A multi-domain cascading steganography tracing method and system based on AB flow adaptive streaming

Abstract

The application discloses a kind of multi-domain cascading steganography tracing methods and systems based on AB flow adaptive streaming, belong to digital watermarking and information hiding technical field, it includes offline pretreatment and double-flow generation, obtains source video and extracts Y brightness channel, generates tracing information and carries out three-level error correction coding, while embedding six orthogonal steganography domains, generates A flow and B flow fragment and pushes to distribution network;Online dynamic distribution, allocate unique tracing ID, according to bit value selects corresponding fragment, dynamically generates play list;Infringement tracing restoration, decode pirate video and extract Y brightness channel, execute three-stage cascading extraction, restore tracing ID by section-level majority vote and multiple sets of redundant vote.The application realizes large-scale user tracing with zero storage overhead, has strong attack resistance, resolution-independent, visually lossless, can be directly connected to CDN system, and is suitable for streaming copyright protection scenarios.

Description

Technical Field

[0001] This invention belongs to the field of digital watermarking and information hiding technology, specifically a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media. Background Technology

[0002] Currently, the streaming media industry generally uses the HLS (HTTP Live Streaming) protocol for content distribution. The core relies on TS segmentation and dynamic m3u8 playlists to achieve adaptive transmission. Existing copyright traceability technologies are mainly divided into three categories: First, single-domain digital watermarking steganography, including single-domain solutions such as DCT-QIM, DWT spread spectrum, and SEI metadata embedding; second, user-specific personalized video generation solutions, which achieve traceability by customizing exclusive video files for individual users; and third, simple bitstream information marking solutions, which only add basic identifiers to the video bitstream without complete traceability logic.

[0003] Existing technologies have five major shortcomings in large-scale distribution scenarios for commercial HLS streaming media: excessively high storage costs (generating personalized videos for each user leads to storage overhead increasing linearly with the number of users); weak attack resistance (single-domain steganography schemes are prone to losing traceability information when facing attacks such as transcoding, compression, screen recording, and resolution scaling); a conflict between real-time performance and robustness (real-time watermark embedding schemes result in excessively high latency for the first frame of playback); poor resolution adaptability (traceability information is bound to a fixed resolution, and information misalignment is easily caused after multi-resolution adaptive playback); and visual quality and signal interference issues (multi-domain steganography schemes lack spatial partitioning design, and signals embedded in different domains interfere with each other, easily leading to video distortion). Summary of the Invention

[0004] The purpose of this invention is to provide a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a multi-domain cascading steganography tracing method based on AB stream adaptive streaming media, comprising the following steps:

[0006] Offline preprocessing and dual-stream generation: The source video is acquired and the Y-luminance channel is extracted; source information is generated and three-level error correction coding is performed, mapping the encoded data to six orthogonal steganography domains; in the Y-luminance channel, through spatial partitioning and pseudo-random block hashing, the mapped data is simultaneously embedded into the supplementary enhancement information domain, the discrete cosine transform-quantization index modulation domain, the discrete cosine transform-spreading domain, the frame-level spreading domain, the resolution-invariant domain, and the discrete wavelet transform / discrete Fourier transform / wavelet packet transform frequency domain, generating A-stream and B-stream media slices carrying differentiated identifiers, and pushing them to the distribution network;

[0007] Online dynamic distribution: Receive playback requests from terminals, assign a unique source ID; convert the source ID into a binary bit sequence, select stream A segment when the bit value is 0, select stream B segment when the bit value is 1, dynamically generate a playlist file and return it to the terminal;

[0008] Infringement tracing and restoration: Obtain suspicious pirated videos, decode and extract Y luminance channel data; perform three-level cascaded extraction in priority order, including extracting identifiers from the supplemental enhancement information domain, decoding bitstreams from the discrete cosine transform-quantization index modulation domain, and robust extraction from the resolution invariant domain and frame-level spread spectrum domain; perform segment-level majority voting and multiple sets of redundant voting on the extraction results to restore the tracing ID.

[0009] As a further preferred embodiment of this technical solution, in the offline preprocessing and dual-stream generation steps, the three-level error correction coding specifically employs a concatenated mode of RS(128,64) external code paired with dual LDPC(1024,512) internal codes; after encrypting and scrambling the traceability information, it is divided into two segments by RS coding, corresponding to the embedded discrete cosine transform-quantization index modulation domain and discrete cosine transform-spreading domain, respectively.

