System and method for rotation invariant fingerprint recognition
US12223764B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Feb 28, 2022 |
| Grant date | Feb 11, 2025 |
| Priority date | — |
| Expiry date | Mar 26, 2043 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06V40/1365
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
The present teaching relates to method, system, medium, and implementations for biometric authentication. Rotation covariant convolution kernels at multiple convolution layers are obtained with weights learned via machine learning based on rotation invariant (RI) training data. For an input image with fingerprint information captured therein related to a person to be authenticated, an initial feature map is obtained and then at each convolution layer, a feature map is processed based on the RC convolution kernels for the layer to output a rotation covariant (RC) feature map, with the feature map being either the initial feature map and an output RC feature map from a previous convolution layer. The last convolution layer outputs a rotation invariant (RI) feature vector representing fingerprint features of the person in a rotation invariant manner, which is then used to authenticate the person.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.