3D face recognition using hog features based on fine-tuning deep residual networks

As the technology for 3D photography has developed rapidly in recent years, an enormous amount of 3D images has been produced, one of the researches for which is face recognition. Improving the accuracy of a number of data is crucial in the 3D face recognition problems. Traditional machine learning methods can be used to recognize 3D faces, but the face recognition rate has declined rapidly with the increasing number of 3D images. As a result, classifying large amounts of 3D image data is time-consuming, expensive, and inefficient. The deep learning methods have become the focus of attention in the 3D face recognition research. Current methods of assessing 3D face recognitions are limited and often subjective, complex or low accuracy. One of the most prominent methods for assessment showing great promise is residual neural network (ResNet), a shortcut connection way of training a very deep network by randomly dropping its layers during training and using the full network in testing time, which allows for a more quantitative evaluation. With the introduction of feature engineering of HOG method for extracting the discriminative information, and especially finetuning method for reconstructing the ResNet learning model, we are able to calculate a relative high accuracy for the extracted face region. This allows also researchers to effectively assess on a continuous accuracy with fine-tuned ResNet learning model of different depths. However, shadow learning technology is not available in many settings (e.g. curse of dimensionality, accuracy decline) yet so there still exists a need for this quantitative assessment from deep learning methods. How to extract the important and discrimative information from the raw data and efficiently recognize a large number of 3D face images with fine-tuned learning framework at high accuracy was the main task of this research. In our experiment, the end-to-end face recognition system based on 3D face texture is proposed, combining the geometric invariants, histogram of oriented gradients and the fine-tuned residual neural networks. The research shows that when the performance is evaluated by the FRGC-v2 dataset, as the fine-tuned ResNet deep neural network layers are increased, the best Top-1 accuracy is up to 98.26% and the Top-2 accuracy is 99.40%. The framework proposed costs less iterations than traditional methods. The analysis suggests that a large number of 3D face data by the proposed recognition framework could significantly improve recognition decisions in realistic 3D face scenarios