Face Alignment using Modified Supervised Descent Method

Abstract

Face alignment has been used on preprocess stage in computer vision’s problems. One of the best methods for face aligment is Supervised Descent Method (SDM). This method seeks the weight of non-linear features which is used for making the product and the feature resulting estimation on the changes of optimal distance of early landmark point towards the actual location of the landmark points (GTS). This article presented modifications of the SDM on the generation of some early forms as a sample on the training stage and an early form on the test stage. In addition, the pyramid image was used as the image for feature extraction process used in the training phase on linear regression. 1€ filter was used to stabilize the movement of estimated landmark points. It was found that the accuracy of the method in BioID dataset with 1000 training images in RMSE is approximately 0.882.