If you're just doing it once, use ikegami's approach. If you're doing it many times (where the term 'many' is intentionally vague) you may benefit from maintaining two hashes; one forward-lookup and one reverse-lookup, if that makes sense.

Bear in mind, however, that in a hash, while the keys are guaranteed to be unique, there is no guarantee that the values are unique. So there's a possibility that 'yellow' as a value could occur several times, each under different unique keys. This must be dealt with somehow when doing reverse-lookups. One way is to just make sure you "don't do that" (don't create hash entries that have equal values). And it happens that maintaining a reverse-lookup hash is an excellent way of keeping track of two-way uniqueness. ...but this is just a tangent...