[0010] As a further preferred embodiment of this technical solution, in the offline preprocessing and dual-stream generation steps, the spatial partitioning and pseudo-random block hashing arrangement are specifically as follows: the Discrete Cosine Transform-Quantization Index Modulation Domain occupies 512 4×4 blocks, the Discrete Cosine Transform-Spreading Domain occupies 1024 4×4 blocks, and the Frame-Level Spreading Domain occupies 800 8×8 blocks, with each domain arranged using different hash seeds; the resolution-invariant domain uses a square cropping area at a 47.5% ratio of the frame center, and is divided into 32×32 normalized grids for embedding.

[0011] As a further preferred embodiment of this technical solution, the offline preprocessing and dual-stream generation steps also include adaptive perceptual masking: dynamically adjusting the pixel modification amount based on the texture variance of a local 8×8 block, skipping embedding when the variance is less than 30, the modification amount is ±1 when the variance is between 30 and 100, the modification amount is ±2 when the variance is between 100 and 500, and the modification amount is ±4 when the variance is not less than 500, with global constraints that the peak signal-to-noise ratio of the video is not less than 42dB and the maximum modification amount of a single pixel does not exceed 4.

[0012] As a further preferred embodiment of this technical solution, in the online dynamic distribution step: the traceability ID is a 32-bit binary sequence, with stream A segment selected when the bit value is 0 and stream B segment selected when the bit value is 1; the traceability ID is cyclically encoded once every 32 segments; the playlist file is in m3u8 format, the media segment is in TS format, and the segment duration is 5 seconds.

[0013] As a further preferred embodiment of this technical solution, the three-level cascaded extraction in the infringement tracing and restoration step specifically comprises: L0 level extraction: directly reading the AB identifier in the supplementary enhancement information domain; L1 level extraction: decoding the bitstream of the discrete cosine transform-quantization index modulation domain, and performing LDPC and RS decoding error correction; L2 level extraction: when L0 and L1 level extraction fail, performing frame-level spread spectrum domain multi-resolution extraction and resolution-invariant domain fault-tolerant extraction; wherein, the frame-level spread spectrum domain multi-resolution extraction has 8 built-in resolution candidates and automatically traverses and matches pseudo-random sequences, and the resolution-invariant domain fault-tolerant extraction searches for a 5×5 candidate grid with ±10% spatial offset based on the standard center clipping position, and selects the result with the highest correlation.

[0014] As a further preferred embodiment of this technical solution, a multi-domain cascading steganography tracing system based on AB-stream adaptive streaming media includes: an offline preprocessing pipeline deployed on an origin server, used to execute the offline preprocessing and dual-stream generation steps in any one of claims 1 to 6, generating A-stream and B-stream TS fragments; an online distribution engine deployed on the edge node of a content distribution network, used to execute the online dynamic distribution step in any one of claims 1 to 6, dynamically generating a playlist based on the tracing ID; and an infringement tracing engine deployed on a copyright protection platform, used to execute the infringement tracing restoration step in any one of claims 1 to 6, restoring the tracing ID from pirated videos.

[0015] As a further preferred embodiment of this technical solution, the offline preprocessing pipeline includes: a standardized transcoding module for source video transcoding and Y-luminance channel extraction; a six-domain cascaded steganography module for encrypting source information, error correction coding, and embedding six orthogonal steganography domains; an A-stream and B-stream TS segment generation module for generating dual-channel TS segments; and the online distribution engine includes: a user authentication module for verifying user identity and assigning a source ID; a bit-level mapping module for implementing A / B selection mapping between source IDs and TS segments; and a dynamic playlist generation module for generating playlists conforming to the HTTP LiveStreaming protocol standard.

[0016] As a further preferred embodiment of this technical solution, the infringement tracing engine includes: a video decoding preprocessing module for decoding pirated videos and normalizing the Y luminance channel; a three-level cascaded extraction module for extracting AB identifiers according to L0, L1, and L2 priority order; a redundant voting module for performing segment-level majority voting on the AB identifiers of frames within the same segment and performing multiple sets of redundant voting on multiple segment encoding results; and a tracing ID restoration module for converting the binary sequence of the AB identifiers after voting into a tracing ID.

[0017] As a further preferred embodiment of this technical solution, a computer-readable storage medium is provided, wherein a computer program is stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 6.

