Noise tolerant classification of aerial images into manmade structures and natural-scene images based on statistical dispersion measures

Objective of this paper is to categorize aerial images into two classes: manmade structures and natural-scene images. A novel noise tolerant approach based on statistical dispersion measures is presented here. In this approach, three statistical dispersion measures namely standard deviation, mean absolute deviation and median absolute deviation are used as features. With these measures, a feature vector of size 3×1 is formed and applied to probabilistic neural network (PNN) for classification purpose. From the database of 112 images, 14 images (7 from each class) are used for training purpose. For testing, we have used remaining 98 images (47 images manmade class and 51 images of natural scene class). The proposed method gives 95.75% correct classification for images with manmade structure and 98.04% for natural scene images.