Analysis of CNN Architectures for Pose Estimation of Noisy 3-D Face Images
Randy Pangestu Kuswana, Faqih Akhmad, Benyamin Kusumoputro · 2019
Convolutional neural networks (CNN) has been used in various applications, especially in computer vision field, due to its superiority compare with that of conventional artificial neural networks. In this paper, CNN is developed as head pose estimator for noisy three dimensional face images and analyzed the recognition accuracy for different architectural architecture of the networks, especially on the feature extraction part. Four different amount of layers are experimented, which resulting different input neurons to the estimator part. Experimental results show that the CNN could estimate the head pose with high enough recognition accuracy, and the CNN with 3 feature extraction layers could obtain the highest accuracy of 81.31% for normal face images.