Facial Keypoints Detection Using CNN And Autoencoders

Abstract

Facial keypoints detection is an essential task in computer vision and augmented reality domains. We put forth a novel solution for this problem which different than existing techniques. We implement a Convolutional Neural Network-based model with Autoencoders to detect facial keypoints. The deep structure of Convolutional Neural Networks (CNN) enables extraction of high-level features and gives more accuracy while locating each key point. Convolutional networks are shaped to predict all the points simultaneously. With autoencoders it is possible to boost the performance of the model. Autoencoders are data-specific and can work effectively if similar data was used while training.