Method and computing device for determining the visibility of human face point

Abstract

The invention discloses a method for judging the visibility of a human face point, which firstly generates a human face point visibility judging model through training, and then performs judging processing on a human face image to be processed through the model. The method for generating a decision model comprises the following steps: obtaining a labeled face image as a training image; constructing a first, a second, and a third processing module; inputting the training image into the first processing module to output the predicted position of each face point, and training the first processingmodule according to the labeling data; using a second processing module to generate a feature map of each human face point according to the position of each human face point; the feature map of eachhuman face point is input to the third processing module to output the confidence level representing the visibility of each human face point, and the third processing module is trained according to the labeling data. Based on the trained first processing module, the trained second processing module and the trained third processing module, a face point visibility determination model is generated. The invention also discloses a corresponding computing device.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and computing device for judging the visibility of human face points. Background technique [0002] Face feature point detection is based on face detection, and it is a solution for locating feature points on the face (such as mouth corners, eye corners, etc.). Face feature point detection has many applications in engineering, such as: (1) Improve face recognition, align faces to average faces through face feature point detection, and the aligned face images can improve the effect of face recognition algorithms ; (2) face fusion, using the results of face feature point detection, two different face images are fused to achieve an effect similar to "face changing"; (3) face beauty, such as on the face On the basis of feature point detection, special effects such as blush, eye shadow and eyeliner, colored contact lenses, and glasses can be added. It can be seen t...

AI Technical Summary

Problems solved by technology

[0003] Deep neural network is widely used in today's facial feature point detection algorithm and has achieved good results, but when the facial features or contours are blocked, errors will occur in subsequent applications

Taking face makeup as an example, when the facial features are partially blocked by an occluder, continuing to make special effects based on facial feature points will lead to the problem of adding makeup or special effects to the occluder

Images