Intelligent verification code system and method based on multi-modal dynamic context perception

Abstract

The application discloses a multi-modal dynamic context-aware intelligent verification code system and method, and relates to the technical field of network security; the system comprises the following steps: 1, capturing the user operation context in real time, and generating visual content associated with the scene through a semantic analysis engine; 2, synchronously collecting touch biometric features, device sensor data and environmental parameters according to the user operation; 3, dynamically selecting a verification code generation strategy based on a risk index, and the strategy gradient comprises: a static default code, behavior verification and live detection; according to the verification code generation strategy, the visual content, the biometric features, the device sensor data and / or the environmental parameters are used to generate the verification code.

Description

Technical Field

[0001] This invention discloses an intelligent CAPTCHA system and method based on multimodal dynamic context awareness, which relates to the field of network security technology. Background Technology

[0002] Currently, there are still some shortcomings in distinguishing between humans and automated programs, such as robots and web crawlers. For example, at the technical level, CAPTCHAs still have certain limitations, which also impairs the user experience; there are also misjudgments in behavioral analysis, as some automated programs can simulate some human behavioral patterns and are difficult to distinguish; and biometric features are also at risk of being forged, posing security risks. Summary of the Invention

[0003] This invention addresses the problems of existing technologies by providing an intelligent CAPTCHA system and method based on multimodal dynamic context awareness. It integrates multimodal CAPTCHA technology with user behavior characteristics, device sensor data, and environmental context to distinguish between human users and automated programs.

[0004] The specific solution proposed in this invention is as follows:

[0005] This invention provides an intelligent CAPTCHA generation method based on multimodal dynamic context awareness, comprising:

[0006] Step 1: Capture user operation context in real time and generate scene-related visual content through a semantic analysis engine;

[0007] Step 2: Synchronously collect touch biometric data, device sensor data, and environmental parameters based on user operations;

[0008] Step 3: Dynamically select the CAPTCHA generation strategy based on the risk index. The strategy gradient includes: static verification → behavioral verification → liveness detection. CAPTCHAs are generated using visual content, biometrics, device sensor data, and / or environmental parameters according to the CAPTCHA generation strategy.

[0009] Furthermore, the visual content generated in step 1 of the intelligent CAPTCHA generation method based on multimodal dynamic context awareness includes:

[0010] Retrieve entities semantically related to the current scene using a knowledge graph;

[0011] Construct challenge content that includes antonymous related distractors;

[0012] An adversarial noise pattern targeting the neural network is superimposed on the image.

[0013] Furthermore, in step 1 of the intelligent CAPTCHA generation method based on multimodal dynamic context awareness, the adversarial noise pattern for the neural network includes:

[0014] Add pixel-level perturbations generated by FGSM;

[0015] Notch filters are used to add interference bands to the sensitive frequency bands of CNNs;

[0016] Add artistic texture obfuscation based on neural style transfer, with style loss weights greater than or equal to preset values.

[0017] Furthermore, step 3 of the aforementioned intelligent CAPTCHA generation method based on multimodal dynamic context awareness utilizes the following formula:

[0018] Risk=A×DeviceTrust+A×BehaviorAnomaly+B×EnvRisk

[0019] The Risk index is calculated as follows: DeviceTrust represents device trustworthiness, scored by a rule engine or machine learning model, with a score range of 0-100. Preset high and low score segments are provided; a high score indicates high device trustworthiness, and a low score indicates a risky device. BehaviorAnomaly measures the degree of behavioral anomaly, measuring the deviation of user behavior from historical patterns, obtained by calculating the deviation of current behavior from a historical baseline. EnvRisk assesses the potential risks in the user's environment, scored by a rule engine according to preset rules. A and B are preset specific values, where A represents the weight of device and behavior, and B represents the weight of environmental risk.

[0020] Liveness detection is forcibly started when EnvRisk≥B.

