Combining features for texture analysis

Abstract

In the present paper we consider building feature vectors for texture analysis by combining information provided by two techniques. The first feature extraction method (the Discrete Wavelet Transform) is applied to the entire image. By computing the Gini index for several subimages of a given texture, we choose one that maximizes this measure. For the selected subimage we apply the second technique (a Gabor filter) for feature extraction. When we combine the two vectors, the classification results are better than the one obtained using only one set of features. The classification was performed on the Brodatz album, using a naive Bayes classifier.