A comparative study of various pixel-based image fusion techniques as applied to an urban environment

Abstract

In this paper, the use of high-resolution images for identification of urban features through pixel-based image fusion techniques is discussed. Fusion techniques are used to merge high spatial resolution panchromatic (PAN) image with low spatial resolution multispectral (MS) image to enhance the visual quality/appearance of some of the urban features present in the image. In this paper nine pixel-based fusion techniques, viz., Principal Component Analysis (PCA), multiplicative, Brovey transformation, wavelet analysis, subtractive, HPF, modified IHS, Ehlers, and hyperspherical color space, are discussed for the fusion of PAN and MS IKONOS images. The fused images are interpreted on the basis of visual comparison, correlation coefficients and histogram statistics. All the nine fusion techniques improved the resolution and the visual appearance of the original MS image. These are comparable to the resolution of the original PAN image. All the three resampling methods (nearest neighborhood, bilinear interpolation and the cubic convolution) do not have any significant effect on the final visual appearance of the fused images. All fusion techniques result in a change of the statistical parameters of the original images. The multiplicative technique results in major changes of the statistical parameters than the other techniques. After doing visual comparison and analyzing the statistical parameters, the Brovey and the PCA techniques seem to gather more information of urban features in fused PAN and MS IKONOS images.