Hyperspectral Data Classification Using Contourlet Transform

For many years, Wavelet Transform was the major feature extraction method for image classification. Since the means of feature extraction directly affects the performance of the classification, it is vital to choose an appropriate method for different types of images. Although the Wavelet Transform provides a common method for this, recent techniques are being studied that can capture further image properties hidden from the Wavelet Transform. One of the alternatives to the Wavelet Transform is the Contourlet Transform. The Contourlet Transform performs better on detecting the smoothness along the edges which is encountered on the boundaries of smooth regions of the image. Furthermore, it has more directionality than the Wavelet counterpart that can improve classification performance significantly when the image has classes with many different directions. This work applies the Contourlet Transform and its variations to the classification of the AVIRIS image data taken from Indiana's Indian Pine test site in June 1992. The data is hyperspectral in nature and hence this work additionally provides new benchmark results on hyperspectral data classification.