Face recognition method based on non-uniform distribution

Abstract

The invention provides a face recognition method based on non-uniform distribution, and belongs to the field of image mode recognition. Firstly, a face detector is used for detecting and interceptingface areas, then an affine transformation matrix is used for carrying out side face correction operation on all the intercepted face areas, and then a deep learning model is used for extracting features of a face picture subjected to side face correction. Training and testing are performed on a deep learning model, and the training and the testing comprise: (1) training the deep learning model; and (2) after side face correction operation is carried out on a face area detected and intercepted by a face detector, extracting original features by utilizing the trained deep learning model, carrying out normalization processing, and calculating cosine similarity between normalized features of different original features to judge whether the features belong to the same person or not. A large-boundary feature enhancement technology and a migratable feature normalization technology are combined, so that the face recognition efficiency is greatly improved, and the stability of a face recognition system is also greatly improved.

Description

technical field [0001] The invention belongs to the field of image pattern recognition, in particular to a face recognition technology based on deep learning. Background technique [0002] Face recognition is a biometric technology for identification based on human facial feature information. It usually detects faces from acquired dynamic or static images, and then identifies the detected faces. A complete face recognition process such as figure 1 shown. The face recognition system first inputs the face detection module through the camera, video files or image files, and then intercepts all face images from the displayed image, and then normalizes the non-normalized faces through the face correction module for the standard face. After the face image is preprocessed, the face features are extracted through the face recognition algorithm model, and the features are calculated and compared, and finally the recognition result is obtained. [0003] The existing face training ...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the serious uneven distribution of the face training set used in the current face recognition system, which affects the generalization ability of the face model and reduces the performance of face recognition, the present invention proposes a face recognition method based on non-uniform distribution. A new way to identify

Images