Neural network for acquisition and discrimination of object images

This paper describes an implementation of a neural network that performs rapid acquisition and discrimination of object images. The network can also identify images of the original object when the images are oriented differently than the original object. The neural network consists of several layers of neurons that are used in acquisition and discrimination of images. The input layer detects simple features of the object image such as elementary geometric shapes of lines, triangles, circles, and so on. The higher layers of neurons are involved in acquisition by combining simple features into more complex features like the shape of a target, irregular surfaces, and others. The network has a built-in learning capability and can reconstruct the whole image from partial data. It is implemented on a Cray X-MP/24 and on a Lisp workstation.