Neural Network Combines with a Rotational Invariant Feature Set in Texture Classification

Abstract

AbstractIn this paper, a new combine method for texture description is introduced, which has successfully applied to pollen surface image discrimination in combination with a multilayer perceptron (MLP) neural network. Through wavelet decomposition and a details reconstruction process, a set of rotation invariant statistic features was formed to characterize textures. In this method, the joint probability of a grey level image and its corresponding details image was calculated. By using MLP as classifier, in experiments with sixteen types of airborne pollen grains, more than 95 percent pollen images were correctly classified.KeywordsTexture FeatureTexture ClassificationWavelet DecompositionDetail ImageAirborne PollenThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.