[0021] Furthermore, in step 3 of the intelligent verification code generation method based on multimodal dynamic context awareness, when generating the verification code based on visual content, a semantic analysis engine is used to dynamically generate topological deformation problems related to the user's operation content. The topological deformation problems include the Möbius strip path selection problem, the prediction problem of cutting results of high-genus surfaces (g≥a specific value), or the equivalence judgment problem of continuous deformation of Klein bottle surface, which require human spatial intuition to solve.

[0022] Furthermore, in step 3 of the intelligent verification code generation method based on multimodal dynamic context awareness, the verification code is generated according to the device sensor data and environmental parameters, including: performing environmental consistency verification, wherein the gravitational acceleration is detected to be consistent with the required spatial direction, and the x / y axis deviation is ≤ a preset value;

[0023] Verify the correlation between screen brightness and ambient light, |lux_env-lux_screen| < preset value;

[0024] Verify the binding relationship between magnetic field strength and device fingerprint.

[0025] Furthermore, in step 2 of the intelligent verification code generation method based on multimodal dynamic context awareness, when collecting touch biometric features, the touch pressure spectrum and the trajectory fractal dimension are measured, D = 2 - H, where D is the fractal dimension and H is the Hurst exponent (Hurst exponent ∈ [0,1]).

[0026] This invention also provides an intelligent CAPTCHA system based on multimodal dynamic context awareness, including a capture module, a collection module, and a generation module.

[0027] The capture module captures the user's operation context in real time and generates scene-related visual content through the semantic analysis engine;

[0028] The data acquisition module synchronously collects touch biometric data, device sensor data, and environmental parameters based on user operations;

[0029] The generation module dynamically selects a CAPTCHA generation strategy based on the risk index. The strategy gradient includes: static verification → behavioral verification → liveness detection. Based on the CAPTCHA generation strategy, the module generates CAPTCHAs using visual content, biometrics, device sensor data, and / or environmental parameters.

[0030] The advantages of this invention are:

[0031] This invention can integrate user behavior characteristics, device sensor data, and environmental context selection strategies to generate multimodal CAPTCHAs, which can be used to distinguish between human users and automated programs. It solves the problems of CAPTCHA limitations, impaired user experience, and misjudgment in behavior analysis, and can effectively distinguish automated programs from human behavior. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0035] Example 1

[0036] This invention provides an intelligent CAPTCHA generation method based on multimodal dynamic context awareness, comprising:

[0037] Step 1: Capture user operation context in real time and generate scene-related visual content through a semantic analysis engine. The generated visual content includes:

[0038] Retrieve entities semantically related to the current scene using a knowledge graph;

[0039] Construct challenge content that includes antonymous related distractors;

[0040] An adversarial noise pattern targeting a neural network is superimposed onto the image. This adversarial noise pattern includes:

[0041] Add pixel-level perturbations generated by FGSM;

[0042] Notch filters are used to add interference bands to the sensitive frequency bands of CNNs;

[0043] Add artistic texture obfuscation based on neural style transfer, with style loss weights greater than or equal to preset values.

[0044] Step 2: Synchronously collect touch biometrics, device sensor data, and environmental parameters based on user operations. When collecting touch biometrics, measure the touch pressure spectrum and the trajectory fractal dimension, D = 2 - H, where D is the fractal dimension and H is the Hurst exponent (Hurst exponent ∈ [0,1]).

[0045] Step 3: Dynamically select the CAPTCHA generation strategy based on the risk index. The strategy gradient includes: static verification → behavioral verification → liveness detection. CAPTCHAs are generated using visual content, biometrics, device sensor data, and / or environmental parameters according to the CAPTCHA generation strategy.

[0046] The following formula can be used:

[0047] Risk=A×DeviceTrust+A×BehaviorAnomaly+B×EnvRisk

[0048] The Risk index is calculated as follows: DeviceTrust represents device trustworthiness, scored by a rule engine or machine learning model, with a score range of 0-100. Preset high and low score segments are provided; a high score indicates high device trustworthiness, and a low score indicates a risky device. BehaviorAnomaly measures the degree of behavioral anomaly, measuring the deviation of user behavior from historical patterns, obtained by calculating the deviation of current behavior from a historical baseline. EnvRisk assesses the potential risks in the user's environment, scored by a rule engine according to preset rules. A and B are preset specific values, where A represents the weight of device and behavior, and B represents the weight of environmental risk.

