Chapter 5 - An efficient pure color image denoising using quantum parallel bidirectional self-organizing neural network architecture

Abstract

This chapter presents a self-supervised learning network in a quantum environment, named a “quantum parallel bidirectional self-organizing neural network (QPBDSONN) architecture” appropriate for pure color image denoising. The suggested QPBDSONN architecture mimics the classical parallel bidirectional self-organizing neural network architecture by embedding quantum computation which invokes the fundamental concepts and principles of quantum mechanics. The QPBDSONN architecture comprises three quantum bidirectional self-organizing neural networks (QBDSONNs) in the input layer to process three distinct color components (red, green, blue) separately used in the color noisy images. The principle of the operation of this QPBDSONN architecture is posed as a cluster of three parallel QBDSONNs after isolation of these three distinct basic color components from the pure noisy color images in the initial phase. Afterward, the basic color components obtained from the input noisy color images are fed simultaneously thorough three parallel different basic color component QBDSONNs for subsequent processing. Each of the three constituent QBDSONNs in the proposed network architecture comprises input, hidden or intermediate, and output layers of neurons. Each neuron of these three layers is intraconnected by means of an eight-connected neighborhood topology designated as qubits. The current parallel network architecture does not rely on a quantum back-propagation algorithm to adjust interconnection weights (described by rotation gates). Instead it uses qubit layers of neurons between the hidden layer and the output layer in a counterpropagation sense. Finally, quantum measurement is done with the aim of diminishing superposition of qubits in the output layer followed by a fusion operation to merge processed color image components in the sink layer and produce the true color output image.