Mobile terminal sensitive information anti-interception method and system based on interface rendering interception

Abstract

The invention discloses an interface rendering interception-based mobile terminal sensitive information anti-interception method and system, and the method comprises the steps: implanting an interception hook in a mobile terminal graphic rendering pipeline, capturing a rendering instruction stream, and recognizing a sensitive information region; in combination with a rendering target context sensing technology, when risk behaviors such as screen capture, screen recording or screen projection are detected, differential rendering processing is performed on a sensitive information area, for example, OCR recognition is injected, false data of semantic simulation is generated for display, or sensitive content is physically eliminated by using a hardware layer isolation technology. According to the scheme, the authenticity of local screen display and the security balance of information in a non-local scene are realized, and the risks of automatic privacy stealing and manual screenshot leakage aiming at the application interface of the mobile terminal are effectively resisted.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of mobile terminal information security and computer graphics, and in particular to a method and system for preventing the interception of sensitive information on mobile terminals based on interface rendering interception. Background Technology

[0002] With the widespread adoption of mobile internet, smart devices carry massive amounts of high-value personal data. Preventing the unauthorized interception of sensitive information (screenshots, screen recordings, screen mirroring) is a crucial issue in information security. Currently, existing mobile protection solutions mainly include global interception based on system tags and local masking based on the application layer.

[0003] Global interception solutions, such as the FLAG_SECURE flag in Android systems, effectively prevent screen recording and screenshot attacks by disabling screenshot behavior at the window manager level. However, this solution has a "one-size-fits-all" problem: once enabled, the entire screen goes black, preventing users from sharing non-sensitive areas, resulting in a poor user experience in scenarios such as remote collaboration, online education, or bug reporting.

[0004] Local masking schemes: such as overlaying color blocks based on application layer code. While this scheme preserves the visibility of non-sensitive areas, its defense methods are limited.

[0005] Analysis of existing technologies reveals that while CN110334522A proposes a watermark protection scheme, its defense against OCR attacks is limited; CN111177770A uses layer overlay for privacy protection, but it has shortcomings in underlying protection. This invention improves protection capabilities while maintaining user experience by combining rendering pipeline interception with anti-perturbation measures. Furthermore, US Patent US2020 / 0210606A1 proposes a screen capture protection scheme, but its consistency handling of multiple outputs (main screen display and capture path) is insufficient.

[0006] Insufficient underlying protection: Malware may directly read video memory by mounting to the underlying framebuffer, or use virtual display technology to intercept data.

[0007] The difference between this invention and the aforementioned prior art is that it achieves end-to-end interception at the rendering pipeline level, introduces adversarial perturbations that are natural to the human eye and garbled by the machine, and constructs a "dual-track rendering synchronization" mechanism between the local and intercepting ends. Summary of the Invention

[0008] To address the aforementioned technical issues, this invention implements a protection strategy of "local realistic display and non-local differentiated processing" by embedding an interception mechanism deep within the graphics rendering pipeline and combining it with panoramic context awareness technology.

[0009] The technical solution of the present invention is as follows: A method for preventing the interception of sensitive information on mobile devices based on interface rendering interception includes the following steps:

[0010] Step 1): Implant a rendering interception hook in the graphics rendering pipeline of the mobile terminal;

[0011] The locations where rendering interception hooks can be implanted include at least one of the following: the layer composition loop entry point of the graphics compositing component SurfaceFlinger; the drawing command call interface of the graphics drawing library Skia; and the command buffer submission interface of the graphics driver layer Vulkan or OpenGLES.

[0012] Step 2): Capture the rendering command stream or layer composition request of the current interface;

[0013] Step 3): Analyze the rendering instruction stream or layer composition request to identify the sensitive information areas contained therein;

[0014] Step 4): Real-time detection of the current rendering target context to determine whether there is any capture behavior that is not displayed on the local physical screen;

[0015] Step 5): If the interception behavior is determined to exist, the sensitive information area is subjected to differentiated rendering processing according to the preset security policy to generate processed rendering data that is different from the original rendering data; if the interception behavior is determined not to exist, the normal original layer composition and display process is maintained.

