High-confidentiality scene confidentiality operation video security management method

Abstract

The invention relates to a high-confidentiality scene confidentiality operation video security management method, and relates to the technical field of security. The method comprises the following steps: identifying and acquiring identity information of a person executing the confidentiality operation in the confidentiality operation of a high-confidentiality scene; in a high-confidentiality scene, a video recorder executes confidentiality operation; the method comprises the following steps: processing a video recording executed secrecy operation through a pre-trained information steganography model, steganography secrecy operation information containing personnel identity information into a video frame of the video, obtaining a traceable video frame containing the secrecy operation information, and replacing an original video frame with the traceable video frame to obtain a traceable video; and identity information and video processing operation of video processing personnel are obtained, and the identity information and the video processing operation are incrementally steganographically written into video frames of the processed traceable video through the information steganography model. And decoding confidential operation information from the traceable video frame through a pre-trained information steganography model to realize high-confidentiality scene confidential operation video security management.

Description

Technical Field

[0001] The present invention relates to the technical field of video processing, and in particular to a video security management method for confidential operation videos in high-security scenarios. Background Art

[0002] Confidential operations occurring in high-security scenarios require strict monitoring. Currently, the monitoring videos of confidential operations in high-security scenarios do not contain any information about the confidential operation personnel, nor can they contain any information about the users who process the monitoring videos. It is impossible to determine which user's operation led to the leakage through the leaked videos, and the source of the leakage cannot be traced after the monitoring videos are leaked. The existing monitoring videos in high-security scenarios use watermark tracing, randomly attaching the user's ID and operation time to the execution positions of certain video frames using watermarks. When the auditor downloads the video, through the function of video transcoding, the auditor's user ID is randomly attached to the specified positions of certain video frames using watermarks. The purpose of video leakage tracing is achieved by adding user information to certain video frames of the confidential operation monitoring video. However, the watermark is visible and can easily be removed by technicians, and ultimately tracing cannot be achieved.

[0003] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present invention provides a video security management method for confidential operation videos in high-security scenarios.

[0005] The present invention provides a video security management method for confidential operation videos in high-security scenarios, including: In a high-security scenario, identify the identity of the confidential operation personnel and obtain the identity information of the confidential operation personnel; Use the monitoring video to record the confidential operations performed by the confidential operation personnel; Through a pre-trained information steganography model, hide the confidential operation information including the identity information of the confidential operation personnel into the video frames of the video, obtain the traceable video frames containing the confidential operation information, and replace the original video frames with the traceable video frames to obtain a traceable video; For the traceable video, before the traceable video management personnel process the traceable video, obtain the identity information of the video processing personnel and their video processing operations, and incrementally hide them into the video frames of the processed traceable video through the information steganography model; Decode the confidential operation information from the traceable video frames through a pre-trained information steganography model to achieve video security management for confidential operation videos in high-security scenarios; Among them, the information steganography model includes: an information graphic code preprocessing module based on a reversible neural network and an information graphic code steganography module; the forward process of the information graphic code preprocessing module is used to encode the spliced information graphic code, downsample the spliced information graphic code into a reduced form that supports hiding in video frames and obtain the corresponding first key map; the reverse process of the information graphic code preprocessing module is used to upsample the reduced form of the spliced information graphic code decoded by the information graphic code steganography module in combination with the first key map to restore the decoded spliced information graphic code; the reduced form of the spliced information graphic code and the video frame to be processed are input into the information graphic code steganography module, and the information graphic code steganography module performs discrete wavelet transform on the reduced form and the video frame to the frequency domain, and then the frequency domain data is input into the reversible neural network of the information graphic code steganography module for steganography to obtain the second key map and the frequency domain data of the video frame with the information graphic code, and then the second key map and the traceable video frame are obtained through inverse frequency domain transformation respectively.

[0006] Furthermore, multiple sampled video frames are extracted from the entire time span of the video recording the confidentiality operation through segmented sampling, and the information steganography model performs confidentiality operation information steganography on each sampled video frame to obtain the sampled traceable video frame, and replaces the original corresponding sampled video frame with the sampled traceable video frame to obtain the traceable video.

[0007] Furthermore, a two-dimensional code generator is used to convert the confidentiality operation information into the corresponding information graphic code, and at least one information graphic code written into the video frame is spliced together according to a set format to obtain the spliced information graphic code.

[0008] Furthermore, when the number of information graphic codes written into the video frame is less than the number of information graphic codes set by the format, the replication result of the information graphic code written into the video frame is filled to meet the format setting.

