Performance evaluation of various resampling techniques on IRS imagery

Resampling is used to calculate pixel values when one raster grid is fitted to another. High-resolution remote sensing satellite images contain more information in the discrete samples and after resampling process, it is desirable that the reconstructed image should maintain the same sharpness as the original image. Although several techniques are available, but it is essential to determine the best one for maintaining the sharpness and the pixel break at higher magnification level for photographic and digital display of high resolution satellite images. To preserve image quality, the interpolating function used for the resampling should be an ideal low-pass filter. In order to determine the best interpolation function, different resampling functions, namely Nearest Neighbor (NN), Bilinear (BL), Cubic Convolution ( α = 0 ), High-resolution Cubic Spline with Edge Enhancement ( α = -1 ), High-resolution Cubic Spline ( α = -0.5 ), Cubic Spline ( α = -2 ), Cubic Spline ( α = -.3/4), Cubic B-spline, Catmull-Rom Cubic, Quadratic Interpolation and Approximating Quadratic B-Spline have been analyzed on different spatial resolution Indian Remote Sensing (IRS) satellite images (LISS-III, LISS-IV and CARTOSAT-1). Performance of the above resampling methods has been evaluated by Visual interpretation, Digital Number Percentage (DN %) Analysis as well as other parameters likes Entropy and Image Noise Index (INI), MSE and PSNR. Investigation results have shown that with the change in the image processing operation, spatial resolution and evaluation parameter, the performance of resampling method changes, thereby emphasizing the need to judiciously select the resampling method.