[0016] In step 5), the differentiated rendering process includes one or any combination of the following processing methods:

[0017] Step 5.1) Adversarial perturbation processing: Calculate the text texture features of the sensitive information region; generate adversarial sample noise based on the neural network, and superimpose the adversarial sample noise onto the texture of the sensitive information region; the superimposed sensitive information region maintains a natural display effect in human vision, the visual quality loss is within a preset threshold, and it can interfere with the optical character recognition (OCR) algorithm's feature extraction of the sensitive information region;

[0018] Step 5.2) Semantic-level simulation desensitization processing: Analyze the semantic category of the original data in the sensitive information area. The semantic category includes at least one of the following: amount, name, ID number, bank card number, mobile phone number, and address; call the simulation data generation engine to generate fake simulation data that is consistent with the semantic category and format of the original data; use the fake simulation data to replace the original rendering parameters in the rendering instruction stream, and construct and display the desensitized interface of semantic simulation.

[0019] Step 5.3) Hardware-level layer separation processing: Utilize the hardware composition capability of the display controller to mark sensitive information areas as protected hardware layers; configure the filtering rules of the hardware compositor so that the protected hardware layers are only output to the local physical screen, and are automatically removed or filled with preset obfuscated content when written to the write-back buffer or frame buffer.

[0020] Step 6): Output the processed rendering data to the rendering buffer corresponding to the interception behavior, while continuing to output the original rendering data to the local physical screen.

[0021] The security level of the sensitive information dynamically selects the type of differentiated rendering processing, wherein the higher the security level, the stronger the corresponding rendering protection.

[0022] An anti-interception system includes: a rendering pipeline interception module for embedding an interceptor in the graphics rendering pipeline to capture rendering command streams or layer compositing requests; a sensitivity identification engine for analyzing the rendering command streams or layer compositing requests to identify sensitive information regions contained therein; a context detection unit for monitoring the state of the rendering target in real time to determine whether there is interception behavior that is not displayed on the local physical screen; and a differential processing engine for performing differential rendering processing on the sensitive information regions according to a preset security policy when interception behavior is determined to exist, including anti-disturbance injection processing, simulation data replacement processing, or hardware layer isolation.

[0023] The differentiated processing engine includes: an adversarial noise generation submodule, used to generate defensive noise against a specific OCR engine through a generative adversarial network model; and a simulation data generation submodule, used to generate simulated data that meets the requirements of the original data format and perform instruction-level replacement.

[0024] An electronic device includes a memory and a processor, wherein the processor implements the method when executing a computer program stored in the memory.

[0025] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method.

[0026] The beneficial effects of this invention are as follows:

[0027] 1. Achieves a perfect balance between security and user experience: Through a unique "dual-track rendering" mechanism, it ensures users have a lossless visual experience when viewing sensitive information locally (path A), and only performs differentiated processing on non-local output when interception risks occur (path B). This completely solves the user experience degradation issues caused by "full-screen black screen" or "forced censorship" in existing technologies.

[0028] 2. Effectively resists intelligent attacks: Adversarial sample technology is introduced, and micro-texture noise is specifically designed for OCR and machine vision algorithms. Compared with traditional physical occlusion, this solution significantly reduces machine recognition accuracy (missed recognition rate ≠ 95%) while maintaining human readability, filling the technological gap in resisting AI reconstruction attacks.

[0029] 3. Extremely high deceptiveness and trapping ability: Through semantic simulation technology that uses "what you see is not what you get," it not only hides the real data but also outputs formatted fake data (such as fake bank card numbers that conform to verification rules). This not only prevents information leakage but also increases the cost for attackers to clean up data, demonstrating proactive defense characteristics.

[0030] 4. End-to-End Hardware Protection Capabilities: The solution covers the entire chain, from application-layer view analysis and system-layer rendering pipeline interception to hardware-layer (HWC) isolation. In particular, the mechanism of directly removing sensitive data using the hardware compositor physically prevents advanced attacks that attempt to access video memory or framebuffers. Attached Figure Description

[0031] Figure 1 This is the main flowchart of the method provided in the embodiments of the present invention.

[0032] Figure 2 This is a system logic architecture diagram provided in an embodiment of the present invention.

[0033] Figure 3 This is a schematic diagram comparing the OCR recognition results before and after anti-perturbation processing.

[0034] Figure 4 These are comparison images of the interface effects of semantic-level simulation desensitization. Detailed Implementation

[0035] Example 1: Implementation of Render Interception and Context Awareness

[0036] A hook is attached at the entry point of the `repaintEverything` function in `SurfaceFlinger`, and the list of currently active displays is traversed by calling the Android system's `DisplayManager.getDisplays()` interface. By checking the `FLAG_PRESENTATION` or `FLAG_PRIVATE` flags of the display objects, it is determined whether a virtual display environment created by screen recording, screen casting, or third-party interception applications exists.

