From Images to Perception: Emergence of Perceptual Properties by Reconstructing Images
This work addresses the fundamental question of how visual perception arises in humans, with implications for neuroscience and AI, though it is incremental in building on existing bio-inspired models.
The study tackled the problem of whether human visual perception emerges from image statistics by developing a bio-inspired neural network, PerceptNet, optimized for image reconstruction tasks. The results showed that the network's encoder layer achieved the highest correlation with human perceptual judgments on image distortion, with optimal alignment at moderate noise, blur, and sparsity levels.
A number of scientists suggested that human visual perception may emerge from image statistics, shaping efficient neural representations in early vision. In this work, a bio-inspired architecture that can accommodate several known facts in the retina-V1 cortex, the PerceptNet, has been end-to-end optimized for different tasks related to image reconstruction: autoencoding, denoising, deblurring, and sparsity regularization. Our results show that the encoder stage (V1-like layer) consistently exhibits the highest correlation with human perceptual judgments on image distortion despite not using perceptual information in the initialization or training. This alignment exhibits an optimum for moderate noise, blur and sparsity. These findings suggest that the visual system may be tuned to remove those particular levels of distortion with that level of sparsity and that biologically inspired models can learn perceptual metrics without human supervision.