We propose a new preprocessing stage destined to convert an ultraspectral image into a speech-like signal that exposes redundancy in such a way that the compression stage removes it more efficiently. In general, preprocessing involves a sequence of reversible operations that in spite of not providing compression they redistribute the input data to better present redundancy in preparation for the following stage. Three techniques are hereby integrated to accomplish this: band normalization, band ordering and image scanning. Band normalization reduces the dynamic range of spectral bands while preserving their integrity, band ordering rearranges them to maximize the interband correlation and image scanning applies specific spatial block scanning techniques to maximize the intraband correlation. These techniques induce intraband periodicity of highly correlated samples transforming the output into a predictable signal that, by means of linear filtering, can be compressed the same way a speech signal is.