[0037] The system maintains a registry of sensitive views internally. This registry uses a high-performance hash table structure to store the ID of the view or the normalized coordinate range to achieve O(1) level fast matching.

[0038] like Figure 1 As shown, when interception is detected (step S104), the system splits the rendering instruction stream into two parallel paths at step S105:

[0039] Path A (Local Path): Maintains the original rendering instructions, ensuring that the Main Display obtains the original memory block, and the user experience remains realistic and lossless.

[0040] Path B (Safe Path): Triggers the differential rendering engine, performs desensitization processing on sensitive areas, and outputs the data to the relevant rendering buffer. Both paths are submitted synchronously in step S106, ensuring a high degree of consistency between local display and captured data in the time dimension.

[0041] Algorithm flow for adversarial noise generation:

[0042] Input: Original text region texture T, pre-trained adversarial noise model M.

[0043] Step 1: Calculate the gradient. Calculate the gradient G = ∇(OCR_Loss(T)) for the general OCR loss function based on model M.

[0044] Step 2: Noise generation. Construct adversarial disturbance noise δ = ε × sign(G), where ε is a coefficient controlling the disturbance intensity (preferably 0.01-0.03).

[0045] Step 3: Texture blending. Calculate T' = T + α×δ, where α is the transparency weight (Alpha must be less than 0.05).

[0046] Step 4: Quality Verification. Evaluate the peak signal-to-noise ratio (PSNR) between T and T'. If PSNR > 35dB (ensuring a natural screen protection display), output the protective texture T'; otherwise, degrade to a normal masking.

[0047] Example 2: A text anti-identification method based on adversarial perturbation. For high-security text (such as ID card numbers and chat logs), the system enables adversarial rendering mode.

[0048] Feature and saliency extraction: Extract gradient features of the text region to be rendered.

[0049] Noise generation algorithm: A lightweight adversarial example generation network is used, and inference is performed on the terminal NPU based on the Android NNAPI or Qualcomm SNPE framework.

[0050] Real-time fusion solution: NPU-accelerated noise operators ensure that the single-frame injection latency is less than 5ms in tests on Snapdragon 8 Gen 2 and above devices.

[0051] Degradation strategy: If the NPU load is too high or the hardware does not support it, the system will automatically switch to the pre-generated static noise pattern mode.

[0052] Technical parameters: In terms of processing latency, the single-frame injection time is less than 2ms; in terms of security, the OCR omission rate of the adversarial image is over 95%, while the loss of human eye readability is less than 3%.

[0053] Example 3: Semantic desensitization in financial application scenarios. For financial APP interfaces, the system provides a seamless desensitization solution.

[0054] Bimodal recognition: The system parses the View Hierarchy tag, or uses the following regular expression library to match sensitive information:

[0055] Amount: ^(\$|￥)\d+(\.\d{2})?$

[0056] Bank card: ^(\d{4} ){3}\d{4}$ (Includes Luhn validation logic implementation)

[0057] ID card: ^\d{17}[\d|X]$

[0058] Simulation data generation: The DataFaker module randomly samples simulation data based on the identified type. Specifically, for bank card numbers, the system incorporates Luhn algorithm verification logic: by performing a weighted sum of the first 15 digits of the card number (odd / even bits) and calculating the check digit, it ensures that the generated 16-digit simulated card number passes the standard format verification, thus greatly enhancing its deceptiveness to interceptors.

[0059] Redraw command: Modify the drawText(fake_content, ...) parameter in the Skia command to ensure that the font size and color are fully inherited from the original drawing object.

[0060] Effect: In the screenshot captured by the attacker, the account balance or account information has been perfectly replaced with fake data, and there are no traces of modification, which greatly increases the complexity of the attacker's data screening.

[0061] Example 4: Hardware-level protection for payment password input for extremely high-security payment scenarios (such as payment password disks):

[0062] HWC layer allocation: Request an independent layer from the display controller and set its Flag to GRALLOC_USAGE_PROTECTED (typical value 0x00004000).

[0063] Physical path protection: The display engine outputs directly. If the system attempts to capture the screen using screencap, the capture area will be empty.

[0064] Limitations and Degradation: When the number of overlay layers exceeds the hardware limit (usually 4-8 layers), the system automatically degrades.

[0065] Additional notes: Legal compliance

[0066] This technology is solely for protecting users' sensitive personal privacy information and does not involve compromising the integrity of the Android system or evading regulation. All interception actions are performed only within a security context authorized by the user or declared by the application developer.

