Active unsupervised texture segmentation on a diffusion based feature space

Abstract: We propose a novel and efficient approach for active unsurpervised texture segmentation. First, we show how we can extract a small set of good features for texture segmentation based on the structure tensor and nonlinear diffusion. Then, we propose a variational framework that allows to incorporate these features in a level set based unsupervised segmentation process that adaptively takes into account their estimated statistical information inside and outside the region to segment. Unlike features obtained by Gabor filters, our approach naturally leads to a significantly reduced number of feature channels. Thus, the supervised part of a texture segmentation algorithm, where the choice of good feature channels has to be learned in advance, can be omitted, and we get an efficient solution for unsupervised texture segmentation. The actual segmentation process based on the new features is an active and adaptative contour model that estimates dynamically probability density functions inside and outside a region and produces very convincing results. It is implemented using a fast level set based active contour technique and has been tested on various real textured images. The performance of the approach is favorably compared to recent studies.