Method and device for determining the position of key points of human face

Abstract

The embodiment of the present invention proposes a method and device for determining the position of the key points of the human face. The method includes: obtaining the preliminary coordinates of the key points of the human face in the image through a coordinate regression network; feature, through the position-sensitive feature extraction network to obtain position-sensitive features, the position-sensitive features include the underlying features of the face corresponding to the coordinates of the preliminary face key points and the semantic features of the face; according to the position-sensitive features, the coordinates of the preliminary face key points are corrected, Determine the final face key point coordinates in the image. The embodiment of the present invention makes full use of the preliminary facial key point coordinates including spatial position information, and the underlying features of the human face with underlying high-resolution features to obtain position-sensitive features, so that the acquired position-sensitive features can have rich semantic features and underlying features. features, thereby improving the accuracy of the final face key point coordinates obtained through correction.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a method and a device for determining the positions of key points of a human face. Background technique [0002] Face key point detection algorithms are mainly divided into two categories according to the technical route: traditional key point coordinate regression methods and deep learning methods using neural networks. Among them, the feature stability used in the traditional key point coordinate regression method is weak, and it is easily affected by factors such as illumination. For deep learning methods, in some application scenarios, due to the small face in the image, the required face feature map after the deep learning network is greatly reduced, and the spatial position information is missing, resulting in the result of face key point detection. Deviation, affect the actual application performance. Contents of the invention [0003] Embodiments of the presen...

AI Technical Summary

Problems solved by technology

Among them, the feature stability used in the traditional key point coordinate regression method is weak, and it is easily affected by factors such as lighting.

For deep learning methods, in some application scenarios, due to the small face in the image, the required face feature map after the deep learning network is greatly reduced, and the spatial position information is missing, resulting in the result of face key point detection. Deviation, affecting actual application performance

Images