Face Recognition Methods in Conditions of Incompleteness of the Source Data

The analysis of face recognition methods in conditions of incompleteness of the source data is presented. Classical approaches (PCA, LDA, SIFT) and their modifications for working with noisy images are considered, including local matching, reconstruction, and detection-rejection algorithms. Solutions to the «single sample» problem, such as SVD/DMMA-based virtual image generation and decomposition are considered. The following deep learning methods are considered: ResNet, GAN and CapsNet. The results are focused on the development of sustainable biometrics and video analytics systems that work with artifacts and small samples count.