Techniques for extracting keys from digital images data are presented. In an embodiment, a method comprises receiving a particular image; using a computer, determining a plurality of characteristics specific to contents of the particular image; using the computer, generating, based at least in part on the plurality of characteristics, a plurality of coefficients; generating a plurality of values by applying a particular function to the plurality of coefficients; selecting, from the plurality of values, a subset of values, each value in the subset satisfying a particular threshold requirement; based, at least in part on the subset of values, generating one or more keys, each of which uniquely encodes data included in the subset of values; generating and storing metadata that associates each of the one or more keys with the particular image.