Quantum computation via neural networks applied to image processing and pattern recognition

Abstract

This thesis explores moving information processing by means of quantum computation technology via neural networks. A new quantum computation algorithm achieves a double-accurate outcome on measuring optical flows in a video. A set of neural networks act as experimental tools that manipulate the applied data. Attempts have been made to calculate a pixel's location, velocity and grey scale value of moving images but the location and velocity could not be simultaneously measured precisely enough in accordance with both classical and quantum uncertainty principles. The error in measurement produced by quantum principles was found to be half that produced by a classical approach. In some circumstances the ratio of a pixel's coordinates and that of velocities could be determined using quantum eigenstate theory. The Hamiltonian of interaction of two NOT gates is most likely to represent the Gibbs potential distribution in calculating the posterior probability of an image. A quantum chain code algorithm was erected to describe the edges of image features. The FACEFLOW experimental system was built in order to classify the moving human faces. Three kinds of neural network models were finally presented.

Date of Award	2001
Original language	English