You have to first think of the strategy you are going to use for combining different classifiers. One way to go is to assign a pixel to most occurring class accross different classifiers ! lets say in your case with svm and NN, you assign a pixel to a class represented by both svm and ANN if they are same otherwise you can look at the rule set, assign one with highest probability if the class from svm and NN is different. In both case, simple band math would not be enough but you have to write your custom function in IDL. Hope this helps.