Low-Touch Biometric Matching With Winnowing and Dominant Face Capture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-touch biometric systems face challenges with accuracy, latency, and throughput due to large gallery sizes, especially in facial recognition, where accuracy degrades significantly as the gallery size grows, and failovers to storage galleries result in increased latency and false positives/negatives.
Innovation Solution
Implementing subset galleries stored on faster, more accessible devices and using winnowing information to obtain biometric data from associated groups when probes fail, along with determining a dominant face for enhanced capture and control, to improve accuracy and reduce latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If biometric probes are matched against large storage galleries, then comprehensive identification coverage is achieved, but accuracy degrades significantly and latency increases
Solution Approach 1:
The patent divides the large storage gallery into multiple subset galleries organized in a hierarchical structure. The system first performs matching against smaller subset galleries, and only fails over to the full storage gallery when necessary. This segmentation maintains identification accuracy by limiting initial comparisons to manageable subsets while preserving comprehensive coverage through the fallback mechanism.
Solution Approach 2:
The patent introduces winnowing information as an intermediary element that bridges subset galleries and the full storage gallery. When a probe fails to match in subset galleries, winnowing information is obtained to filter and narrow down the search space in the full gallery, enabling accurate matching without comparing against all entries simultaneously.
2Reliability
If failover to storage galleries is implemented when subset matching fails, then identification coverage is maintained, but latency increases
Solution Approach 1:
The patent performs preliminary matching against subset galleries before attempting full gallery matching. This preliminary action filters out many obvious matches early in the process, reducing the number of cases that require expensive full gallery searches and thereby reducing overall latency while maintaining coverage.
Solution Approach 2:
Winnowing information serves as an intermediary that accelerates the failover process. When subset matching fails, the winnowing information pre-filters the full gallery candidates, so the system doesn't need to perform exhaustive comparisons against all storage gallery entries, thus reducing the latency penalty of failover.
3Reliability
If full gallery matching is performed, then all possible matches are identified, but false positives and false negatives increase
Solution Approach 1:
By segmenting the gallery into subsets, the patent reduces the probability of false positives in initial matching stages. The hierarchical structure allows the system to make confident matches against small subsets before considering the full gallery, thereby improving overall match precision while maintaining completeness through controlled failover.
Solution Approach 2:
Winnowing information acts as an intermediary filtering layer that reduces false positives during failover. It pre-processes the full gallery candidates to eliminate unlikely matches before detailed comparison, thereby maintaining match completeness while improving accuracy by reducing false positive rates.
Data Source
AI summary
In various embodiments, enhancements for low-touch biometric systems include obtaining winnowing information when a biometric probe unsuccessfully matches against a subset gallery. The winnowing information is used to obtain biometric data from an associated group of people and the biometric probe is then compared against the biometric data. If this also fails, identity information may be obtained and used to obtain a corresponding biometric template to match the biometric probe against. In some embodiments, enhancements for low-touch biometric systems determining a dominant by selecting the largest live face in an image then performing one or more actions using the dominant. Such one or more actions may include selecting the dominant for capture, performing exposure control and/or gain control on the dominant as opposed to the rest of an image, determining guidance to provide for biometric capture based at least on a position of the dominant, providing such guidance, etc.