Improving the accuracy of neural network methods of verification of persons by spatial-weighted normalization of brightness image

In this article, we propose a method of spatially weighted brightness normalization for facial grayscale images which retains more information during the normalization process. An experimental study is being conducted of the effect of various brightness normalization options on the accuracy of a fixed neural network classifier in the verification problem. It is experimentally shown that the proposed brightness normalization can improve the accuracy of facial images verification in complex lighting conditions and compensate for the samples that were not present in the training data.