On the Importance of Feature Descriptors for the Characterisation of Texture

In this contribution is investigated the use of textural descriptors as descriptors for primitive textural information in medical images. A few approaches have been presented in the literature to the direction of the discrimination of texture in medical images compared with similar approaches used for texture recognition. In this paper we try to prove that texture exists in medical images and can be encoded using the proposed statistical based descriptors. We have selected four different methods for the estimation of such descriptors. These have been tested in various texture images and in endoscopic medical images, attempting this way to create the texture models possibly exist in the images. The texture spaces described by the corresponding vectors of the features are used as input to different multilayer perceptron type neural networks for the characterization of images from their texture content. An in depth experimental study has been conducted comparing textural feature extraction techniques on various images along with a novel discrete wavelet transform based methodology.