Following the segmentation, each region is individually validated to determine if it represents valid terrain. Large area regions are automatically accepted as terrain using a spatial threshold of . Although the roof area of the Pentagon is equal to this threshold value, in practice, very few buildings approach this size so in most cases regions of this magnitude (or larger) would, in fact represent a ground surface. Regions with elevations significantly above their neighbors (especially if the neighbors are confirmed terrain regions) are also rejected. Given that there are numerous regions, large and small, within the scene, TEXAS attempts to build connections between regions that are not immediately adjacent, and then combines them into larger regions. A downward flood fill is performed after the classification step to add points in the high curvature area back in, this serves a dual purpose. First, it adds legitimate ground points that were previously excluded because of nearby high curvature, points immediately adjacent to buildings and ground points underneath vegetation are the two most common examples. Secondly, if a building was mistakenly left in the image, it ensures that the sides of the building are properly preserved for the next iteration. If this method were not used, the building sides would be eliminated and interpolated for the next iteration, smoothing them and reducing the ability of the next iteration to correctly identify the building. By preserving the sharp edges, more aggressive iterations can correctly identify a building and remove it entirely.