[0018] Compared with the prior art, the beneficial effects of the present invention are:

[0019] 1. In this invention, an AB stream bit-level zero-storage traceability mechanism is adopted, which converts the traceability ID into a binary bit sequence. Bit 0 corresponds to the A stream fragment and bit 1 corresponds to the B stream fragment. The CDN only needs to host two static TS fragments, A and B, without generating a dedicated video file for each user. This completely abandons the traditional user-customized video model, with zero storage overhead. It can support up to 4,294,967,296 unique users. Even if 50% of the fragments are damaged, the complete traceability ID can still be restored through redundant voting, supporting accurate traceability for a large number of users.

[0020] 2. In this invention, a six-domain orthogonal concatenated steganography and three-level error correction coding scheme is adopted. Six mutually orthogonal steganography domains (SEI domain, DCT-QIM domain, DCT-SS domain, FS frame-level spread spectrum domain, RIS resolution invariant domain, and DWT / DFT / WPT frequency domain) are embedded in the Y luminance channel of the video. Each domain adopts strict spatial partitioning and pseudo-random block hashing arrangement. Combined with the three-level concatenated error correction of RS(128,64) external code and dual LDPC(1024,512) internal code, it can resist 16 mainstream streaming media attacks such as transcoding, extreme compression, screen recording, and resolution scaling. The overall source tracing success rate is not less than 75%, which far exceeds the traditional single-domain steganography scheme.

[0021] 3. This invention employs RIS resolution-invariant steganography, using a 47.5% square cropping area at the frame center combined with a 32×32 normalized grid design to completely eliminate dependence on absolute resolution. It eliminates the need to know the specific resolution of the pirated video; regardless of how the video has been scaled or cropped, accurate extraction can be achieved, perfectly adapting to HLS multi-resolution adaptive playback scenarios.

[0022] 4. This invention employs a three-level cascaded extraction strategy, with extraction prioritized according to L0-L1-L2 levels: L0 level is fast extraction of the SEI domain, suitable for lossless transmission scenarios; L1 level is DCT-QIM domain decoding extraction, suitable for medium compression scenarios; L2 level is FS+RIS joint robust extraction, suitable for heavy attack scenarios such as screen recording, extreme compression, and scaling. Lossless scenario extraction takes no more than 1 millisecond per frame, significantly improving batch detection efficiency.

[0023] 5. In this invention, an adaptive perceptual masking technique is employed, dynamically adjusting pixel modification amounts based on the local 8×8 block texture variance: solid color areas skip embedding, flat areas have a modification amount of ±1, medium texture areas have a modification amount of ±2, and rich texture areas have a modification amount of ±4. Global constraints ensure the video peak signal-to-noise ratio is no less than 42dB, and the maximum modification amount per pixel does not exceed 4, ensuring that the embedded video exhibits no perceptible distortion, does not affect the user's viewing experience, and meets commercial streaming media quality standards.

[0024] 6. This invention is based on Rust+FFmpeg, is compatible with H.264 / H.265 encoding and mainstream CDN platforms, and adopts a three-stage overall system architecture (offline preprocessing pipeline - online distribution engine - infringement tracing engine). It is fully compatible with existing CDN and streaming media platform architectures, does not require reconstruction of the existing distribution system, has low deployment costs, strong engineering feasibility, and is suitable for various streaming media copyright protection scenarios such as long videos, live broadcasts, and educational videos. Attached Figure Description

[0025] Figure 1 This is an overall architecture diagram of a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media according to the present invention;

[0026] Figure 2 The flowchart below shows the offline preprocessing and dual-stream generation process of a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media according to the present invention.

[0027] Figure 3 This is a flowchart illustrating the online distribution and dynamic playlist generation process of a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media according to the present invention.

[0028] Figure 4 This is a flowchart illustrating the infringement tracing and source ID restoration process of a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media according to the present invention.

[0029] Figure 5 This is a schematic diagram of the six-domain orthogonal steganography space partitioning of a multi-domain cascaded steganography tracing method and system based on AB stream adaptive streaming media according to the present invention.

[0030] Figure 6 This is a schematic diagram of the RIS resolution invariant domain normalized grid of a multi-domain cascaded steganography tracing method and system based on AB stream adaptive streaming media according to the present invention.

