NEURAL NETWORK-BASED METHOD AND DEVICE FOR FACE FEATURE EXTRACTION AND MODELING, AND FACE RECOGNITION

摘要

Provided in the present invention are a neural network-based method and device for face feature extraction and modeling, and face recognition. The method comprises: acquiring face image triplets from a training set of a predetermined application scenario; acquiring a trained face recognition neural network, and determining, according to the face recognition neural network, a triplet deep neural network; employing the face image triplets as an input of the triplet deep neural network, and determining a loss function value; training, according to the loss function value and training parameters, the triplet deep neural network with the training set; and performing, by means of a test set of the predetermined application scenario, a face recognition test on the triplet deep neural network, determining, according to the test result, test accuracy, and determining, according to the test accuracy and the triplet deep neural network, a target face feature extraction model. The method and device of the present invention provide an advantageous effect of superior face recognition accuracy when a face feature model obtained via modeling is applied to a specific face recognition application scenario. Also provided in the present invention are a face recognition method and device.