We consider the problem of compressing an image into packets sets that are mutually refinable. This problem is applicable to distributed storage of data: each user owns a different compressed version of the same original picture. Each version by itself is a coarse approximation of the original picture. Then, people can decide to share their versions. The more versions people contribute, the better the resulting approximation becomes, independently of which people are actually involved: only the number of users that collaborate matters. We will address this problem by using multiple description techniques combined with sub-band coders.