[0031] Figure 7 This is a schematic diagram illustrating the priority of the three-level cascading extraction strategy of a multi-domain cascading steganography tracing method and system based on AB stream adaptive streaming media according to the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example

[0034] Please see Figure 1 - Figure 7 As shown, the objectives, technical solutions, and advantages of the present invention will be made clearer. The specific embodiments of the present invention will be described in detail below.

[0035] I. Implementation Environment

[0036] Hardware environment:

[0037] Origin server: Intel i7 or higher processor, 16GB or more of memory

[0038] CDN edge nodes: Standard configuration nodes

[0039] Copyright detection server: Intel Xeon E5 or higher processor, 32GB or more of memory

[0040] Software environment:

[0041] Operating systems: Windows Server, CentOS, Ubuntu

[0042] Programming language: Rust 2021 Edition

[0043] Encoding / decoding library: FFmpeg

[0044] Dependencies: aes 0.8.3, crc32fast 1.32, rayon 1.7.0, tokio 1.32.0, ffmpeg-sys-next 8.0

[0045] Basic parameters:

[0046] Video base format: 1080p / 30fps

[0047] TS segment duration: 5 seconds

[0048] CRF transcoding standard: 23

[0049] RIS embedding strength: 3.0

[0050] DCT-QIM quantization step size: 24.0

[0051] II. Implementation Steps

[0052] Step 1: Offline Preprocessing

[0053] Source video standardization transcoding: Transcode the source MP4 video into a YUV420P format frame sequence and extract the Y luminance channel data.

[0054] Generate traceability information: Generate traceability information in 64-byte format, which is: 4B magic number "ABST" + 16B video UUID + 1B AB identifier + 4B keyframe index + 2B fragment ID + 4B CRC32 + 33B reserved bits.

[0055] Encryption and Scrambling: The traceability information is encrypted with AES-256 and scrambled using pseudo-random methods.

[0056] Three-level error correction coding: Employs a concatenated pattern of RS(128,64) outer code combined with dual LDPC(1024,512) inner code, where RS coding is based on... For a finite field, the irreducible polynomial is: The primitive element is α=2, which can correct 32-byte errors; the LDPC encoding code rate is 1 / 2, and the parity check matrix... The row weight is 6, the maximum number of BP iterations is 50, and the encoded data is divided into two halves, which correspond to the embedded DCT-QIM field and DCT-SS field respectively.

[0057] Six-domain orthogonal embedding: In the Y-luminance channel, according to the spatial partitioning and block hashing strategy, the encoded bitstream is simultaneously embedded into six orthogonal steganography domains: SEI, DCT-QIM, DCT-SS, FS, RIS, and DWT / DFT / WPT. The spatial partitioning strategy is as follows: the 4×4 DCT-QIM domain occupies 512 blocks, the DCT-SS domain occupies 1024 blocks, and the 8×8 FS domain occupies 800 blocks. The block hashing seeds are 0xAB5F44AD2026 and 0xAB5FB10C2026, respectively. The RIS domain is a square cropped area of ​​47.5%×min(w,h) at the frame center, divided into a 32×32 normalized grid.

[0058] Adaptive perceptual masking: Pixel modification is dynamically adjusted based on the local 8×8 block texture variance. Embedding is skipped when the variance is less than 30, the modification is ±1 for 30-100, ±2 for 100-500, and ±4 for greater than 500. Global constraints include a PSNR of at least 42dB and a maximum single-pixel modification of no more than 4.

[0059] Generate A / B stream TS fragments: Embed AB identifiers 0x01 and 0x02 respectively, encode to generate A stream TS fragments and B stream TS fragments, and push them to CDN node hosting.

[0060] Step Two: Online Distribution

[0061] User authentication request: Receive user playback requests and verify user identity.

[0062] Assign a traceability ID: Assign a unique 32-bit traceability ID (the traceability ID mentioned in this invention refers to an identifier used to uniquely identify a user and track the source of piracy).

[0063] Bit-level mapping: The traceability ID is converted into a binary bit sequence. When the bit value is 0, the A stream fragment is selected, and when the bit value is 1, the B stream fragment is selected. The traceability ID is encoded once every 32 fragments.

[0064] Dynamically generate playlists: Dynamically generate m3u8 playlists containing CDN access addresses for TS segments and return them to the user.

[0065] Step 3: Tracing the source of infringement

[0066] Pirated Video Decoding: Decodes suspicious pirated videos into YUV420P format.

