Aerial images often show regions that may offer high local color dynamics but nevertheless appear globally regular in a stochastical way so that humans easily recognize them as associated structures like a forest, a field, a pathway, or a waterbody. This work intends to algorithmically recognize and segment such regions. To do this, first, texture models are generated for different image regions, their structural similarities are assessed and, then, sufficiently different texture models are determined. Afterwards, each pixel in the image is assigned to one of those texture models with a structural metric and is then colored according to the assignment. Three different methods of texture modelling are examined and compared in view of the goal to combine as many regions as possible that are seen to be similar by a human observer (e. g., treetops, fields) and to insert segment margins where there are apparent transitions (e. g., boundary of a forest, vegetation changes).