Enhancing CNNs robustness to occlusions with bioinspired filters for border completion

2025/04/24 by Coutinho, Catarina P., Merhab, Aneeqa, Petkovic, Janko +2
#Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2504.17619

Abstract

We exploit the mathematical modeling of the visual cortex mechanism for border completion to define custom filters for CNNs. We see a consistent improvement in performance, particularly in accuracy, when our modified LeNet 5 is tested with occluded MNIST images.

Citations

• Leveraging dendritic properties to advance machine learning and neuro-inspired computing
• A new perspective on border completion in visual cortex as bicycle rear wheel geodesics paths via sub-Riemannian Hamiltonian formalism
• Normative theory of visual receptive fields
• Random wiring, ganglion cell mosaics, and the functional architecture of the visual cortex
• Pyramidal Neuron as Two-Layer Neural Network
• Gradient-based learning applied to document recognition

Related

• Learning on the Edge: Explicit Boundary Handling in CNNs
• BIFRNet: A Brain-Inspired Feature Restoration DNN for Partially Occluded Image Recognition
• Robustness of Object Recognition under Extreme Occlusion in Humans and Computational Models
• CNNPruner: Pruning Convolutional Neural Networks with Visual Analytics
• Are Deep Learning Models Robust to Partial Object Occlusion in Visual Recognition Tasks?