|
It is possible to get better accuracy when coregistering images, but it all depends on the quality and accuracy of your GCPs. If you have done the best job you can picking GCPs, then there probably is not a way to get greater accuracy.
That said, you might want to consider different warp methods. If you have lots of GCPs, well spread throughout the image, then the Triangulation method can give more accurate results than polynomial warps. This is especially true for airborne sensors, like Hymap, where the nature of the distortion can vary a lot through the image, as the aircraft attitude changes during flight.
The algorithms that you mention are resampling algorithms, which is different than warping methods. Resampling method refers to the method used to decide exactly which pixel values should go in the output pixels, whereas the warp method refers to the algorithm used to decide how to spatially warp the image (i.e., where the output pixels go). In terms of resampling, nearest neighbor exactly preserves the original pixel values, but can result in a more choppy appearance. Bilinear and cubic convolution provide a more smooth appearance, but that is because the original pixel values are resampled. For a change detection analysis, you may not want that.
If you are only calculating vegetation indices, then the mismatch between wavelengths of your two images is probably not going to have a huge affect on your results. Your plan to choose more or less equivalent bands for the calculation of the indices seems good to me.
Peg
|