Department of Applied Mathematics, College of Science, Zhongyuan University of Technology, Zhengzhou, Henan, China.

---

Abstract—The facial keypoints detection is a challenging task due to the large variation of facial features, the change in 3D viewing angle, and difference in size and position of the face. Over the years, researchers have proposed a variety of algorithms such as combining multiple weak classifiers in cascade. However, a lot of work still needs to be done to further improve the detection accuracy and to accommodate for extreme cases. In this project, we proposed to use deep convolutional neural networks to locate the facial keypoints. Specifically, we experimented with LeNet, VGGNet and a 14-layer CNN on the Kaggle dataset. We also adopted image augmentation techniques to further increase the training set size. Finally, we were able to achieve a MSE of 3.02 with the VGGNet. The result indicated that deep CNNs have fairly good performance for the facial keypoints detection task.

Index Terms—Facial keypoints detection, deep convolutional neural network, LeNet, VGGNet, data augmentation.

[PDF]

Cite: Ran Gao, Qi Liu, "Facial Keypoints Detection with Deep Learning," Journal of Computers vol. 13, no. 12, pp. 1403-1410, 2018.