Classifying synthesized optical codes using polarimetric information and machine learning algorithms for optical security applications

Resumen

Given the increasing use of the internet and the transfer of information in this era, it is crucial to focus on encryption and data security. According to the technological advances in optics and photonics and their multiple applications, many researchers have been urging to apply optics to encrypt and authenticate information in the last decades. In other words, optical waveforms involve many complex degrees of freedom, such as polarization, amplitude, phase, large bandwidth, quantum properties of photons, and multiplexing that can be combined in many ways to produce high-security information systems. In this thesis, we have been investigating different photonics techniques appropriate for optical security applications. This interdisciplinary investigation includes photonics techniques such as digital holography, beam shaping, Fourier optics, polarization optics, diffractive imaging system, and interferometry. Besides, our applied approaches demanded extensive research in computational methods such as pattern classification by means of machine learning algorithms, computer simulation, fringe analysis, statistical analysis, and binary encoding. However, despite the defined thesis title, our achievement has not been limited to optical security. Classifying synthesized (unique) optical codes can be mainly split into two categories. The first one is an approach for obtaining unique optical codes. The second one is a method or technique for classifying and distinguishing synthesized optical codes. Regarding the first category, in this thesis, on the one hand, we propose a method to obtain unique optical codes (polarimetric signature codes) from illuminating 3D printed samples by linearly polarized beams. Also, the ability of 3D printed samples to be considered as Physical Unclonable Functions based on polarimetric information is discovered in this thesis. Hence, we consider 3D printed samples as physical keys able to produce unique polarized optical codes. On the other hand, we obtain unique polarized optical codes by synthesizing a laser beam at the entrance pupil of a highly focusing system. Accordingly, we developed a binary approach for encoding character codes into holographic cells appropriate for transferring information in free space. Regarding the second category, on one side, we classify the polarimetric signature codes obtained by a physical key (3D printed sample) by means of the Support Vector Machine classifier using feature vectors extracted from statistical analysis on speckle patterns. On the other side, we introduce polarimetric mapping images as multidimensional arrays to be inputs of a convolutional neural network model for the autodetection of character codes obscured in the longitudinal component of a highly focused electromagnetic field. This approach might be considered as an alternative method, which eliminates the necessity of phase retrieval algorithms in particular cases.