Adaptive multi-modal iris-periorbital fusion recognition system and method based on transformer

Abstract

This invention discloses an adaptive multimodal iris-periocular fusion recognition system and method based on Transformer, belonging to the field of biometric recognition technology. It includes: an image preprocessing module for enhancing and denoising images through contrast-limited adaptive histogram equalization and guided filtering; a bimodal segmentation module for generating iris and periocular region masks using a U-Net network; a feature extraction module for extracting global feature vectors using a Vision Transformer architecture and outputting a quality score by combining it with a quality regression head, or simulating feature vectors based on Log-Gabor density statistics in verification mode; and an adaptive fusion matching module for calculating a comprehensive quality score based on image contrast, sharpness, and coverage, dynamically adjusting the weighting of iris and periocular features using this score, and calculating the final matching score. This invention effectively solves the problem of poor robustness in low-quality image recognition in non-cooperative scenarios, significantly improving recognition accuracy and system efficiency.

Images