Texture segmentation by change detection in second and higher order statistics

In this study we consider unsupervised texture segmentation via non-parametric change detection of locally estimated image features. The change detection approach is straightforward and may be used with first, second, or higher order statistics. Adaptive smoothing is used to reduce the variance in local statistic estimates while preserving step behavior due to the presence of a region boundary. The approach is combined with Laws (1980) energy operators by detecting a change in the variance of the image resulting from application of a Laws mask to the original texture image