In this work we model the problem of identifying how close structurally two proteins are as a problem of measuring the similarity between color images that represent their contact maps, where the chromatic information encodes the chemical nature of the contacts. We study two conceptually distinct methods to measure the similarity between such contact maps: a content-based image retrieval one and another based on image registration. In experiments with contact maps constructed from the PDB, the image registration approach was able to identify with 100% precision 8 instances of a protein class mixed with 28 proteins of other classes. The content-based image retrieval approach had an accuracy only a little worse than that.