Enhancing CNNs robustness to occlusions with bioinspired filters for border completion
arXiv:2504.17619v11 citationsh-index: 4GSI
Originality Incremental advance
AI Analysis
This addresses robustness issues in CNNs for image recognition tasks, but it is incremental as it builds on existing methods with a specific bioinspired modification.
The paper tackled the problem of improving CNNs' robustness to occlusions by using bioinspired filters based on visual cortex mechanisms for border completion, resulting in consistent accuracy improvements on occluded MNIST images with a modified LeNet 5.
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.