A Noise-Tolerant Framework for Aerial Images Classification Based on Gabor Energy Feature

This paper presents a novel framework to categorize aerial images into two classes: images with manmade structures and natural scene images. A novel noise-tolerant three-stage feature extraction framework is presented here which includes extracting edges from the input gray image, applying Gabor filter to compute Gabor energy feature and wavelet decomposition technique to extract the feature vector of computationally affordable size. A probabilistic neural network (PNN) is employed to classify the aerial images. From the database of 112 images (58 are natural scenes and 54 are images with manmade structures), total of 30 images, 15 from each class are used for training phase. For testing the algorithm, 82 images (39 manmade class and 43 of natural class) are used. The proposed method gives 94.87% correct classification for images with manmade structure and 97.67% for natural scene images.