Document Security via Feature Vector Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Devices such as printers, laptops, and smartphones often lack effective mechanisms to prevent unauthorized capture and sharing of documents with confidential or privileged information, despite existing security measures, as they typically require capturing and comparing the document's content to enforce restrictions.
Innovation Solution
A document security system that uses a machine-learning based heuristic extraction model to generate a feature vector representation of a document, which obfuscates the content and provides a unique signature for security rules, allowing for privacy and security guarantees without storing the document itself, using a multimodal space to compare and identify matching features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the document content is captured and stored to enforce security restrictions, then security enforcement capability is improved, but confidentiality and privacy protection deteriorate
Solution Approach 1:
The patent extracts only the essential features needed for security verification (text content, layout, images) from the complete document, storing them as separate feature vectors rather than storing the entire document. This allows security checks to be performed on extracted features without exposing or storing the full confidential document content.
Solution Approach 2:
The patent transforms the document from its original form into a different representation (feature vectors with specific parameters like text content, layout coordinates, image hashes). These parameter-based representations enable security enforcement while maintaining confidentiality, as the parameters can be verified without reconstructing the original confidential document.
2Measurement precision
If the complete document is stored in the database to check for unauthorized copies, then accuracy of security checking is improved, but storage requirements and processing time deteriorate
Solution Approach 1:
The patent segments the document into multiple independent feature vectors (text features, layout features, image features, metadata features). Each feature can be processed and compared independently during security checks, reducing the overall processing time while maintaining accurate verification through the combination of multiple feature comparisons.
Solution Approach 2:
The patent extracts only the necessary feature vectors from the complete document for storage and comparison. By taking out and storing only these essential features rather than the entire document, the system achieves fast processing during security checks while maintaining sufficient accuracy through multi-feature verification.
3Reliability
If the document content is captured to verify security rules, then security rule enforcement is improved, but privacy protection and security guarantees deteriorate
Solution Approach 1:
The patent introduces feature vectors as intermediary representations between the original confidential document and the security verification process. These intermediaries (extracted features like text content, layout, images) enable security rule enforcement without directly exposing or storing the full confidential document, thus protecting privacy while maintaining enforcement capability.
4Measurement precision
If multiple feature types are extracted and stored for comprehensive security checking, then security verification accuracy is improved, but device complexity and storage requirements deteriorate
Solution Approach 1:
The patent segments the security verification system into multiple independent feature extraction and comparison modules (text feature module, layout feature module, image feature module, metadata feature module). Each module handles a specific feature type independently, which improves verification accuracy through comprehensive checking while managing complexity through modular organization.
Data Source
AI summary
Examples disclosed herein relate to extracting a plurality of features from a document according to a heuristic extraction model, generating a vector representation of the document according to the plurality of extracted features, associating at least one security rule with the document, determining if a received vector representation matches the stored vector representation of the document, and providing a response to a requested action on the document according to the determination.