[0067] Informed consent: Before enabling anti-interception functions, systems or applications must clearly inform users of their protection mechanisms (such as the possible subtle background textures) and explain them in the privacy policy.

[0068] Processing principle: Following the "minimum necessary" principle, only predefined or dynamically identified sensitive areas are processed, and data in non-sensitive areas is not retained or modified.

[0069] Through the above-mentioned multi-layered and multi-dimensional combination of technologies, this invention greatly enhances the protection boundary of sensitive information on mobile devices.

Claims

1. A method for preventing the interception of sensitive information on mobile devices based on interface rendering interception, characterized in that, Includes the following steps: Step 1): Implant a rendering interception hook in the graphics rendering pipeline of the mobile terminal; Step 2): Capture the rendering command stream or layer composition request of the current interface; Step 3): Analyze the rendering instruction stream or layer composition request to identify the sensitive information areas contained therein; Step 4): Real-time detection of the current rendering target context to determine whether there is any capture behavior that is not displayed on the local physical screen; Step 5): If the interception behavior is determined to exist, the sensitive information area is subjected to differentiated rendering processing according to the preset security policy to generate processed rendering data that is different from the original rendering data. If it is determined that the aforementioned interception behavior does not exist, the normal original layer composition and display process is maintained; Step 6): Output the processed rendering data to the rendering buffer corresponding to the interception behavior, while continuing to output the original rendering data to the local physical screen.

2. The method for preventing interception of sensitive information on mobile devices based on interface rendering interception according to claim 1, characterized in that: In step 1), the location where the rendering interception hook is implanted includes at least one of the following: the layer composition loop entry of the graphics compositing component SurfaceFlinger; the drawing command call interface of the graphics drawing library Skia; and the command buffer submission interface of the graphics driver layer Vulkan or OpenGLES.

3. The method for preventing interception of sensitive information on mobile devices based on interface rendering interception according to claim 1, wherein its features are as follows: The characteristic is that, in step 5), the differentiated rendering process includes one or any combination of the following processing methods: Step 5.1) Anti-perturbation processing: Calculate the text texture features of the sensitive information region; Adversarial sample noise is generated based on neural networks and then superimposed onto the texture of sensitive information regions. The superimposed sensitive information area maintains a natural display effect in human vision, and the visual quality loss is within a preset threshold. It can interfere with the optical character recognition (OCR) algorithm's feature extraction of the sensitive information area. Step 5.2) Semantic-level simulation desensitization processing: Analyze the semantic category of the original data in the sensitive information area. The semantic category includes at least one of the following: amount, name, ID card number, bank card number, mobile phone number, and address; call the simulation data generation engine to generate fake simulation data that is consistent with the semantic category and format of the original data; By replacing the original rendering parameters in the rendering instruction stream with fake simulation data, a desensitized interface for semantic simulation is constructed and displayed. Step 5.3) Hardware-level layer separation processing: Utilize the hardware composition capability of the display controller to mark sensitive information areas as protected hardware layers; configure the filtering rules of the hardware compositor so that the protected hardware layers are only output to the local physical screen, and are automatically removed or filled with preset obfuscated content when written to the write-back buffer or frame buffer.

4. The method for preventing interception of sensitive information on mobile devices based on interface rendering interception according to claim 1, characterized in that: The security level of the sensitive information dynamically selects the type of differentiated rendering processing, wherein the higher the security level, the stronger the corresponding rendering protection.

5. An anti-interception system used in the mobile terminal sensitive information anti-interception method based on interface rendering interception as described in any one of claims 1-3, characterized in that, include: The rendering pipeline interception module is used to implant interceptors in the graphics rendering pipeline to capture rendering command streams or layer composition requests. A sensitive identification engine is used to analyze the rendering instruction stream or layer composition request to identify the sensitive information areas contained therein; The context detection unit is used to monitor the state of the rendering target in real time and determine whether there is any interception behavior that is not displayed on the local physical screen. The differential processing engine is used to perform differential rendering processing on the sensitive information area according to a preset security policy when interception behavior is detected, including anti-disturbance injection processing, simulation data replacement processing, or hardware layer isolation.

6. An anti-interception system according to claim 5, characterized in that: The differentiated processing engine includes: an adversarial noise generation submodule, used to generate defensive noise against a specific OCR engine through a generative adversarial network model; and a simulation data generation submodule, used to generate simulated data that meets the requirements of the original data format and perform instruction-level replacement.

7. An electronic device comprising a memory and a processor, characterized in that, When the processor executes the computer program stored in the memory, it implements the method as described in any one of claims 1 to 3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.

Images