Figure 1: Image corruption may well be the end of us all!
Take an image. Split it up in a bunch of quadratic sub-images. Let each of them store low resolution copies of it's Von Neuman neighborhood's contents. Give the blocks checksums, and compress them individually. If a block fails to pass the checksum, you can now reconstruct it using interpolation from it's neighbor's low-res copies. It wouldn't be picture perfect, but you'd be able to deduce the contents of the image even at pretty bad corruption (information recovery at 75% data loss at a best case scenario).
Do the same with video, while you're at it. Seriously, give me a video codec that does this, and you will make the world a better place.
Better yet, add redundancy to text. Add a plaintext Huffman coded copy of the textual contents of a document to it's metadata.