Document Access Control via Token Alignment Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current document-based access control methods face challenges with high error rates in automated identification due to poor scanning quality and variability in credentials, leading to insufficient security and performance issues, particularly in real-time applications like border control.
Innovation Solution
A computer-implemented method that tokenizes optical patterns from documents, creates extended sets of tokens through concatenation, and computes candidate-alignment scores to accurately match search strings against reference strings, ensuring fast and accurate access control by evaluating millions of entries in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated optical scanning and OCR processes are used for document-based access control, then processing speed and automation are improved, but error rates increase leading to false negatives and misreads
Solution Approach 1:
The patent segments the credential string into multiple tokens (words or character groups) rather than processing it as a single continuous string. This segmentation allows the system to create multiple candidate alignments by different ways of grouping and matching tokens, thereby reducing the impact of individual character recognition errors while maintaining fast automated processing.
2Measurement precision
If strict exact matching is used for credential verification, then false positives are reduced, but false negatives increase due to scanning errors
Solution Approach 1:
The patent implements a dynamic matching approach that generates multiple candidate alignments between search and reference tokens, then evaluates them using scoring functions. Instead of a static exact match, the system adapts by considering various possible alignments and selecting the best match based on computed scores, thereby reducing false negatives caused by scanning errors while maintaining precision.
3Adaptability or versatility
If multiple spelling variants and name formats are accommodated, then adaptability is improved, but processing complexity increases
Solution Approach 1:
By segmenting credentials into tokens and creating extended sets of candidate alignments through different grouping methods, the system handles spelling variants and format differences systematically. This segmentation approach manages complexity by breaking down the matching problem into smaller, more manageable token-level operations rather than dealing with entire variable-format strings at once.
4Measurement precision
If extensive candidate alignment evaluation is performed, then matching accuracy is improved, but processing time increases
Solution Approach 1:
The patent evaluates multiple candidate alignments beyond what a simple exact match would require, but not all possible combinations are fully processed. The system uses scoring functions to evaluate candidates and can stop when a sufficient match is found, performing partial evaluation that balances accuracy improvement with time constraints. This excessive action principle allows the system to consider more options than minimum required while avoiding exhaustive computation.
Data Source
Figure 1a
Figure 1b
Figure 2A
AI summary
The invention relates to an access control system (800,850) comprising: - a volatile or non-volatile data store (808) comprising one or more reference strings (204, 810) respectively representing a reference entity; - an optical sensor (864) configured to capture (110) an optical pattern (868) presented to the sensor on a document (866) assigned to a requesting entity (856); - a distributed or monolithic data processing system (800, 850, {900, 824}, {900, 802}) configured for: • decoding (111) the optical pattern to obtain a search string (202) representing the requesting entity (856); • tokenizing (112) the search string into search tokens (232, 234, 236); • creating (127, 128) an extended set of search tokens and reference tokens; • for each of the reference strings, computing a prediction score based on an artificially extended set of search tokens and reference tokens aligned to each other; and o selectively in case the prediction score obtained for all reference strings is above or below a prediction threshold, granting the requesting entity (856) access to an access-controlled resource.