I would like to explore the possibility of using texture in an ecognition algorithm for classifying vegetation. Though, it seems to be based on sub-objects I am not sure how to actually create the sub-objects to get the specific texture value. ecognition "help" does a decent job of explaining what the texture values are, but does not really tell you how to get those numbers or how to add them to an algorithm.
First you "Edit NN feature space" which lets you choose by which features your objects will be given membership to classes by double clicking, then you "Apply NN standard"; What I prefer is, after having segmented the image, to export results and add up several features (pixel related, class related, texture, brightness) to the "layer values", and then I open this dbf of the shapefile in R to use RWeka, caret or randomForest machine learning algorithms that give you a decision tree to perform the classification based on your training.
Creating sub-object can be done by selecting "create below" when running a segmentation with a smaller scale parameter than you original level.
There are also texture indices that are not based on sub-object. The most straightforward is the variance of spectral values (on a noise free image, this is a quite robust and fast to compute texture index depites its "simplicity").
For more advanced indices, you can use the texture after Haralick. Those are more complex (more computation time), but very usefull if you know what you are looking for (e.g. a directional texture due to sun shadows)