Explicit Over Implicit: Enhancing CNNs Via Complex Structure Tensor Representations for Periocular Recognition
Computer Vision and Pattern Recognition
Summary
The authors found that standard CNNs have trouble recognizing orientation information in images. They used a special input called the Complex Structure Tensor, which packages orientation details with confidence levels, to help CNNs perform better. Their experiments showed that CNNs fed with these orientation-based inputs worked better and even allowed the use of smaller CNN models without losing accuracy. They tested this approach on eye region image datasets for biometric ID and showed it reduced error rates by 5-26%. This suggests adding orientation information upfront helps CNNs understand images more effectively and reliably.
Authors
Kevin Hernandez-Diaz, Josef Bigun, Fernando Alonso-Fernandez
Abstract
Our study provides evidence that CNNs struggle to extract orientation features effectively. We show that using the Complex Structure Tensor, which contains compact orientation features with certainties, as input to CNNs consistently improves identification accuracy compared to grayscale inputs alone. Experiments also demonstrated that our inputs, provided by mini-complex convnets, combined with reduced CNN sizes, outperformed full-fledged, prevailing CNN architectures. This suggests that the upfront use of orientation features in CNNs, a strategy seen in mammalian vision, not only mitigates their limitations but also enhances their explainability and relevance to thin-clients. Experiments were conducted on publicly available datasets comprising periocular images (Cross-Eyed and PolyU) for biometric identification and verification in both Close-World and Open-World Scenarios using six CNN architectures. Our experiments on the Cross-Eyed and PolyU datasets yield a 5-26% reduction in EER, providing strong empirical evidence that explicit orientation priors mitigate CNN representational limits in Open-World and Close-World scenarios.