Well, hhEb09'1, there are just two forms of compression and decompression: lossless, and lossy. With the lossless variety, one can cycle a photo through a billion times and it's bit-for-bit identical to the original. With lossy, after a billion cycles, it's, well, it's visual good-for-nothing goo.
In bitmap reconstruction, pattern matching and transition algorithms are used, along with smoothing filters, in a higher resolution than the original image, in order to theoretically estimate the missing information in the image and fill it in accordingly.
With simple patterns, it does a pretty good job. With complex image patterns, particularly the factal-like nature of edges found in nature, where the same pattern is found throughout a very wide range of scales, from the jagged coastline of a country to the jagged edge of a single rock, it does a very poor job.
Regardless, when dealing with priceless images of other worlds, it should be a felony to compress them in a lossy mode for any purpose other than to use as thumbnails (but the underlying image should be made available), or printed pictures in a magazine.
Unfortunately, it's lossly compressed pictures on various websites which get so jacked up that people can't tell the difference between
cornfields in Iowa and compression artifacts.