Texture classification using a two-stage neural network approach

In this article, we present a two stage neural network structure which combines the self-organizing map (SOM) and the multilayer perceptron (MLP) for the problem of texture classification. The texture features are extracted using a multichannel approach. These channels comprise of a set of Gabor filters having different sizes, orientations and frequencies to constitute N-dimensional feature vectors. The SOM acts as a clustering mechanism to map these N-dimensional feature vectors onto a 2D space. This in turn forms the feature space to feed into MLP for training and subsequent classification. It is shown that this mechanism increases the inter-class separation and decreases the intra-class distance in the feature space, hence reduces the classification complexity. Also, the reduction in the dimensionality of the feature space results in reduction of the learning time of the MLP.