Identity identification method for autonomous learning

Abstract

The invention relates to a method for member identity identification under a similar home environment. Specific steps comprise: first, automatically screening face image quality, preliminarily completing home member register; second, under near distance and good light, identifying identity through faces; third, tracking a face, and obtaining information of height, body type, costume of a same person, and labeling samples according to the face information; fourth, according to information of tracking faces and heights, through a clustering method, obtaining multi-dimensional labeled samples; fifth, using the multi-dimensional labeled samples to train a target environment classifier. The method can track and learn initial no-label data according to environment, and makes up problems that face recognition is low in recognition rate under conditions of small faces, poor light, and excessive face angles, and greatly improves home identity identification practicality.

Description

technical field [0001] The invention belongs to the technical field of computer pattern recognition, and in particular relates to a self-learning identity recognition method. Background technique [0002] At present, for identity recognition in the home environment, facial features obtained by deep learning are mainly used for authentication and recognition. Due to the distribution gap between the training set and the objects in the home environment, and the lighting conditions in the home environment are not good, and the use scene is not suitable. Due to manual registration of high-quality face images, the angle of the face changes greatly, resulting in unstable face alignment and low reliability of face recognition in actual use, and there has been no stable and effective solution to these problems Program. [0003] Chinese patent application 201610544157.X discloses a face recognition method. By performing Gamma correction on the face image, the gray value of the shadow...

AI Technical Summary

Problems solved by technology

[0002] At present, for identity recognition in the home environment, facial features obtained by deep learning are mainly used for authentication and recognition. Due to the distribution gap between the training set and the objects in the home environment, and the lighting conditions in the home environment are not good, and the use scene is not suitable. Due to manual registration of high-quality face images, the angle of the face changes greatly, resulting in unstable face alignment and low reliability of face recognition in actual use, and there has been no stable and effective solution to these problems Program

However, in the home environment, the light may be very weak, and people are very relaxed in the home environment, the angle of the face changes greatly, and occlusion is very likely to occur, the quality of the face is low, and it cannot meet the requirements of automatic registration and recognition. It is not practical in the actual scene. Low-quality and large-angle face images affect the final recognition effect; there is also the problem that the distribution of the training set and the target set are different, and the transfer loss is very large. Even if the experimental performance reaches a very high level, the actual Still can't meet the requirements in use

Images