Human face pose estimation method fusing manual design descriptor and depth features

Abstract

The invention discloses a human face pose estimation method fusing a manual design descriptor and depth features, which is commonly applied to human face image quality evaluation for safety protectionand human face recognition application. According to the method, an SIFT descriptor is used for extracting the contour and local information of a face image, a DeepID deep network is used for extracting appearance and structure information of a face image. The method mainly comprises the following steps: training a deep neural network model for extracting deep features, inputting a face image obtained through detection, and extracting the deep features of the face image by using the deep neural network model; extracting SIFT feature vectors with scale space invariant features, connecting theSIFT features and the depth features of the face images in series, inputting the SIFT features and the depth features into a trained SVM classifier for classification, and determining the posture category of the to-be-classified face. According to the invention, the face pose estimation can be effectively carried out, and the pose estimation accuracy is improved.

Description

technical field [0001] The invention relates to the research fields of face recognition and face image quality evaluation, in particular to a face pose estimation method that combines hand-designed descriptors and depth features. Background technique [0002] In face recognition applications and monitoring security precautions, due to the uncertainty of the image collection environment and conditions, it is difficult to control the quality of the collected face images, and the image quality varies greatly, which will affect the effect of face recognition. In particular, the complex and changeable face poses make it difficult to detect key points of the face, especially when the face angle is seriously deflected, which will affect the effect of face recognition. Therefore, it is necessary to predict and estimate the pose of the face in the image to control the impact of face pose on the recognition performance. How to accurately and quickly estimate the face pose in an image...

AI Technical Summary

Problems solved by technology

[0002] In face recognition applications and monitoring security precautions, due to the uncertainty of the image collection environment and conditions, it is difficult to control the quality of the collected face images, and the image quality varies greatly, which will affect the effect of face recognition

In particular, the complex and changeable face poses make it difficult to detect key points of the face, especially when the face angle is seriously deflected, which will affect the effect of face recognition

Images