Remote sensing panchromatic images classification using moment features and decision fusion

Panchromatic remote sensing images have useful information about textural classification in land-use and land-cover applications. Various methods model texture and extract features for classification tasks. In supervised classification, all of the feature extraction methods try to increase the accuracy of classification and simultaneously decrease the computationally load. At the present work, we use the moments to extract texture information. The moments are categorized into different kernel functions. We use three conventional moments in pattern recognition such as Geometric, Legendre and Zernike moments and we obtain strong performance from almost unknown Chebyshev moment in feature extraction. The challenging part of pixel-wised image classification is classifying pixels in the near of different classes margin. To overcome the problem, we use majority voting as decision fusion rule. The well-known support vector machine (SVM) is used for supervised classification. We compare our proposed method with gray-level co-occurrence matrix (GLCM), Gabor filters and attribute profiles (APs). Different criteria such as average accuracy, overall accuracy, κ statistic and computation time are used for assessment of classification performance.