[0067] Y Channel Extraction and Normalization: Extract the Y luminance channel and perform luminance normalization preprocessing. The normalization formula is as follows:

[0068]

[0069] in The average global brightness value for the Y channel. This represents the standard deviation of brightness.

[0070] Three-level cascade extraction:

[0071] Level L0: Directly reads the AB identifier in the SEI field, completed in milliseconds, suitable for lossless transmission scenarios.

[0072] Level L1: Decodes the bitstream of the DCT-QIM domain, performs error correction via LDPC and RS decoding, and is suitable for medium compression scenarios.

[0073] Level L2: When Level L0 and Level L1 extraction fail, multi-resolution extraction in the FS domain and fault-tolerant extraction in the RIS domain are performed. The FS domain multi-resolution extraction has 8 built-in resolution candidates (including 1920×1080, 1728×972, 1600×900, 1440×810, 1280×720, 1152×648, 960×540, and 854×480), and automatically traverses and matches the PN sequence. The RIS domain fault-tolerant extraction searches for 5×5 candidate grids with ±10% spatial offset based on the standard center clipping position, and selects the result with the highest correlation. The RIS domain PN sequence seed is 0xAB5F72152026, and the embedding strength is 3.0.

[0074] Voting Restoration: Segment-level majority voting is performed on the AB identifiers of frames within the same TS segment, and multiple sets of redundant voting are performed on the encoding results of 32 TS segments. Even if 50% of the fragments are damaged, the complete traceability ID can still be restored through redundant voting. The 32-bit AB identifier binary sequence after voting is converted into a decimal traceability ID.

[0075] III. Implementation Results Verification

[0076] After testing in 16 typical streaming media attack scenarios (including transcoding, extreme compression, screen recording, resolution scaling, cropping, etc.), the time for batch tracing of 1 hour of video does not exceed 5 minutes, the overall tracing success rate is not less than 75%, the PSNR of the embedded video is not less than 42dB, it supports adaptive extraction of 8 mainstream resolutions, and can trace the source normally in ±10% cropping scenarios, fully meeting the needs of commercial copyright protection.

[0077] IV. Terminology Definitions Source ID: The source ID mentioned in this invention refers to an identifier used to uniquely identify a user and track the source of piracy, and its form is a 32-bit binary sequence.

[0079] AB stream: refers to two static TS fragments, A stream and B stream, which only embed differentiated identifiers and are shared by all users.

[0080] The six orthogonal steganography domains include the SEI domain, DCT-QIM domain, DCT-SS domain, FS domain, RIS domain, and DWT / DFT / WPT domain. Each domain adopts strict spatial partitioning and pseudo-random block hashing to prevent signal interference.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-domain cascading steganography tracing method based on AB stream adaptive streaming media, characterized in that, Includes the following steps: Offline preprocessing and dual-stream generation: The source video is acquired and the Y-luminance channel is extracted; source information is generated and three-level error correction coding is performed, mapping the encoded data to six orthogonal steganography domains; in the Y-luminance channel, through spatial partitioning and pseudo-random block hashing, the mapped data is simultaneously embedded into the supplementary enhancement information domain, the discrete cosine transform-quantization index modulation domain, the discrete cosine transform-spreading domain, the frame-level spreading domain, the resolution-invariant domain, and the discrete wavelet transform / discrete Fourier transform / wavelet packet transform frequency domain, generating A-stream and B-stream media slices carrying differentiated identifiers, and pushing them to the distribution network; Online dynamic distribution: Receive playback requests from terminals, assign a unique source ID; convert the source ID into a binary bit sequence, select stream A segment when the bit value is 0, select stream B segment when the bit value is 1, dynamically generate a playlist file and return it to the terminal; Infringement tracing and restoration: Obtain suspicious pirated videos, decode and extract Y luminance channel data; The three-level cascaded extraction is performed in order of priority, including extracting the identifier from the supplementary enhancement information domain, decoding the bitstream from the discrete cosine transform-quantization index modulation domain, and performing robust extraction from the resolution invariant domain and frame-level spread spectrum domain; the extraction results are then subjected to segment-level majority voting and multiple sets of redundant voting to restore the source ID.

2. The method according to claim 1, characterized in that, In the offline preprocessing and dual-stream generation steps, the three-level error correction coding specifically employs a concatenated mode of RS(128,64) external code paired with dual LDPC(1024,512) internal codes. After encrypting and scrambling the traceability information, it is divided into two segments by RS coding, corresponding to the embedded discrete cosine transform-quantization index modulation domain and discrete cosine transform-spreading domain, respectively.

