Inverse FFT with odd dimensions gives unexpected results

If you calculate an FFT image in ENVI, and then perform an inverse FFT transform on that FFT image, you may see unexpected, wide, dark and light striping in the resulting inverse FFT image. This happens only if the input image has an odd number of samples and/or lines. This behavior is due to IDL using a different storage scheme for the FFT complex values in even- versus odd-dimensioned FFT results.

You can avoid the striping in inverse FFT images by simply making sure that the image used in a forward FFT transform has an even number of samples and lines. For example, you can choose an even-dimensioned spatial subset of the image as it is being selected in the file selection dialog for the forward FFT transform.

It has been suggested that ENVI should automatically pad or trim images with odd dimensions to make the dimensions even before an FFT image is calculated. The developers have considered this option and have decided not to change the default behavior of ENVI's forward FFT routine because leaving the spatial subsetting up to the user allows for the maximum flexibility of this tool.