Neural network methods for image segmentation

Abstract

Segmentation is the fundamental step in many image processing algorithms. In this paper, a simplified neuro-computing structure in feed forward form for use in segmentation of images in complex background is proposed. The work considers the formation and training a neuro-computing structure in which the pixel values of various region of the image are used as target. The method does not require any feature extraction, labeling of objects, region growing or splitting methods to configure and train a neuro-computing structure, which for the work is a Multi Layer Perceptron (MLP) trained with (error) Back Propagation learning. The neuro-computing structure is trained with different training functions. The network is also trained with single, double and triple hidden layers. The training is also done with Generalized Regression Neural Network for different values of spread function. Then the mean square error between the output image and desired image and the time required for training has been calculated.