3. The method according to claim 1, characterized in that, In the offline preprocessing and dual-stream generation steps, the spatial partitioning and pseudo-random block hashing arrangement are as follows: the Discrete Cosine Transform-Quantization Index Modulation Domain occupies 512 4×4 blocks, the Discrete Cosine Transform-Spreading Domain occupies 1024 4×4 blocks, and the Frame-Level Spreading Domain occupies 800 8×8 blocks. Each domain is arranged using different hash seeds. The resolution-invariant domain uses a square cropping area at 47.5% of the frame center and is divided into 32×32 normalized grids for embedding.

4. The method according to claim 1, characterized in that, The offline preprocessing and dual-stream generation steps also include adaptive perceptual masking: dynamically adjusting the pixel modification amount based on the texture variance of a local 8×8 block. When the variance is less than 30, the embedding is skipped; when the variance is between 30 and 100, the modification amount is ±1; when the variance is between 100 and 500, the modification amount is ±2; when the variance is not less than 500, the modification amount is ±4. The global constraint is that the peak signal-to-noise ratio of the video is not less than 42dB and the maximum modification amount of a single pixel does not exceed 4.

5. The method according to claim 1, characterized in that, In the online dynamic distribution step: the traceability ID is a 32-bit binary sequence, with stream A segment selected when the bit value is 0 and stream B segment selected when the bit value is 1; the traceability ID is cyclically encoded once every 32 segments; the playlist file is in m3u8 format, the media segment is in TS format, and the segment duration is 5 seconds.

6. The method according to claim 1, characterized in that, In the infringement tracing and restoration steps, the three-level cascaded extraction specifically includes: L0 level extraction: directly reading the AB identifier in the supplementary enhancement information domain; L1 level extraction: decoding the bitstream of the discrete cosine transform-quantization index modulation domain, and performing LDPC and RS decoding error correction; L2 level extraction: when L0 and L1 level extraction fail, performing frame-level spread spectrum domain multi-resolution extraction and resolution-invariant domain fault-tolerant extraction; wherein, the frame-level spread spectrum domain multi-resolution extraction has 8 built-in resolution candidates and automatically traverses and matches pseudo-random sequences, and the resolution-invariant domain fault-tolerant extraction searches for a 5×5 candidate grid with ±10% spatial offset based on the standard center clipping position, and selects the result with the highest correlation.

7. A multi-domain cascading steganography tracing system based on AB-stream adaptive streaming media, characterized in that, include: An offline preprocessing pipeline, deployed on an origin server, is used to execute the offline preprocessing and dual-stream generation steps in the method of any one of claims 1 to 6, generating A-stream and B-stream TS segments; an online distribution engine, deployed on an edge node of a content delivery network, is used to execute the online dynamic distribution steps in the method of any one of claims 1 to 6, dynamically generating a playlist based on the source ID; An infringement tracing engine, deployed on a copyright protection platform, is used to execute the infringement tracing and restoration steps in the method described in any one of claims 1 to 6, and to restore the tracing ID from pirated videos.

8. The system according to claim 7, characterized in that, The offline preprocessing pipeline includes: a standardized transcoding module for source video transcoding and Y-luminance channel extraction; a six-domain cascaded steganography module for encrypting traceability information, error correction coding, and embedding six orthogonal steganography domains; and an A-stream and B-stream TS segment generation module for generating dual-channel TS segments. The online distribution engine includes: a user authentication module for verifying user identity and assigning traceability IDs; a bit-level mapping module for implementing A / B selection mapping between traceability IDs and TS segments; and a dynamic playlist generation module for generating playlists conforming to the HTTP LiveStreaming protocol standard.

9. The system according to claim 7, characterized in that, The infringement tracing engine includes: a video decoding preprocessing module for decoding pirated videos and normalizing the Y luminance channel; a three-level cascaded extraction module for extracting AB identifiers according to L0, L1, and L2 priority order; a redundant voting module for performing segment-level majority voting on AB identifiers of frames within the same segment and performing multiple sets of redundant voting on multiple segment encoding results; and a tracing ID restoration module for converting the binary sequence of AB identifiers after voting into a tracing ID.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 6.

Images