Texture extraction and segmentation via statistical geometric features

The statistical geometric features (SGF) are a new approach to texture analysis combining statistics with geometrical attributes to give a powerful discriminatory ability. The original scheme considered the approach in principal and did not address factors important to its eventual application, namely its implementation and the segmentation of texture imagery. We show how the implementation factors can affect performance and how it can be used for segmentation. Using SGF, a new adaptively-positioned windowing strategy for segmentation delivers a performance which, in terms of speed and accuracy, gives a performance between the traditional tiled- and sliding-window approaches.