Texture fusion and classification based on flexible discriminant analysis

We apply texture fusion to combine texture features computed using different texture models for classification purposes. Texture features are computed using four different models. We compare the performance of flexible discriminant analysis based on multivariate regression splines and generalized additive models to well-known classifiers like traditional discriminant analysis and neural nets. Two main conclusions can be drawn from this study: 1) texture fusion by combining features computed using different texture models improves the classification accuracy significantly compared to using a single texture model; and 2) flexible discriminant analysis and classification trees can be valuable tools in classifying non-Gaussian features.