This letter reports the production and optical polarimetric verification of codes based on thin-film technology for security applications. Because thin-film structures display distinctive polarization signatures, this data is used to authenticate the message encoded. Samples are analyzed using an imaging ellipsometer able to measure the 16 components of the Mueller matrix. As a result, the behavior of the thin-film under polarized light becomes completely characterized. This information is utilized to distinguish among true and false codes by means of correlation. Without the imaging optics the components of the Mueller matrix become noise-like distributions and, consequently, the message encoded is no longer available. Then, a set of Stokes vectors are generated numerically for any polarization state of the illuminating beam and thus, machine learning techniques can be used to perform classification. We show that successful authentication is possible using the knearest neighbors algorithm in thin-films codes that have been anisotropically phase-encoded with pseudorandom phase code.