The statement of basic material. As it is known-the human brains vision seems to be very easy functioning. It does not take any difficulty to tell apart a dog and a cat, read a word or recognize a human face. But in difference from human-these tasks are really difficult problems for solving with a computer. Recognition process only seems easy because human brain is really good at perception and as a result in understanding images. During last years, machine learning has made great progress in solving these difficult problems. In particular, model called-deep convolutional neural network can result in reasonable performance on solving difficult visual recognition tasks which are matching or exceeding human performance in some aspects. Conclusions. This paper introduced a new method of obtaining data for training security systems from social media and human interaction for future use in authentication process in various informational systems. There are several advantages of proposed system which can be described. First of all, one should mention that using of TensorFlow can be adaptive, powerful, and flexible. Moreover, training time is acceptable in comparison with other frameworks and much more faster if one uses distributed TensorFlow.