Color wavelet cross co-occurrence matrices for endoscopy image classification

In this work we propose a set of new wavelet-domain based color-texture features for the classification of zoom-endoscopy images in the field of medical imaging. We extend the concept of classic co-occurrence matrices to capture information between detail-subband pairs of different color channels. Our results show, that the proposed features outperform other wavelet-domain based color-texture features in terms of leave-one-out cross-validation accuracy.