Combined 2D/3D approach for improving the accuracy of face recognition systems using deep learning

Face recognition systems using color images are widely known. However, their main problem is the instability to various lighting conditions, emotional and facial expressions, overlaps and rotation angles. The article proposed a new approach that combines the processing of two-dimensional data on a color image and three-dimensional data on a point cloud or depth map. The main attention was paid to the architecture of neural networks, the quality and accuracy of recognition. Some combined methods for the face recognition using machine learning and deep learning have been proposed. A comparative analysis of the results of experiments in terms of the recognition accuracy on open face databases was carried out. The best combined method was chosen. The main goal is to build a reliable, accurate combined biometric face recognition system that is resistant to complex external factors, such as facial expressions, scale changes, lighting, partial overlap with foreign objects, large rotation angles.