[0049] Liveness detection is forcibly started when EnvRisk≥B.

[0050] When generating CAPTCHAs based on visual content, a semantic analysis engine can be used to dynamically generate topological deformation problems related to user operations. These topological deformation problems include the Möbius strip path selection problem, the prediction problem of cutting results of high-genus surfaces (g≥a specific value), or the problem of determining the equivalence of continuous deformation on the surface of a Klein bottle, which require human spatial intuition to solve.

[0051] The verification code is generated based on the device sensor data and environmental parameters. It may include: performing environmental consistency verification, which includes detecting the consistency between the gravitational acceleration and the required spatial direction, and the x / y axis deviation ≤ preset value.

[0052] Verify the correlation between screen brightness and ambient light, |lux_env-lux_screen| < preset value;

[0053] Verify the binding relationship between magnetic field strength and device fingerprint.

[0054] For example, in an e-commerce scenario, generating a mobile immersive CAPTCHA involves generating semantically relevant visual content based on the product categories the user is browsing. If the user is viewing "mountaineering equipment," the code could be represented as:

[0055]

[0056] Output effect: Against a snow-capped mountain background, the user is prompted to "select essential mountaineering items".

[0057] Example 2

[0058] This invention also provides an intelligent CAPTCHA system based on multimodal dynamic context awareness, including a capture module, a collection module, and a generation module.

[0059] The capture module captures the user's operation context in real time and generates scene-related visual content through the semantic analysis engine;

[0060] The data acquisition module synchronously collects touch biometric data, device sensor data, and environmental parameters based on user operations;

[0061] The generation module dynamically selects a CAPTCHA generation strategy based on the risk index. The strategy gradient includes: static verification → behavioral verification → liveness detection. Based on the CAPTCHA generation strategy, the module generates CAPTCHAs using visual content, biometrics, device sensor data, and / or environmental parameters.

[0062] The information interaction and execution process between the modules in the above system are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description in the method embodiment of the present invention, and will not be repeated here.

[0063] Similarly, the system of the present invention can integrate user behavior characteristics, device sensor data and environmental context selection strategies to generate multimodal CAPTCHAs, which can be used to distinguish between human users and automated programs, solve the problems of CAPTCHA limitations, impaired user experience and misjudgment in behavior analysis, and effectively distinguish automated programs from human behavior.

[0064] It should be noted that not all steps and modules in the above processes and some system structures are necessary; certain steps or modules can be omitted as needed. The execution order of each step is not fixed and can be adjusted as required. The system structures described in the above embodiments can be physical or logical structures. That is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0065] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A method for generating intelligent CAPTCHAs based on multimodal dynamic context awareness, characterized by: include: Step 1: Capture user operation context in real time and generate scene-related visual content through a semantic analysis engine; The visual content includes: retrieving entities semantically related to the current scene through a knowledge graph; constructing challenging content containing antonymous associations; and overlaying adversarial noise patterns targeting neural networks onto the image. Adversarial noise patterns for neural networks include: adding pixel-level perturbations generated by FGSM; adding interference bands using notch filters for CNN-sensitive frequency bands; and adding artistic texture obfuscation based on neural style transfer, with style loss weights ≥ preset values. Step 2: Synchronously collect touch biometric data, device sensor data, and environmental parameters based on user operations; Step 3: Dynamically select the CAPTCHA generation strategy based on the risk index, using the following formula: Risk=A×DeviceTrust+A×BehaviorAnomaly+B×EnvRisk The Risk index is calculated as follows: DeviceTrust represents device trustworthiness, scored by a rule engine or machine learning model, with a score range of 0-100. Preset high and low score ranges are used; high scores indicate high device trustworthiness, while low scores indicate risky devices. BehaviorAnomaly measures the degree of behavioral anomaly, measuring the deviation of user behavior from historical patterns, obtained by calculating the deviation of current behavior from a historical baseline. EnvRisk assesses environmental risk, evaluating the potential risks in the user's environment. It is scored by a rule engine according to preset rules, with A and B being preset specific values. A represents the weight of device and behavior, and B represents the environmental risk weight. When EnvRisk ≥ B, liveness detection is forcibly initiated. The strategy gradient includes: static verification → behavioral verification → liveness detection. Based on the verification code generation strategy, the verification code is generated using visual content, biometrics, device sensor data and / or environmental parameters. The generation of the verification code based on device sensor data and environmental parameters includes: performing environmental consistency verification, which involves detecting the consistency between the gravitational acceleration and the required spatial direction, and the x / y axis deviation ≤ preset value. Verify the correlation between screen brightness and ambient light, |lux_env-lux_screen| < preset value; Verify the binding relationship between magnetic field strength and device fingerprint.

