Effect of Training Sample on Reconstructing 3D Face Shapes from Feature Points

Abstract

Statistical learning models are recognized as important emerging methodologies in the area of data reconstruction. The learning models are trained from a set of examples to reach a state where the model will be able to predict the correct output for other examples. PCA-based statistical modeling is a popular technique for modeling 3D faces, which can be used for 3D face reconstruction. The capability of 3D face leaning models in depicting new 3D faces can be considered as the representational power (RP) of the model. This paper focuses on examining the effect of the sample size (number of training faces) on the accuracy of the 3D face shape reconstruction from a small set of feature points. It also aims to visualize the effect of the training sample size on the RP of PCA-based models. Experiments were designed and carried out on testing and training data of the USF Human ID 3D database. We found …