Face modeling process based on Dlib

In this paper, the face modeling problem, a random forest model on each feature point by pixel difference feature, by regression estimation of forest model shape training samples; to estimate the shape of training samples for linear least squares fitting and real shape, a global optimization model; and then use the model to test the sample feature point location regression estimation and shape optimization, so as to realize the automatic localization of facial feature points. A method based on gradient enhancement is proposed to deal with the feature data and solve the problem of missing feature points by means of cascade learning of regression tree. In addition, the relationship between regularization parameters and over-fitting phenomena is also explored. At the same time, the ratio of the number of training data to the prediction accuracy of the model is studied. At the same time, the data is synthesized by the affine transformation of the existing data in the case of insufficient data.