[0009] Furthermore, the forward process of the reversible neural network of the information graphic code preprocessing module is as follows: The feature map of the spliced information graphic code input into any k-th reversible neural module is divided into sub-feature maps along the channel dimension and the sub-feature maps , and the sub-feature maps are calculated by the first residual dense convolutional network for transformation and added to the sub-feature maps for combination, and the result of the addition and combination is used as the input feature map of the subsequent (k + 1)-th reversible neural module that makes up the information graphic code preprocessing module ; the sub-feature maps In the k-th reversible neural module, the scaling rotation factor is also learned through the second residual dense convolutional network and the scalable Sigmoid activation function respectively, and the translation factor is learned through the third residual dense convolutional network. The sub-feature map is multiplied by the scaling rotation factor and then added to the translation factor. The result of the addition combination is used as the feature map of the subsequent (k + 1)-th reversible neural module that constitutes the information graphic code preprocessing module of the sub-feature map .

[0010] Furthermore, the reverse process of the reversible neural network of the information graphic code preprocessing module is as follows: The feature map of the concatenated information graphic code input to any (k + 1)-th reversible neural module of the information graphic code preprocessing module is split into sub-feature maps along the channel dimension and sub-feature maps . The sub-feature map is used to learn the scaling rotation factor through the second residual dense convolutional network and the scalable Sigmoid activation function respectively, and the translation factor is learned through the third residual dense convolutional network. The sub-feature map is divided by the scaling rotation factor after subtracting the translation factor, and is used as the sub-feature map of the k-th reversible neural module of the information graphic code preprocessing module ; The sub-feature map subtracts the transformation result calculated by the first residual dense convolutional network from the sub-feature map to be used as the sub-feature map of the k-th reversible neural module of the information graphic code preprocessing module .

[0011] Furthermore, the forward process of the reversible neural network of the information graphic code steganography module is as follows: The reduced form of the concatenated information graphic code input to the first reversible neural module of the information graphic code steganography module and the video frame are respectively transformed to the frequency domain through frequency domain transformation, and then the feature map of the reduced form of the concatenated information graphic code input to any k-th reversible neural module of the information graphic code steganography module is calculated by the fourth residual dense convolutional network for transformation and then added to the feature map of the video frame . The result of the addition combination is used as the feature map of the video frame of the subsequent (k + 1)-th reversible neural module that constitutes the information graphic code steganography module ; The feature map of the video frame In the k-th reversible neural module of the information graphic code steganography module, the non-scaling rotation factor is also learned through the fifth residual dense convolutional network and the scalable Sigmoid activation function, and the translation factor is learned through the sixth residual dense convolutional network. The feature map of the spliced information graphic code is multiplied by the non-scaling rotation factor and then added to the translation factor as the feature map of the spliced information graphic code of the subsequent (k + 1)-th reversible neural module that constitutes the information graphic code steganography module .

[0012] Furthermore, the reverse process in the reversible neural network of the information graphic code steganography module is as follows: The second key map and the traceable video frame input to the information graphic code steganography module are transformed to the frequency domain through frequency domain transformation processing, and the feature map of the traceable video frame in the frequency domain received by the final K-th reversible neural module and the feature map of the second key map in the frequency domain ; Then, the feature map of the traceable video frame in the frequency domain input to any (k + 1)-th reversible neural module of the information graphic code steganography module respectively calculates the translation factor through the sixth residual dense convolutional network, and calculates the non-scaling rotation factor through the fifth residual dense convolutional network and the scalable Sigmoid activation function; the feature map of the spliced information graphic code in the frequency domain subtracts the translation factor and then divides by the non-scaling rotation factor, and the result is used as the feature map of the spliced information graphic code of the k-th reversible neural module of the information graphic code steganography module : The feature map subtracts the feature map transformed by calculating through the fourth residual dense convolutional network and is used as the feature map of the spliced information graphic code of the k-th reversible neural module of the information graphic code steganography module .

[0013] Furthermore, the implementation of the scalable Sigmoid activation function is as follows: The Sigmoid activation function compresses the output of the residual dense convolutional network to the range of (0, 1), and then translates it to the range of (-0.5, 0.5) by subtracting 0.5, so that the final scaling or non-scaling rotation factor is symmetric about 0. Then, the output value is scaled by the scaling or non-scaling coefficient and the exponential function of the natural exponential e is taken to obtain the scaling or non-scaling rotation factor.

[0014] Furthermore, during training, the learning objective of the information steganography model includes minimizing the following loss terms: The difference between the video frame to be processed and the traceable video frame; The difference between the frequency domain of the video frame to be processed and the frequency domain of the traceable video frame; The difference between the reduced form of the spliced information graphic code encoded by the forward process of the information graphic code steganography module and the prediction of the reduced form of the spliced information graphic code restored from the traceable video frame by combining the reverse process of the information graphic code steganography module with the second key graph; The difference between the spliced information graphic code decoded by the reverse process of the information graphic code steganography module and the original information graphic code.

