Enhancing CNNs robustness to occlusions with bioinspired filters for border completion

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. Keywords: Convolutional Neural Networks Visual Cortex 1 Introduction Visual perception has evolved as a fundamental tool for living organisms to extract information from their surroundings and adapt their behavior. However, encoding visual information presents several challenges. One major issue is occlusion, i.e. an object's outline is partially hidden by an obstacle.