Performance Comparison of Multiwavelet and Multicontourlet Frame Based Features for Improving Classification Accuracy in Remote Sensing Images

Scalar wavelet based contourlet frame based features are used for improving the classification of remote sensing images. Multiwavelet an extension to scalar wavelets provides higher degree of freedom, which possess two or more scaling function and wavelet function. Unlike scalar wavelets, which has single scaling and wavelet function. Multiwavelet satisfies several mathematical properties simultaneously such as orthogonality, compact support, linear phase symmetry and higher order approximation. The multiwavelets considered here are Geronimo-Hardin-Massopust (GHM) and Chui Lian (CL). In this paper the performance of GHM and CL multiwavelet is compared. Finally CL based multicontourlet frame based features are used for improving the classification accuracy of remote sensing images as it has directional filter banks. Principal component analysis based feature reduction is performed and Gaussian Kernel Fuzzy C means classifiers are used to improve the classification accuracy. The experimental results shows that the CL based multicontourlet overall accuracy is improved to 5.3% (for LISS-IV(i)), 2.09% (for LISS IV(ii)) 4.17% (for LISS IV(iii)) and 4.2% (for LISS IV-(iv)) the kappa coefficient is improved to 0.04 (for LISS IV-(i)), 0.029 (for LISS IV-(ii)), 0.031 (for LISS IV-(iii)) and 0.05 (for LISS IV-(iv)) compared to Wavelet based Contourlet transform.