MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202631069698 A) filed by National Institute Of Technology Jamshedpur on June 03, 2026, for System And Method For Rotation-Invariant Iris Recognition Through Graph-Based Structural Analysis.
Inventors include Nushrat Praveen; Prof. Dr. Danish Ali Khan; Dr. Alekha Kumar Mishra; Dr. Rashmi Panda; and Dr. Deepak Rai.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: The invention relates to a system and method for rotation-invariant iris recognition utilizing graph-based structural analysis. The system comprises an acquisition interface for receiving a digital iris image and a spatial-to-graph transformation unit that maps sampling points in concentric rings to extract localized patches. These patches are converted into high-dimensional node representations using convolutional operations and positional encodings. A topological edge construction unit establishes connectivity between nodes based on polar distances, accounting for angular wrap-around to ensure rotational invariance. A multi-stage graph signal processor utilizes multi-head attention mechanisms to refine structural embeddings by aggregating features from neighboring nodes. A feature fusion unit performs dual-pooling to generate a composite graph-level vector, which is projected into a discriminative embedding space for identity verification. By modeling the iris as a polar-coordinate graph and employing contrastive learning with hard negative mining, the invention effectively handles natural angular variations and occlusions for robust biometric matching.
Disclaimer: Curated by HT Syndication.
