Multi-angle face recognition method based on deep learning and space conversion network

Abstract

The invention discloses a multi-angle face recognition method based on deep learning and a space conversion network, and the method comprises the steps: firstly building a convolutional neural networkmodel, improving a loss function of the convolutional neural network model, and carrying out the training of the model through a pre-obtained image; secondly, performing face alignment based on a space conversion network: performing data acquisition on a pre-acquired picture by utilizing a transformation matrix, and generating a corresponding face with the same angle size as the template; and finally, detecting the human face based on the YOLOV2. According to the method, generalized face feature representation is extracted through training on massive face data, the distinguishability betweenfeatures is highlighted, the face recognition accuracy is remarkably improved, and the method can achieve face recognition in a natural non-cooperative scene; and face alignment and a face recognitionnetwork can be unified to form an end-to-end learning system, so that the face alignment flexibility is greatly improved.

Description

technical field [0001] The invention belongs to the field of artificial intelligence face recognition, relates to the determination of human faces under multi-angle conditions, and specifically relates to a multi-angle face recognition method based on deep learning and a space conversion network. Background technique [0002] Face recognition can obtain the identity information of the corresponding person based on the face image. Because of its convenience, ease of use and universality, face recognition has been widely used in fields such as finance, criminal investigation and national defense. Due to the rapid development of deep learning, face recognition technology has become more and more popular, and it has received extensive attention from the academic community. With the improvement of computer performance and the increase of databases, the accuracy of face recognition has been continuously increased. At present, the highest accuracy rate is close to 100%, but this d...

AI Technical Summary

Problems solved by technology

There are currently two problems with this kind of face alignment. The first is that it is very dependent on the accuracy of feature points. Deviation or failure to detect feature points will greatly affect its effect.

Training a feature point detection network with excellent performance requires a large amount of feature point data, and the cost of collecting such a database is huge

The second is that the alignment of fixed geometric shapes will produce distortion of the geometric information of the face

In large-scale face recognition scenarios, due to face differences caused by uncertain factors such as illumination and pose, it is difficult to define a fixed set and template to adapt to all states

Images