My idea is simple. For media that depends on lossy compression it should be possible to ”patch” the media file with a smaller file of same resolution but a lower quality. The backup file is smaller due to more lossy compression. For this to work on image files the format probably need to loose some dependency on the previous block. One error should not affect the rest of the image. On the blocks you then perform some checks. The idea is then that if you encounter a error in the original big image file then you can patch it with the block of pixels in the smaller backup file. The block in the backup file would be of lower quality but look similar. You probable would not notice that the image has been fixed.
For a .jpg photo of 3.3MB I saw that I would only need a 350k version. To keep it simple one could store both versions in the same file as an option. For those who don’t make backups I think this would make it easier to recover otherwise lost photos.