Drawing Capability Authentication via Generative Adversarial Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern authentication systems are vulnerable to hacking and exploits, as they rely on measurable biometrics or passcodes that can be compromised, and lack supplementary protection against unauthorized access, especially for intangible personal traits.
Innovation Solution
A method utilizing a generative adversarial network (GAN) to authenticate users based on their unique drawing capabilities, where users are presented with drawing challenges, and their submissions are compared to historical drawing samples to determine authenticity, making it impractical to reverse-engineer the credentials.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional authentication methods (passwords, biometrics) are used, then user identification can be achieved, but the system becomes vulnerable to hacking and credential compromise
Solution Approach 1:
The patent transforms authentication from verifying static credentials (passwords, biometric templates) to verifying dynamic drawing behavior parameters. The system analyzes multiple parameters including stroke pressure, velocity, acceleration, and temporal patterns to create a behavioral biometric profile that is extremely difficult to compromise through traditional hacking methods
Solution Approach 2:
The patent replaces mechanical authentication systems (physical passwords, fingerprint sensors) with an AI-based analytical system using deep learning models. The neural network processes drawing behavior data and generates authentication decisions, substituting physical credential verification with computational behavioral analysis
2Measurement precision
If biometric authentication is implemented, then unique user identification is possible, but the biometric data can be compromised and reverse-engineered
Solution Approach 1:
The patent segments the authentication process into multiple independent behavioral dimensions (stroke pressure, velocity, acceleration, temporal patterns). Each dimension is analyzed separately and combined to form a comprehensive authentication decision, making it difficult for attackers to compromise the entire system by obtaining a single data point
Solution Approach 2:
The patent adds temporal and dynamic dimensions to authentication by analyzing how drawing actions unfold over time. Instead of static biometric images, the system processes time-series data from drawing motions, creating a multi-dimensional behavioral signature that is far more resistant to spoofing and reverse-engineering
3Reliability
If drawing capability authentication is used, then intangible personal traits can be leveraged for security, but the system complexity increases
Solution Approach 1:
The patent implements self-service through automatic behavioral analysis and continuous profile updates. The system automatically captures drawing behavior, processes it through AI models, and updates authentication profiles without manual intervention, reducing operational complexity despite the advanced algorithms involved
Solution Approach 2:
The patent performs preliminary training during an onboarding phase where users create reference drawing samples. This preliminary action establishes baseline behavioral profiles before actual authentication begins, simplifying the ongoing authentication process by having pre-computed reference data for comparison
4Reliability
If multiple authentication layers are implemented, then security is improved, but the user operation time increases
Solution Approach 1:
The patent merges multiple authentication验证 into a single drawing action. The drawing behavior simultaneously verifies user identity, device ownership, and intent, combining what would traditionally require separate steps (password entry, CAPTCHA verification, biometric scan) into one unified authentication gesture
Data Source
AI summary
A person's drawing capability is used as an authentication credential. During a training phase, a user is asked to hand draw various reference shapes such as a rectangle, flower, etc. These user drawings for a given shape are input to a training discriminator (with an “authentic” label) along with drawings automatically generated from a latent sample (with a “not authentic” label), and the training discriminator computes positive discrimination vectors for this shape that are unique to this user. Thereafter, when the user wants access to a resource (such as an online account, a mobile computing device, or an electronic document), the user is presented with a drawing challenge for one of the reference shapes, and they draw a corresponding challenge image. An image vector for the challenge image is generated, and if the image vector falls within the positive discrimination vectors, access to the resource is granted.