Human peripheral blur is optimal for object recognition
This research addresses the evolutionary basis of human vision for object recognition, offering a novel perspective that could influence computational models in computer vision and neuroscience.
The study tested whether the human eye's blurry peripheral vision is optimal for object recognition by training deep neural networks on foveated images with varying blur profiles. They found that networks using a blur profile matching the human eye performed best, with categorization accuracy deteriorating only for steeper profiles, suggesting this sampling scheme optimizes recognition rather than being a constraint.
Our vision is sharpest at the center of our gaze and becomes progressively blurry into the periphery. It is widely believed that this high foveal resolution evolved at the expense of peripheral acuity. But what if this sampling scheme is actually optimal for object recognition? To test this hypothesis, we trained deep neural networks on 'foveated' images with high resolution near objects and increasingly sparse sampling into the periphery. Neural networks trained using a blur profile matching the human eye yielded the best performance compared to shallower and steeper blur profiles. Even in humans, categorization accuracy deteriorated only for steeper blur profiles. Thus, our blurry peripheral vision may have evolved to optimize object recognition rather than merely due to wiring constraints.