[0015] The above technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art: In the information steganography model of the present application, the forward process of the information graphic code preprocessing module is used to encode the spliced information graphic code, downsample the spliced information graphic code into a reduced form that supports hiding in the video frame, and obtain the corresponding first key graph; the reverse process of the information graphic code preprocessing module is used to upsample the reduced form of the spliced information graphic code decoded by the information graphic code steganography module in combination with the first key graph to restore the decoded spliced information graphic code; the reduced form of the spliced information graphic code and the video frame to be processed are input into the information graphic code steganography module. The information graphic code steganography module performs discrete wavelet transform on the reduced form and the video frame to the frequency domain, and then the frequency domain data is input into the reversible neural network of the information graphic code steganography module for steganography to obtain the second key graph and the frequency domain data of the video frame with the information graphic code. Then, the second key graph and the traceable video frame are obtained through inverse frequency domain transformation respectively. Through the forward and reverse processes of the information steganography model, the confidentiality operation information is effectively recorded into the video frame of the confidentiality operation video in a high-confidentiality scenario, and the confidentiality operation video in the high-confidentiality scenario can be implicitly traced. The steganography and decoding processes of the confidentiality operation information depend on the information steganography model, the first key graph, and the second key graph, and are not easily cracked. The present application hides the original confidentiality operation and subsequent video processing operations into the traceable video, which is convenient for tracing the leakage of secrets. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of a method for managing the security of a confidentiality operation video in a high-confidentiality scenario provided by an open embodiment of the present invention.

[0019] Figure 2 Schematic diagram of converting confidential operation information into information graphic codes provided by the disclosed embodiments of the present invention.

[0020] Figure 3 Schematic diagram of the overall architecture of the information steganography model provided by the disclosed embodiments of the present invention.

[0021] Figure 4 Schematic diagram of the forward process of the information graphic code preprocessing module provided by the disclosed embodiments of the present invention.

[0022] Figure 5 Schematic diagram of the reverse process of the information graphic code preprocessing module provided by the disclosed embodiments of the present invention.

[0023] Figure 6 Schematic diagram of the forward process of the information graphic code steganography module provided by the disclosed embodiments of the present invention.

[0024] Figure 7 Schematic diagram of the reverse process of the information graphic code steganography module provided by the disclosed embodiments of the present invention.

[0025] Figure 8 Schematic diagram of the residual dense convolutional network provided by the disclosed embodiments of the present invention.

[0026] Figure 9 Schematic diagram of the secure operation video security management device for high-security scenarios provided by the disclosed embodiments of the present invention. Detailed implementation manners

[0027] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitations, the element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0029] Embodiment 1 Refer toFigure 1 As shown in the figure, an embodiment of the present invention provides a video security management method for confidential operation in a high-security scenario, including: S100, in a high-security scenario, identify the identity of the confidential operation personnel, obtain the identity information of the confidential operation personnel, and the identity information of the confidential operation personnel includes: the name and job number of the confidential operation personnel.

[0030] S200, use the surveillance video to record the confidential operations performed by the confidential operation personnel.

[0031] S300, process the video recording the confidential operation through a pre-trained information steganography model, hide the confidential operation information including the identity information of the confidential operation personnel into the video frames of the video, obtain the traceable video frames containing the confidential operation information, and replace the original video frames with the traceable video frames to obtain a traceable video.

[0032] S400, this application supports steganography for secondary processing operations of the traceable video. For the traceable video, before the traceable video management personnel process the traceable video, obtain the identity information of the video processing personnel and their video processing operations, and incrementally hide them into the video frames of the traceable video to be processed through the information steganography model. In the specific implementation process, before the authorized traceable video management personnel process the traceable video, obtain the identity information and operation information of the management personnel, and incrementally hide and update the identity information and operation information of the management personnel into the video frames of the current traceable video through the pre-trained information steganography model. The traceable video after the information is updated replaces the current traceable video and is provided to the management personnel who initiate the operation. The incremental steganography update is achieved by adding new information graphic codes to the spliced information graphic codes.

[0033] In one implementation, in order to reduce the video frame processing volume of the information steganography model, the video recording the confidential operation is segmented and sampled, and multiple sampled video frames for hiding the confidential operation information are extracted from the entire time span of the video recording the confidential operation through segmented sampling. The information steganography model performs steganography of the confidential operation information on each sampled video frame to obtain the sampled traceable video frames, and replaces the original corresponding sampled video frames with the sampled and steganographed traceable video frames to obtain a traceable video of the video recording the confidential operation.

