WAVELET BASED IMAGE FUSION FOR CLASSIFICATION PROBLEMS

Abstract

With the improvement in spatial and spectral resolutions of remote sensing images, the concept of image fusion is gaining momentum to merge the images for further analysis so that spatial and spectral characteristics of the images can be exploited. The fused images are expected to provide enhanced information to be extracted for various applications. A number of image fusion techniques can be utilized to perform image fusion. However, most of the image fusion techniques fail to preserve the spectral characteristics. The aim of this dissertation is to utilize an advanced mathematical tool namely wavelet transform for fusion of images at different resolution. A two-dimensional Discrete Wavelet Transformation (DWT) has been implemented to perform image fusion. In the fusion process, smooth or approximate part has been taken from low resolution image and detail part, which contains maximum information, has been taken by comparing the wavelet coefficients of either images. Wavelet transformation based image fusion was compared with conventional fusion techniques namely Brovey Transformation and IHS (Intensity, Hue, and Saturation) transformation on two different data sets. The evaluation of results on two different data sets carried out on the basis of statistical analysis clearly suggests that wavelet based image fusion, not only increases the spatial resolution but also preserves the spectral resolution of original low resolution multispectral image. Wavelet based image fusion performs the best. As an example, to study the effect of image fusion on information extraction, the fused images obtained from different techniques were classified using maximum likelihood classification. The results clearly indicate that the accuracy of classification performed on wavelet based fused images was the highest among all the fused images.