Multi-scale object extraction using a self organizing neural network with a multi-level beta activation function

A multi-level beta activation function is proposed in this article for the extraction of multi-scale objects from an image scene. The beta function with equal class responses is generated using the number of classes in the image scene. A three layer self-organizing neural network comprising an input layer, a hidden layer and an output layer, is then used to extract multi-scale objects using this activation function. The system error is calculated based on some fuzzy measures in the output status of the neurons in the output layer of the network. An application of the proposed activation function for the extraction of objects using a three layer self-organizing neural network is demonstrated with two images. The standard correlation factor and the discrepancy index (DI) between the extracted images and the original images are used as the figures of merit to evaluate the quality of the extracted images.