[0034] For the case of segmented sampling, obtain the sampled traceable video frames according to the sampled video frame index, and the information steganography model decodes the confidential operation information for each sampled traceable video frame.

[0035] The information steganography model realizes the steganography and decoding of the confidential operation information. The working principle of the information steganography model is as follows: This application first uses a QR code generator to convert the confidential operation information into corresponding information graphic codes. Such asFigure 2 As shown, the confidentiality operation information includes the identity information of the confidentiality operator, the confidentiality operation time, and the planned confidentiality operation. The confidentiality operation information containing the identity information of the confidentiality operator is used to generate a corresponding information graphic code in the form of a two-dimensional code through a two-dimensional code generator. At least one information graphic code written into the video frame is spliced together according to a set format to obtain a spliced information graphic code. For the convenience of subsequent description, the spliced information graphic code is defined as . When the number of information graphic codes written into the video frame is less than the number of information graphic codes set by the format, the replication result of the information graphic codes written into the video frame is filled to meet the format setting. For example, if the format is set to a spliced information graphic code composed of 5×5 information graphic codes and the number of information graphic codes written into the video frame is 5, then the 5 information graphic codes written into the video frame are copied 5 times and filled into the spliced information graphic code.

[0036] This application constructs and trains the information steganography model. As Figure 3 shown, the information steganography model includes: an information graphic code preprocessing module and an information graphic code steganography module based on a reversible neural network.

[0037] In the specific implementation process, as Figure 4 and Figure 5 shown, the information graphic code preprocessing module is implemented by a reversible neural network formed by stacking reversible neural modules, and the information graphic code preprocessing module includes a forward process and a reverse process.

[0038] The forward process of the information graphic code preprocessing module is used to encode the spliced information graphic code, downsample the spliced information graphic code into a reduced form that supports hiding in the video frame and obtain a corresponding first key map. The reverse process of the information graphic code preprocessing module is used to upsample the reduced form of the spliced information graphic code decoded by the information graphic code steganography module in combination with the first key map to restore the decoded spliced information graphic code.

[0039] The forward process of the reversible neural network of the information graphic code preprocessing module is as Figure 4 shown. The feature map of the spliced information graphic code input to any k-th reversible neural module is divided into sub-feature maps and sub-feature maps along the channel dimension. The sub-feature map is calculated by a first residual dense convolutional network and combined with the sub-feature map by addition. The result of the addition combination is used as the sub-feature map of the feature map of the subsequent (k + 1)-th reversible neural module that makes up the information graphic code preprocessing module : , where is the first residual dense convolutional network in the k-th invertible neural module. The sub-feature map also learns the scaling and rotation factors through the second residual dense convolutional network and the scalable Sigmoid activation function respectively in the k-th invertible neural module, and learns the translation factor through the third residual dense convolutional network. The sub-feature map is multiplied by the scaling and rotation factors and then added to the translation factor. The result of the addition is used as the sub-feature map of the subsequent (k + 1)-th invertible neural module that constitutes the information graphic code preprocessing module : : ; where is the second residual dense convolutional network in the k-th invertible neural module of the information graphic code preprocessing module, is the third residual dense convolutional network in the k-th invertible neural module of the information graphic code preprocessing module. The implementation of the scalable Sigmoid activation function is as follows: the output of the second residual dense convolutional network is compressed to the range (0, 1) by the Sigmoid activation function, and then translated to the range (-0.5, 0.5) by subtracting 0.5, so that the final scaling and rotation factor is symmetric about 0. Then, through the scaling coefficient the output value is scaled and then the exponential function of the natural exponential e is taken to obtain the scaling and rotation factor, and the scaling and rotation factor is positive and the range is controlled.

[0040] In the forward process of the invertible neural network of the information graphic code preprocessing module, the first invertible neural module in the information graphic code preprocessing module receives the spliced information graphic code, and the sub-feature map output by the final K-th invertible neural module in the information graphic code preprocessing module is used as the reduced form of the final spliced information graphic code , and the sub-feature map output by the final K-th invertible neural module in the information graphic code preprocessing module is used as the first key map of the corresponding reduced form .

