Mechanism of Formation of Vision Based on Learning of Correlation between Sensation and Motion

In this paper, we aim to support the scheme that visual cognition is organized by natural neural networks based on experiences. In this scheme the networks learn correlation between temporal variation of visual stimulation and self-motion to organize visual cognition. We propose a neural network that detects temporal variations of visual stimulation and combine them with corresponding self-motion. The network does not use any prior knowledge about an image such as optical process of formation or pixels arranged in two dimensional plane. By experiments, we show that the proposed network can combine sensation and motion correctly and can recall pattern of one side by input pattern of the other side.