2. The intelligent CAPTCHA generation method based on multimodal dynamic context awareness according to claim 1, characterized in that: In step 3, when generating the verification code based on the visual content, the semantic analysis engine is used to dynamically generate topological deformation problems related to the user's operation content. These topological deformation problems include the Möbius strip path selection problem that requires human spatial intuition to solve, the problem of predicting the cutting result of a high-genus surface g ≥ a specific value, or the problem of judging the equivalence of continuous deformation on the surface of a Klein bottle.

3. The intelligent CAPTCHA generation method based on multimodal dynamic context awareness according to claim 1, characterized in that: In step 2, when collecting touch biometric features, the touch pressure spectrum and the trajectory fractal dimension are measured, D=2-H, where D is the fractal dimension and H is the Hurst exponent, where the Hurst exponent ∈ [0,1].

4. An intelligent CAPTCHA system based on multimodal dynamic context awareness, characterized in that: It includes a capture module, an acquisition module, and a generation module. The capture module captures the user's operation context in real time and generates scene-related visual content through a semantic analysis engine. The visual content includes: entities that are semantically related to the current scene by retrieving them from a knowledge graph; constructing challenge content that includes antonymous interference items; and overlaying adversarial noise patterns targeting neural networks into the image. Adversarial noise patterns for neural networks include: adding pixel-level perturbations generated by FGSM; adding interference bands using notch filters for CNN-sensitive frequency bands; and adding artistic texture obfuscation based on neural style transfer, with style loss weights ≥ preset values. The data acquisition module synchronously collects touch biometric data, device sensor data, and environmental parameters based on user operations; The generation module dynamically selects the CAPTCHA generation strategy based on the risk index, using the following formula: Risk=A×DeviceTrust+A×BehaviorAnomaly+B×EnvRisk The generation module calculates the Risk index. DeviceTrust represents device trustworthiness, scored by a rule engine or machine learning model, with a score range of 0-100. Preset high and low score segments are provided; a high score indicates high device trustworthiness, and a low score indicates a risky device. BehaviorAnomaly measures the degree of behavioral anomaly, quantifying the deviation of user behavior from historical patterns, obtained by calculating the deviation of current behavior from a historical baseline. EnvRisk assesses environmental risk, evaluating the potential risks of the user's environment. It is scored by a rule engine based on preset rules, with A and B being preset specific values. A represents the weight of device and behavior, and B represents the environmental risk weight. When EnvRisk ≥ B, liveness detection is forcibly initiated. The strategy gradient includes: static verification → behavioral verification → liveness detection. The generation module generates a verification code based on the verification code generation strategy, using visual content, biometrics, device sensor data and / or environmental parameters. The generation module generates a verification code based on device sensor data and environmental parameters, including: performing environmental consistency verification, which involves detecting the consistency between the gravitational acceleration and the required spatial direction, and the x / y axis deviation ≤ preset value. Verify the correlation between screen brightness and ambient light, |lux_env-lux_screen| < preset value; Verify the binding relationship between magnetic field strength and device fingerprint.

Images