[0041] The reverse process in the invertible neural network of the information graphic code preprocessing module is as Figure 5 shown. The feature map of the spliced information graphic code input to any (k + 1)-th invertible neural module of the information graphic code preprocessing module is split into the sub-feature map and the sub-feature map along the channel dimension. The sub-feature map The scaling and rotation factors are learned through the second residual dense convolutional network and the scalable Sigmoid activation function respectively, and the translation factor is learned through the third residual dense convolutional network. The sub-feature map is divided by the scaling and rotation factor after subtracting the translation factor, and used as the sub-feature map of the k-th reversible neural module in the information graphic code preprocessing module : ; The sub-feature map subtracts the sub-feature map The transformation result calculated through the first residual dense convolutional network is used as the sub-feature map of the k-th reversible neural module in the information graphic code preprocessing module : .

[0042] In the reverse process of the reversible neural network of the information graphic code preprocessing module, the final k-th reversible neural module in the information graphic code preprocessing module receives the reduced form of the spliced information graphic code decoded by the information graphic code steganography module , and uses it as the sub-feature map of the feature map of the k-th reversible neural module in the information graphic code preprocessing module , and uses the first key map as the sub-feature map of the feature map of the k-th reversible neural module in the information graphic code preprocessing module to perform the reverse process. The first reversible neural module in the information graphic code preprocessing module outputs the decoded spliced information graphic code.

[0043] The reduced form of the spliced information graphic code obtained by the conversion of the information graphic code preprocessing module and the video frame to be processed are input into the information graphic code steganography module. The information graphic code steganography module first performs discrete wavelet transform on the reduced form of the spliced information graphic code and the video frame to the frequency domain. Then the frequency domain data is input into the reversible neural network of the information graphic code steganography module for steganography to obtain the second key map and the frequency domain data of the video frame with the information graphic code. Then, the second key map and the traceable video frame are obtained through inverse frequency domain transformation respectively.

[0044] As Figure 8 shown, the structures of the residual dense convolutional networks involved in this application are the same. It is implemented by convolutional blocks, and each convolutional block contains a convolutional layer, a LeakyReLU activation function, and a concatenation operation that performs residual and dense concatenation on the output of the convolutional block.

[0045] In the specific implementation process, the information graphic code steganography module is implemented by a reversible neural network formed by stacking reversible neural modules. The information graphic code steganography module includes forward and reverse processes. The forward process of the information graphic code steganography module is used to encode the reduced form of the spliced information graphic code and steganographically embed it into the video frame In it. The reverse process of the information graphic code steganography module is used to combine the second key map to restore the predicted reduced form of the spliced information graphic code and the corresponding video frame .

[0046] The forward process of the reversible neural network of the information graphic code steganography module is as Figure 6 shown. The reduced form of the spliced information graphic code input to the first reversible neural module of the information graphic code steganography module and the video frame are respectively transformed to the frequency domain through frequency domain transformation, and then input to the feature map of the reduced form of the spliced information graphic code of any k-th reversible neural module of the information graphic code steganography module is calculated by the fourth residual dense convolutional network and added to the feature map of the video frame as the feature map of the video frame of the subsequent k + 1-th reversible neural module that makes up the information graphic code steganography module : where is the fourth residual dense convolutional network in the k-th reversible neural module of the information graphic code steganography module. The feature map of the video frame also learns the non-scaling rotation factor respectively through the fifth residual dense convolutional network and the scalable Sigmoid activation function in the k-th reversible neural module of the information graphic code steganography module, and learns the translation factor through the sixth residual dense convolutional network. The feature map is multiplied by the non-scaling rotation factor and then added to the translation factor as the feature map of the spliced information graphic code of the subsequent k + 1-th reversible neural module that makes up the information graphic code steganography module ; where is the fifth residual dense convolutional network in the k-th reversible neural module of the information graphic code steganography module, It is the sixth residual dense convolutional network in the k-th reversible neural module of the information graphic code steganography module. The implementation of the scalable Sigmoid activation function in each reversible neural module of the information graphic code steganography module is as follows: The output of the fifth residual dense convolutional network is compressed to the range of (0, 1) by the Sigmoid activation function, and then translated to the range of (-0.5, 0.5) by subtracting 0.5, so that the final rotation factor is symmetric about 0. Then, through the unscaling coefficient The output value is scaled and then the exponential function of the natural exponential e is taken to obtain the rotation factor.

[0047] In the forward process of the information graphic code steganography module, the first reversible neural module in the information graphic code steganography module receives the downscaled form of the spliced information graphic code of the feature map and the video frame of the feature map . The feature map output by the final k-th reversible neural module in the information graphic code steganography module is used as the frequency domain form of the final second key map. The feature map output by the final k-th reversible neural module in the information graphic code steganography module is used as the frequency domain form of the traceable video frame. The frequency domain forms of the second key map and the traceable video frame are transformed to the spatio-temporal domain through the inverse frequency domain transformation.

