Robust segmentation of noisy images using a neural network model

摘要

In this paper we present a robust algorithm for segmentation of noisy images using a neural network. A neural network model is employed to represent a segmented image in such a way that the region type of each pixel simply corresponds to the neuron state. The image segmentation procedure consists of two phases: initial segmentation, and refining segmentation. The initial segmentation is implemented iteratively by using a dynamic evolution algorithm to minimize the energy function of the neural network. During the second phase, a learning procedure is described to determine the interconnection weights based on a multi-layer logic neural network, and the refining segmentation is achieved by a retrieval procedure. Several experimental results involving both synthetic and real images are reported.