Fast fingerprint civil identification system based on algorithms derived from deep learning neural networks. The system generates fingerprint templates as a vector 512 bytes long. Comparison of templates is performed by measuring the distance between vectors in L2 metric. 1:1, 1:N, M:N (batch mode) comparison modes are supported.
Identification system for criminal police works with the entire set of fingerprint information: contact and rolling fingerprints; latent prints and palm prints. A neural network-based identification algorithm is used for most queries. For latent prints of poor quality with a low number of features (minutiae), as well as for palms, combined type algorithms are used: a first pass with a neural algorithm and a second pass with a certified Sonda algorithm. When inputting information from paper fingerprints, neural algorithms are used to segment the prints, providing high quality that does not require operator correction.
Face recognition algorithm is based on neural networks and deep learning technologies for face detection and pattern extraction. The networks are trained on multinational datasets and have no significant bias towards any nation. Key features: face capture; biometric pattern generation; pattern matching 1:1, 1:N, M:N (batch mode). Templates are formed by the neural network in the form of a vector with a length of 1024 bytes.
Iris recognition algorithm is based on neural network and deep learning technologies. The algorithm includes an algorithm for iris detection, pattern formation and pattern matching 1:1, 1:N. The algorithm works with standard scanned iris images: near-infrared, preferably 640x480. The technology has greater resistance to artefacts such as eyelids, eyelashes, etc. compared to 'traditional' algorithms.