[0048] The reverse process in the reversible neural network of the information graphic code steganography module is as Figure 7 shown. The second key map and the traceable video frame input to the information graphic code steganography module are processed through the frequency domain transformation and converted to the frequency domain. The feature map of the traceable video frame in the frequency domain received by the final k-th reversible neural module and the feature map of the second key map in the frequency domain .

[0049] Then, the feature map of the traceable video frame in the frequency domain input to any k + 1-th reversible neural module of the information graphic code steganography module respectively calculates the translation factor through the sixth residual dense convolutional network, and calculates the unscaling rotation factor through the fifth residual dense convolutional network and the scalable Sigmoid activation function; the feature map of the spliced information graphic code in the frequency domain subtracts the translation factor and then divides by the unscaling rotation factor, and the result is used as the feature map of the spliced information graphic code of the k-th reversible neural module of the information graphic code steganography module : .

[0050] The feature map subtracts the feature map The feature map obtained after transformation calculated by the fourth residual dense convolutional network serves as the k-th invertible neural module of the information graphic code steganography module : 。

[0051] To enable the information steganography model to have the above effects, the information steganography model is trained as follows: During training, the learning objectives of the information steganography model include minimizing the following loss terms: The difference between the video frame to be processed and the traceable video frame; The difference between the frequency domains of the video frame to be processed and the traceable video frame; The reduced form of the spliced information graphic code encoded by the forward process of the information graphic code steganography module and the reduced form of the predicted spliced information graphic code restored from the traceable video frame by combining the reverse process of the information graphic code steganography module with the second key map The difference between; The difference between the spliced information graphic code decoded by the reverse process of the information graphic code steganography module and the original information graphic code.

[0052] S500, Process the traceable video frames in the traceable video through the pre-trained information steganography model, and decode the confidentiality operation information from the traceable video frames.

[0053] Embodiment 2 Refer to Figure 9 As shown, the present invention provides a high-security scenario confidentiality operation video security management device, including: at least one processing unit, the processing unit is connected to a storage unit through a bus unit, the storage unit stores a computer program, and when the computer program is executed by the processing unit, the high-security scenario confidentiality operation video security management method is implemented, including: During the high-security scenario confidentiality operation process, identify the identity of the personnel performing the confidentiality operation and obtain the confidentiality operator identity information; During the high-security scenario confidentiality operation process, the video records the confidentiality operation of the confidentiality operator; Through the pre-trained information steganography model, the confidentiality operation information including the identified confidentiality operator identity information is hidden into the video frames of the video, obtain the traceable video frames containing the confidentiality operation information, and replace the original video frames with the traceable video frames to obtain the traceable video, Decode the confidentiality operation information from the traceable video frames through the pre-trained information steganography model to achieve high-security scenario confidentiality operation video security management; Among them, the information steganography model includes: an information graphic code preprocessing module based on a reversible neural network and an information graphic code steganography module; the forward process of the information graphic code preprocessing module is used to encode the spliced information graphic code, downsample the spliced information graphic code into a reduced form that supports hiding in video frames, and obtain a corresponding first key map; the reverse process of the information graphic code preprocessing module is used to upsample the reduced form of the spliced information graphic code decoded by the information graphic code steganography module in combination with the first key map to restore the decoded spliced information graphic code; the reduced form of the spliced information graphic code and the video frame to be processed are input into the information graphic code steganography module, and the information graphic code steganography module performs discrete wavelet transform on the reduced form and the video frame to the frequency domain, and then the frequency domain data is input into the reversible neural network of the information graphic code steganography module for steganography to obtain a second key map and the frequency domain data of the video frame with the information graphic code, and then the second key map and the traceable video frame are obtained through inverse frequency domain transformation respectively.

[0054] Of course, the computer program stored in the storage unit of a high-security scenario confidentiality operation video security management device provided by an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in a high-security scenario confidentiality operation video security management method provided by any embodiment of the present invention.

