Explicit Over Implicit: Enhancing CNNs Via Complex Structure Tensor Representations for Periocular Recognition

2026-07-16 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

A recent study demonstrates that Convolutional Neural Networks (CNNs) can be significantly enhanced for periocular recognition by explicitly providing Complex Structure Tensor (CST) representations as input. The research indicates that CNNs inherently struggle to extract orientation features effectively. By feeding CST, which contains compact orientation features with associated certainties, into CNNs, identification accuracy consistently improves compared to using grayscale inputs alone. Furthermore, the study found that inputs generated by mini-complex convnets, when combined with reduced CNN sizes, surpassed the performance of larger, prevailing CNN architectures. This strategy, mirroring mammalian vision's upfront use of orientation features, not only addresses CNN limitations but also boosts their explainability and suitability for thin-client applications. Experiments on the Cross-Eyed and PolyU datasets yielded a 5-26% reduction in Equal Error Rate (EER) across six CNN architectures in both Close-World and Open-World biometric scenarios.

Key takeaway

For Computer Vision Engineers developing biometric identification systems or resource-constrained vision applications, you should consider integrating explicit orientation features into your Convolutional Neural Network inputs. This approach, using Complex Structure Tensors, can significantly reduce Equal Error Rates by 5-26% and enable the use of smaller, more efficient CNN architectures. Evaluate the Complex Structure Tensor method to enhance both accuracy and explainability, especially for thin-client deployments where computational efficiency is critical.

Key insights

Explicit orientation features from Complex Structure Tensor inputs enhance CNN accuracy and efficiency for periocular recognition.

Principles

• CNNs struggle with implicit orientation features.
• Explicit orientation priors mitigate CNN limits.
• Upfront feature engineering boosts explainability.

Method

Generate compact orientation features with certainties using the Complex Structure Tensor. Feed these explicit features as input to CNNs, potentially via mini-complex convnets, allowing for reduced CNN sizes.

In practice

• Apply CST for periocular biometric tasks.
• Use reduced CNNs with explicit feature inputs.
• Explore orientation features for thin-client vision.

Topics

• Periocular Recognition
• Complex Structure Tensor
• Convolutional Neural Networks
• Biometric Identification
• Thin-Client AI
• Orientation Features

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Harnessing Generative Image Models for Training-Free Primitive Shape Abstraction — Pretrained generative image models can be directly harnessed for training-free 3D primitive shape abstraction, bypassing fine-tuning.
• Computer Vision : Chapter 1 — Convolutional Neural Networks evolved from biological vision models and mathematical convolution to automatically learn hierarchical image features.
• Upsampling method sharpens AI vision with up to 16 times less GPU memory — A new upsampling method sharpens AI vision and increases GPU memory efficiency by up to 16 times, accelerating on-device AI and robotics.
• FPN Paper Walkthrough: Leveraging the Internal Pyramid — FPN enriches multi-scale feature maps with both high spatial and semantic information for improved object detection.
• IVGT: Implicit Visual Geometry Transformer for Neural Scene Representation — IVGT implicitly models continuous 3D geometry from unposed multi-view images using a neural scene representation.
• Generation Models Know Space: Unleashing Implicit 3D Priors for Scene Understanding — Video generation models implicitly learn robust 3D structural priors, which can enhance MLLM spatial reasoning.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.