Face search and recognition in historical photos

The work is devoted to the development of a system for recognizing and verifying faces in historical (pre-revolutionary and pre-war) photographs. The paper reviews the existing methods of face recognition. The Siamese neural network model is considered as the main architecture for system development. The software implementation of the Siamese neural network model in the Python programming language using the TensorFlow and Keras frameworks is described. Three variants of training the Siamese neural network model are considered. The data preparation procedure was carried out. A marked-up set of training examples was collected from the data set. The model was tested and improved using the method of additional training of neural networks. The maximum accuracy that was achieved during the execution of the work is 94%. The methods of recognizing graphic images can be used to solve genealogical problems, such as searching for family ties, searching for information about living and deceased relatives, etc. For residents of Russia and the CIS countries, such tasks are relevant due to the historical events of the last hundred years. The results of numerical experiments on the study and comparison of the proposed methods are also presented.