latin1 is not a subset of UTF-16!
I think you misunderstood text encoding and binary encoding. UTF-16 is used for character encoding, the base unit is a 16-bits integer (UTF-8 is using 8 bits integer).
A JPEG picture (JFIF) is binary encoded, and its data should never get converted via character encoding algorithm.
If you actually did so, you're out of luck, since using a character conversion algorithm on a binary stream depends on whatever "source" text charset that was used at the time.
You can probably try to convert that (UTF-16) binary data back to binary by guessing the initial source charset, using iconv.