[0055] Embodiment 3 An embodiment of the present invention provides a computer-readable storage medium, and the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the high-security scenario confidentiality operation video security management method is implemented, including: During the high-security scenario confidentiality operation process, identify the identity of the person performing the confidentiality operation and obtain the identity information of the confidentiality operator; During the high-security scenario confidentiality operation process, video-record the confidentiality operation of the confidentiality operator; Through a pre-trained information steganography model, hide the confidentiality operation information including the identified identity information of the confidentiality operator into the video frames of the video, obtain a traceable video frame containing the confidentiality operation information, and replace the original video frame with the traceable video frame to obtain a traceable video; Decode the confidentiality operation information from the traceable video frame through a pre-trained information steganography model to implement high-security scenario confidentiality operation video security management; Among them, the information steganography model includes: an information graphic code preprocessing module based on a reversible neural network and an information graphic code steganography module; the forward process of the information graphic code preprocessing module is used to encode the spliced information graphic code, downsample the spliced information graphic code into a reduced form that supports hiding in a video frame, and obtain a corresponding first key map; the reverse process of the information graphic code preprocessing module is used to upsample the reduced form of the spliced information graphic code decoded by the information graphic code steganography module in combination with the first key map to restore the decoded spliced information graphic code; the reduced form of the spliced information graphic code and the video frame to be processed are input into the information graphic code steganography module, and the information graphic code steganography module performs discrete wavelet transform on the reduced form and the video frame to the frequency domain. Then, the frequency domain data is input into the reversible neural network of the information graphic code steganography module for steganography to obtain a second key map and the frequency domain data of the video frame with the information graphic code. Then, through inverse frequency domain transformation, the second key map and the traceable video frame are obtained respectively.

[0056] Certainly, the computer-readable storage medium provided by the embodiments of the present invention stores computer programs that are not limited to the method operations described above, and can also execute relevant operations in a high-security scenario secure operation video security management method provided by any embodiment of the present invention.

[0057] In the embodiments provided by the present invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces, indirect couplings or communication connections of structures or units, and can be in electrical, mechanical or other forms.

[0058] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0059] In addition, the functional units in various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0060] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features claimed herein.

Claims

1. A video security management method for confidential operations in high-secrecy scenarios, characterized in that, Including: In a high-security scenario, identify the identity of the security operation personnel and obtain their identity information; Use surveillance videos to record the security operations performed by the security operation personnel; Through a pre-trained information steganography model, hide the security operation information containing the identity information of the security operation personnel into the video frames of the video, obtain traceable video frames containing the security operation information, and replace the original video frames with the traceable video frames to obtain a traceable video; For the traceable video, before the traceable video management personnel process the traceable video, obtain the identity information of the video processing personnel and their video processing operations, and incrementally hide them into the video frames of the traceable video to be processed through the information steganography model; Decode the security operation information from the traceable video frames through a pre-trained information steganography model to achieve the security management of the security operation video in a high-security scenario; Among them, the information steganography model includes: an information graphic code preprocessing module and an information graphic code steganography module based on a reversible neural network; the forward process of the information graphic code preprocessing module is used to encode the spliced information graphic code, downsample the spliced information graphic code into a reduced form that supports hiding in the video frame and obtain the corresponding first key map; the reverse process of the information graphic code preprocessing module is used to upsample the reduced form of the spliced information graphic code decoded by the information graphic code steganography module in combination with the first key map to restore the decoded spliced information graphic code; input the reduced form of the spliced information graphic code and the video frame to be processed into the information graphic code steganography module, and the information graphic code steganography module performs discrete wavelet transform on the reduced form and the video frame to the frequency domain, and then the frequency domain data is input into the reversible neural network of the information graphic code steganography module for steganography to obtain the second key map and the frequency domain data of the video frame with the information graphic code, and then respectively perform inverse frequency domain transformation to obtain the second key map and the traceable video frame.

2. The high-security scenario confidentiality operation video security management method according to claim 1, wherein Extract multiple sampled video frames from the entire time span of the video recording the security operation through segmented sampling, and the information steganography model performs information steganography of the security operation information on each sampled video frame to obtain sampled traceable video frames, and replace the original corresponding sampled video frames with the sampled traceable video frames to obtain a traceable video.

3. The high-security scenario confidentiality operation video security management method according to claim 1, characterized in that, Use a QR code generator to convert the security operation information into corresponding information graphic codes, and write the at least one information graphic code written into the video frame together in a set format to obtain a spliced information graphic code.

4. The high-security scenario confidentiality operation video security management method according to claim 3, characterized in that When the number of information graphic codes written into the video frame is less than the number of information graphic codes set by the format, fill it with the replication result of the information graphic code written into the video frame to meet the format setting.

5. The high-secrecy scenario confidentiality operation video security management method according to claim 1, characterized in that The forward process of the reversible neural network of the information graphic code preprocessing module is as follows: The feature map of the spliced information graphic code input to any k-th reversible neural module is split into sub-feature maps along the channel dimension and the sub-feature maps , and the sub-feature maps are calculated by the first residual dense convolutional network and then added to the sub-feature maps . The result of the addition is used as the input feature map of the subsequent (k + 1)-th reversible neural module that composes the information graphic code preprocessing module 's sub-feature map ; The sub-feature maps also learn the scaling and rotation factors through the second residual dense convolutional network and the scalable Sigmoid activation function respectively in the k-th reversible neural module, and learn the translation factor through the third residual dense convolutional network. The sub-feature maps are multiplied by the scaling and rotation factors and then added to the translation factor. The result of the addition is used as the feature map of the subsequent (k + 1)-th reversible neural module that composes the information graphic code preprocessing module 's sub-feature map .

