Cyclic Generative Neural Networks for Improved Face Recognition in Nonstandard Domains
- Authors: Grinchuk O.V.1, Tsurkov V.I.2
-
Affiliations:
- Moscow Institute of Physics and Technology
- Dorodnicyn Computing Center, Federal Research Center Computer Science and Control
- Issue: Vol 57, No 4 (2018)
- Pages: 620-625
- Section: Pattern Recognition and Image Processing
- URL: https://journals.rcsi.science/1064-2307/article/view/220174
- DOI: https://doi.org/10.1134/S1064230718040093
- ID: 220174
Cite item
Abstract
A system of methods for improving the quality of face recognition from infrared images is described. For testing the recognition algorithm in a multidomain environment, a database of ordinary and infrared face images is collected. An algorithm based on cyclic generative neural networks is developed. This algorithm makes it possible to transform images from the color domain into the infrared domain, which significantly increases the size of the training sample. It is shown that fine-tuning the recognition algorithm using the generated infrared images improves the recognition result on the test sample.
About the authors
O. V. Grinchuk
Moscow Institute of Physics and Technology
Author for correspondence.
Email: oleg.grinchuk@phystech.edu
Russian Federation, Dolgoprudnyi, Moscow oblast, 141700
V. I. Tsurkov
Dorodnicyn Computing Center, Federal Research Center Computer Science and Control
Email: oleg.grinchuk@phystech.edu
Russian Federation, Moscow, 119333