Goal-Directed Textured-Image Segmentation

Abstract

The SLICE textured-image segmentation system identifies image regions that differ in gray-level distribution, color, spatial texture, or other local property. It has been developed for the analysis of aerial imagery, although it can be used for any domain in which homogeneous image regions must be found prior to interpretation or enhancement. This report concentrates on textured-image segmentation using local texture-energy measures and user-delimited training regions. The SLICE algorithm combines knowledge of target textures or signatures with knowledge of background textures by using histogram-similarity transforms. Regions of high similarity to a target texture and of low similarity to any negative examples are identified and then mapped back to the original image. This use of texture-similarity transforms during the segmentation process improves segmenter performance and focuses segmentation activity on material types of greatest interest. The system can also be used for goal-independent texture segmentation by omitting the similarity-transform computations, and its hierarchical, recursive segmentation strategy integrates very well with other image-analysis techniques.