6. The high-security scenario confidentiality operation video security management method according to claim 1, characterized in that The reverse process of the reversible neural network of the information graphic code preprocessing module is as follows: The feature map of the spliced information graphic code input to any (k + 1)-th reversible neural module of the information graphic code preprocessing module is split into sub-feature maps along the channel dimension and the sub-feature maps , and the sub-feature maps respectively learn the scaling and rotation factors through the second residual dense convolutional network and the scalable Sigmoid activation function, and learn the translation factor through the third residual dense convolutional network. After the sub-feature maps are subtracted by the translation factor and then divided by the scaling and rotation factor, they are used as the sub-feature maps of the k-th reversible neural module of the information graphic code preprocessing module ; the sub-feature maps subtract the transformation result calculated by the first residual dense convolutional network from the sub-feature maps to be used as the sub-feature maps of the k-th reversible neural module of the information graphic code preprocessing module .

7. The high-security scenario confidentiality operation video security management method according to claim 1, characterized in that The forward process of the reversible neural network of the information graphic code steganography module is as follows: The reduced form of the spliced information graphic code input to the first reversible neural module of the information graphic code steganography module and the video frame are respectively transformed to the frequency domain through frequency domain transformation, and then input to the reduced form of the spliced information graphic code of any k-th reversible neural module of the information graphic code steganography module feature map Calculate the transformed one and the video frame feature map through the fourth residual dense convolutional network, and add them together. The addition result is used as the feature map of the video frame of the subsequent k + 1-th reversible neural module that constitutes the information graphic code steganography module ; Feature map of video frame In the k-th reversible neural module of the information graphic code steganography module, non-scaled rotation factors are also learned through the fifth residual dense convolutional network and the scalable Sigmoid activation function, and translation factors are learned through the sixth residual dense convolutional network. The feature map of the spliced information graphic code After being multiplied by the non-scaled rotation factor and then added to the translation factor, it serves as the feature map of the spliced information graphic code for the subsequent (k + 1)-th reversible neural module that constitutes the information graphic code steganography module 。 8. The high-security scenario confidentiality operation video security management method according to claim 1, characterized in that, The reverse process in the reversible neural network of the information graphic code steganography module is as follows: The second key map and the traceable video frame input into the information graphic code steganography module are transformed into the frequency domain through frequency domain transformation processing, and the feature map of the traceable video frame in the receiving frequency domain of the final Kth reversible neural module and the feature map of the second key map in the frequency domain ; Then, it is input into the feature map of the traceable video frame in the frequency domain of any (k + 1)-th invertible neural module of the information graphic code steganography module The translation factor is calculated through the sixth residual dense convolutional network, and the non-scaled rotation factor is calculated through the fifth residual dense convolutional network and the scalable Sigmoid activation function; the feature map of the spliced information graphic code in the frequency domain After subtracting the translation factor, it is divided by the non-scaled rotation factor, and the result is used as the feature map of the spliced information graphic code of the k-th invertible neural module of the information graphic code steganography module : Feature map Subtract the feature map The feature map of the information graphic code that is transformed by the fourth residual dense convolutional network and serves as the splicing of the k-th invertible neural module of the information graphic code steganography module 。 9. The high-security scenario confidentiality operation video security management method according to claim 1, characterized in that The implementation of the scalable Sigmoid activation function is as follows: The Sigmoid activation function compresses the output of the residual dense convolutional network to the range of (0, 1), and then translates it to the range of (-0.5, 0.5) by subtracting 0.5, so that the final scaled or unscaled rotation factor is symmetric about 0. Then, the output value is scaled by the scaled or unscaled coefficient and the exponential function of the natural exponent e is taken to obtain the scaled or unscaled rotation factor.

10. The high-security scenario confidentiality operation video security management method according to claim 1, characterized in that, During training, the learning objective of the information steganography model includes minimizing the following loss terms: The difference between the video frame to be processed and the traceable video frame; The difference between the frequency domains of the video frame to be processed and the traceable video frame; The difference between the reduced form of the spliced information graphic code encoded by the forward process of the information graphic code steganography module and the reduced form of the predicted spliced information graphic code restored from the traceable video frame by combining the reverse process of the information graphic code steganography module with the second key graph; minimizing the difference between the spliced information graphic code decoded by the reverse process of the information graphic code steganography module and the original information graphic code.

Images