Sign-correlation partition based on global supervised descent method for face alignment
Authors: Zhang, Y., Liu, S., Yang, X., Shi, D. and Zhang, J.J.
Journal: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume: 10113 LNCS
Pages: 281-295
eISSN: 1611-3349
ISBN: 9783319541860
ISSN: 0302-9743
DOI: 10.1007/978-3-319-54187-7_19
Abstract:Face alignment is an essential task for facial performance capture and expression analysis. As a complex nonlinear problem in computer vision, face alignment across poses is still not studied well. Although the state-of-the-art Supervised Descent Method (SDM) has shown good performance, it learns conflict descent direction in the whole complex space due to various poses and expressions. Global SDM has been presented to deal with this case by domain partition in feature and shape PCA spaces for face tracking and pose estimation. However, it is not suitable for the face alignment problem due to unknown ground truth shapes. In this paper we propose a sign-correlation subspace method for the domain partition of global SDM. In our method only one reduced low dimensional subspace is enough for domain partition, thus adjusting the global SDM efficiently for face alignment. Unlike previous methods, we analyze the sign correlation between features and shapes, and project both of them into a mutual sign-correlation subspace. Each pair of projected shape and feature keep sign consistent in each dimension of the subspace, so that each hyperoctant holds the condition that one general descent exists. Then a set of general descent directions are learned from the samples in different hyperoctants. Our sign-correlation partition method is validated in the public face datasets, which includes a range of poses. It indicates that our methods can reveal their latent relationships to poses. The comparison with state-of-the-art methods for face alignment demonstrates that our method outperforms them especially in uncontrolled conditions with various poses, while keeping comparable speed.
https://eprints.bournemouth.ac.uk/29290/
Source: Scopus
Sign-Correlation Partition Based on Global Supervised Descent Method for Face Alignment
Authors: Zhang, Y., Liu, S., Yang, X., Shi, D. and Zhang, J.J.
Journal: COMPUTER VISION - ACCV 2016, PT III
Volume: 10113
Pages: 281-295
eISSN: 1611-3349
ISBN: 978-3-319-54186-0
ISSN: 0302-9743
DOI: 10.1007/978-3-319-54187-7_19
https://eprints.bournemouth.ac.uk/29290/
Source: Web of Science (Lite)
Sign-correlation partition based on global supervised descent method for face alignment
Authors: Zhang, Y., Liu, S., Yang, X., Shi, D. and Zhang, J.J.
Conference: ACCV 2016: 13th Asian Conference on Computer Vision
Pages: 281-295
Publisher: Lecture Notes in Computer Science, vol 10113.
ISBN: 9783319541860
ISSN: 0302-9743
Abstract:Face alignment is an essential task for facial performance capture and expression analysis. As a complex nonlinear problem in computer vision, face alignment across poses is still not studied well. Although the state-of-the-art Supervised Descent Method (SDM) has shown good performance, it learns conflict descent direction in the whole complex space due to various poses and expressions. Global SDM has been presented to deal with this case by domain partition in feature and shape PCA spaces for face tracking and pose estimation. However, it is not suitable for the face alignment problem due to unknown ground truth shapes. In this paper we propose a sign-correlation subspace method for the domain partition of global SDM. In our method only one reduced low dimensional subspace is enough for domain partition, thus adjusting the global SDM efficiently for face alignment. Unlike previous methods, we analyze the sign correlation between features and shapes, and project both of them into a mutual sign-correlation subspace. Each pair of projected shape and feature keep sign consistent in each dimension of the subspace, so that each hyperoctant holds the condition that one general descent exists. Then a set of general descent directions are learned from the samples in different hyperoctants. Our sign-correlation partition method is validated in the public face datasets, which includes a range of poses. It indicates that our methods can reveal their latent relationships to poses. The comparison with state-of-the-art methods for face alignment demonstrates that our method outperforms them especially in uncontrolled conditions with various poses, while keeping comparable speed.
https://eprints.bournemouth.ac.uk/29290/
Source: BURO EPrints