Fuzzy Context Sensitive Thresholding Guided Bidirectional Self Organizing Neural Network (BDSONN): A Gray Scale Object Extractor

Abstract

A three layer counter-propagating fully connected bidirectional self organizing neural network (BDSONN) architecture comprising an input layer, an intermediate layer and an output layer of fully connected neurons, driven by the fuzzy membership values of the image scene and efficient for gray scale object extraction from a multiscale image scene is presented in this article. The neurons at each of the input and intermediate layers of the network are connected to the next layer neurons using a neighborhood based topology. The output layer neurons are connected to the intermediate layer neurons forming a counter-propagating structure. The proposed architecture, guided by a multilevel sigmoidal activation function, uses fuzzy image context based thresholding information for self-organizing multiscale input information into different extracted scales of gray by means of counter-propagation of network states. Application of the proposed architecture for gray scale object extraction is demonstrated using real life gray scale images.