Contourlet-based multiband image fusion for improving classification accuracy in IRS LISS III images

Abstract

Unsupervised classification plays a vital role in overseeing the transformations on the earth surface. Unsupervised classification has an indispensable role in an immense range of applications such as remote sensing, motion detection, environmental monitoring, medical diagnosis, damage assessment, agricultural surveys, and surveillance. In this paper, a novel method for unsupervised classification in multitemporal optical images based on image fusion and Gaussian RBF kernel K-means clustering is proposed. Here, the image is generated by performing contourlet-based multiband image fusion on the red, green, and near- IR images. On the finest image generated by collecting the information from three bands, Gaussian RBF kernel K-means clustering is performed. In Gaussian RBF kernel K-means, nonlinear clustering is performed, as a result the false alarm rate is reduced and accuracy of the clustering process is enhanced. The aggregation of image fusion and RBF kernel K-means clustering is seen to be more effective in detecting the changes than its preexistences.