Combining Gaussian Markov random fields with the discrete-wavelet transform for endoscopic image classification
Michael R. Hafner, Alfred Gangl, Michael Liedlgruber, Andreas Uhl, Andreas Vécsei, Friedrich Wrba · 2009
In this work we present a method for automated classification of endoscopic images according to the pit pattern classification scheme. Images taken during colonoscopy are transformed to the wavelet domain using the pyramidal discrete wavelet transform. Then, Gaussian Markov random fields are used to extract features from the resulting wavelet coefficients. Finally, these features are used for a classification using the k-NN classifier and the Bayes classifier. To enhance the classification results feature subset selection is used to reduce the dimensionality of the features. Apart from that, directional neighborhoods for the Markov random fields are introduced. These are exploiting the orientation of the details within the wavelet detail subbands with the goal of further improving the classification performance. The experimental results show that an automated classification using the presented method is feasible.