A Neural Network Model for Storing and Retrieving 2D Images of Rotated 3D Object Using Principal Components

摘要

A neural network model for storing and retrieving two-dimensional (2D) images of rotated three-dimensional (3D) object is presented, where principal components of the 2D images are used for data compression. The network is for examining how we can store and retrieve huge amount of images, and how we can construct a neural model of the mental rotation which is supposed to play important roles in human perception. Numerical experiments with the present model show that we can store and retrieve a huge amount of 2D images due to the eigenspace method utilizing principal components, while achieving the calculation time for retrieving rotated images being proportional to the rotation angle.