Child identity recognition method based on face recognition and head and shoulder re-recognition

Abstract

The invention relates to a child identity recognition method based on face recognition and head and shoulder re-recognition, and the method comprises the steps: carrying out the face detection of an obtained child photo, obtaining the position coordinates of a face key point, carrying out the alignment of the position coordinates of the face key point through a similarity transformation algorithm, and obtaining a head and shoulder alignment image; carrying out feature extraction on the head and shoulder alignment image through an SE-Resnet structure, carrying out face and head and shoulder feature down-sampling on feature maps, and generating face and head and shoulder region feature maps respectively; calculating a loss function, and generating a child identity recognition model; inputting the head and shoulder alignment image into a trained child identity recognition model to obtain face and head and shoulder feature vectors, and performing feature fusion on the face and head and shoulder feature vectors to obtain a fused feature vector; and finally, establishing a feature registration database according to the fused feature vectors, and performing child identity recognition in the feature registration database through the cosine similarity of the fused feature vectors, so that the error recognition rate is remarkably reduced.

Description

technical field [0001] This application relates to the computer field, in particular to a child identification method based on face recognition and head-and-shoulder recognition. Background technique [0002] With the promotion of campus digital management at any time, more and more campus access control has applied the face recognition system. However, face recognition for children aged 2-14 is more difficult, and the misrecognition rate is higher in actual use. The main reason is that children’s facial development is immature, and features such as contours, edges and corners, and facial features are less prominent in two-dimensional imaging, which increases the difficulty of algorithmic feature expression. Therefore, a specific face recognition algorithm is needed to improve the identification rate of children aged 2-14. [0003] In related technologies, face recognition algorithms based on deep learning are mainly used. However, the most advanced face recognition algori...

AI Technical Summary

Problems solved by technology

[0005] The embodiment of the present application provides a child identity recognition method based on face recognition and head-and-shoulder re-recognition to at least solve the problem of the single face recognition model in the case of identifying and retrieving children's identities in the related art. Insufficient ability leads to high misrecognition rate

Images