicted label, and uses the first predicted label, the second predicted label, the first training label, and the second training label to calculate an additive angular interval loss to generate a loss. The second processor uses the loss to perform backpropagation processing on the recognition model to update the recognition model.

7. The encryption / decryption system according to claim 6, wherein the recognition model is a version 2 deformable face recognition network, the version 2 deformable face recognition network includes a feature extraction layer and a 2 fully connected layer, wherein the input end of the 2 fully connected layer is connected to the output end of the feature extraction layer, the feature extraction layer is a deep residual neural network, and the 2 fully connected layer is a 2 fully connected L2 normalization layer.

8. The encryption / decryption system according to claim 1, wherein the number of the plurality of original images is N, and the number of the plurality of deformed images is And the number of bits in the number of the plurality of deformed images is .

9. The encryption / decryption system according to claim 1, wherein the numbers of the plurality of original images are in decimal and the numbers of the plurality of deformed images are in binary.

10. An encryption / decryption method, characterized in that, include: An encrypted string is generated, and multiple deformed images are generated by performing image deformation processing on multiple original images in pairs. The correspondence between the numbers of the multiple original images and the numbers of the multiple deformed images is then recorded in a correspondence table. A representative deformed image is selected from the plurality of deformed images according to the encrypted string, wherein the number of the representative deformed image is the same as the encrypted string; Transmit the distorted image, but do not transmit the encrypted string; When the representative deformed image is received, the first original image number and the second original image number are identified from the representative deformed image. as well as The encrypted string is generated by looking up the corresponding table based on the first original image number and the second original image number.

11. The encryption / decryption method according to claim 10, further comprising: The data is encrypted using a shared key to generate the encrypted string; as well as When the encrypted string is generated by looking up the corresponding table based on the first original image number and the second original image number, the encrypted string is decrypted using the key to generate the data.

12. The encryption / decryption method according to claim 10, further comprising: Select a first image and a second image from the plurality of original images; Feature point recognition processing is performed on the first image and the second image to generate multiple first target points in the first image and multiple second target points in the second image, and a target point correspondence relationship is established between the multiple first target points and the multiple second target points; Based on the adjustment parameters and the correspondence of the target points, the coordinates of multiple deformation points are generated from the coordinates of the multiple first target points and the coordinates of the multiple second target points. The first image is aligned using the coordinates of the plurality of first target points and the coordinates of the plurality of deformation points to generate a first distorted image; and the second image is aligned using the coordinates of the plurality of second target points and the coordinates of the plurality of deformation points to generate a second distorted image; and The first distorted image and the second distorted image are converted into corresponding deformed images according to the adjustment parameters.

13. The encryption / decryption method according to claim 12, further comprising: The correspondence between the first image and the second image and the corresponding deformed image is recorded in the correspondence table.

14. The encryption / decryption method according to claim 10, further comprising: Based on multiple adjustment parameters, the multiple original images are paired and image deformation processing is performed to generate multiple corresponding deformed images, and The recognition model is updated using the plurality of corresponding deformed images and the plurality of original images, so as to identify the first original image number and the second original image number from the representative deformed image using the recognition model.

15. The encryption / decryption method according to claim 10, further comprising: A first image and a second image are selected from the plurality of original images, and image deformation processing is performed on the first image and the second image according to adjustment parameters to generate corresponding deformed images. The corresponding deformed images are used as training samples, and the first image and the second image are used as the first training label and the second training label, respectively. The training samples are input into the recognition model to generate a first predicted label and a second predicted label. An additive angular interval loss is then calculated using the first predicted label, the second predicted label, the first training label, and the second training label to generate a loss. The recognition model is updated by backpropagating the loss.

16. The encryption / decryption method according to claim 15, wherein the recognition model is a version 2 deformable face recognition network, the version 2 deformable face recognition network includes a feature extraction layer and a 2 fully connected layer, wherein the input end of the 2 fully connected layer is connected to the output end of the feature extraction layer, the feature extraction layer is a deep residual neural network, and the 2 fully connected layer is a 2 fully connected L2 normalization layer.

17. The encryption / decryption method according to claim 10, wherein the number of the plurality of original images is N, and the number of the plurality of deformed images is And the number of bits in the number of the plurality of deformed images is .

18. The encryption / decryption method according to claim 10, wherein the numbers of the plurality of original images are in decimal and the numbers of the plurality of deformed images are in binary.

Images