Statistical non-rigid ICP algorithm and its application to 3D face alignment*
Introduction
Three-dimensional representation of face has always been a valuable source for face recognition and facial behaviour analysis. The huge descriptive power and rich shape variation of 3D human face make it much more informative than the corresponding 2D projection. Due to its complex nature, the structure and topology of the 3D facial scan varies from frame to frame. To obtain a relatively consistent facial representation, it is always preferable to use a 3D deformable facial model. Thus, the building and fitting of the 3D deformable model becomes a key step in retrieving 3D facial information, and not surprisingly, it has received a lot of attention in the past decade [1], [2], [3], [4], [5], [6], [7], [8]. Owing to the recent development on cost-effective depth cameras, such as Kinect 2, Creative Senz3D™ and Intel® RealSense™ Camera (F200), further attention has been drawn on this field of study.
3D face fitting, also referred to as 3D face registration [5], [8], [9] in the literature, aims at aligning two sets of point clouds or meshes, and overlapping the face template as close as possible with the target surface (i.e. facial scan). In the scope of 3D deformable model fitting, the face template refers to the 3D deformable face model. 3D Morphable Model (3DMM) [1] is the most commonly used technique to introduce prior knowledge on 3D human face. 3DMM has been widely adopted in various 3D fitting methods [2], [10], [11], [12], since it can well estimate an unseen 3D facial shape by solving the shape, texture, pose and illumination parameters simultaneously. Fitting of 3D facial models is very important, since they can be used to identify particular facial landmarks or to recognize and define facial deformations that benefits face recognition [3], [13] and facial performance transfer [14]. Furthermore, accurate fitting and modelling of human expressions is able to boost the performance of facial expression recognition[15], [16], [17].
Despite the importance, the majority of existing methodologies use just a simple pre-defined mesh model, parameterised or not, to fit a target mesh [2], [3], [5], [8], [11], [18]. In [2], [11], in order to handle recognition under different pose and lightening conditions, a 3DMM that separates parameters for shape, head pose and illumination is employed. Inspired by this, Amberg et al. [10] developed an expression-invariant 3D face recognition algorithm, in that, they fit an identity/expression separated 3DMM to the facial scan and normalize the resulting face by removing the pose and expression components [10]. Unfortunately, these methodologies may fail to describe properly the complex, non-linear and highly deformable structure of the face. Especially when fitting the data captured by high-resolution face capturing system, such as Di3D (Dimensional Imaging [19]), the results are likely to be over-smooth and thus lose important facial details and micro-expression.
In this paper, we examine the problem of fitting a 3D facial models to high-resolution depth scan. Our key contribution is a new active method for describing and fitting 3D faces, which is achieved by learning a set of local statistical model for facial parts, and combining them with the non-rigid Iterative Closest Point (ICP) algorithm [5]. Besides it, we propose a dynamic local fitting procedure that makes full use of dynamic subdivision framework. To this end, we successfully adopt the proposed active method in the fitting procedure, and show that it manages to accurately model the subtle facial feature. Additionally, we provide a point-driven mesh deformation procedure in the data pre-processing stage that helps to prevent incorrect facial part fitting. It deforms 3D template model under the guidance of the state-of-the-art 2D face alignment algorithm [20].
The remainder of the paper is organized as follows. Section 2 gives a brief introduction to the existing works that are related to this paper, and a short discussion over the merits and demerits of them. Section 3 explains our dynamic subdivision framework as well as the local fitting procedure. Next, in Section 4, we describe the proposed novel Active Non-rigid ICP algorithm. In Section 5, we show the results of our experiments followed by an in-depth discussion. Lastly, we conclude the paper in Section 6.
Section snippets
Related work
Various 3D face fitting methods have been proposed to address specific problems in different scenarios, such as the situation with high-resolution data [3], partial range scan [5], [21], [22] and normal maps [23]. In this section, we will give a brief introduction to different kinds of 3D fitting algorithms.
Dynamic subdivision framework
The core idea of our framework is to dynamically fit facial data using a deformable 3D face model, and to provide an accurately fitted surface. In contrast to previous works on 3D surface registration [3], [18], [24] that subsample the data using an annotated template to gain efficiency, our method starts from a sparse level and dynamically propagates to subsequent levels, in which the fittings are performed locally to model regional deformation. We argue that the subsampling step sacrifices
Statistical Non-rigid ICP algorithm
To capture more local variations, we perform local fitting based on the segmented template of subdivision levels. Inspired by the recent success in region-based face modelling [31], we employ a statistical shape model in non-rigid ICP algorithm (see Section 5 for details of shape model building), and propose to solve the optimal mesh controlling parameters in an alternating manner. We refer to this method as Dynamic Active Non-rigid ICP (DA-NICP) in this paper.
Experiments
Apart from visual comparison between fittings, we conduct three experiments to provide quantitative measures of our proposed dynamic Active Non-rigid ICP (DA-NICP) algorithm. To demonstrate the advantage of putting subspace constraint on fitting, we introduce D-NICP— a method similar to DA-NICP that uses dynamic subdivision surfaces and perform local fitting, but with NICP [5] chosen as the only fitting strategy. The third method to compare is the deformable fitting with subdivision on AFM in
Conclusion
We propose a dynamic local fitting procedure that gains benefits from dynamic subdivision framework, and show how to adopt the NICP algorithm to our procedure. The proposed fitting procedure is shown capable of modelling subtle facial details. More importantly, we present a statistical model for describing the faces and we combined it with NICP for 3D face alignment. We have shown that the proposed algorithm largely outperforms state-of-the-art 3D facial deformable models, such as the ones that
Acknowledgements
The work of S. Cheng and I. Marras is funded by the EPSRC project EP/J017787/1 (4D-FAB). M. Pantic acknowledges support by the European Community Horizon 2020 [H2020/2014-2020] under grant agreement no. 645094 (SEWA). S. Zafeiriou also acknowledges support from EPSRC project EP/L026813/1Adaptive Facial Deformable Models for Tracking (ADAManT).
References (52)
- et al.
3D/4D facial expression analysis: an advanced annotated face model approach
Image Vis. Comput. (IVC)
(2012) - et al.
Reduced surface point selection options for efficient mesh deformation using radial basis functions
J. Comput. Phys.
(2010) - et al.
Robust Euclidean alignment of 3D point sets: the trimmed iterative closest point algorithm
Image Vis. Comput. (IVC)
(2005) - et al.
A Morphable Model for the Synthesis of 3D Faces
- et al.
A 3D Face Model for Pose and Illumination Invariant Face Recognition
- et al.
Intraclass retrieval of nonrigid 3D objects: application to face recognition
IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI)
(2007) - et al.
3D Face Recognition in the Presence of Expression: A Guidance-based Constraint Deformation Approach
- et al.
Optimal Step Nonrigid ICP Algorithms for Surface Registration
- et al.
The space of human body shapes: reconstruction and parameterization from range scans
ACM Trans. Graph. (TOG)
(2003) - et al.
Establishing point correspondence of 3D faces via sparse facial deformable model
IEEE Trans. Image Process. (TIP)
(2013)
Fast nonrigid mesh registration with a data-driven deformation prior
Registration of 3D point clouds and meshes: a survey from rigid to nonrigid
IEEE Trans. Vis. Comput. Graph.
Expression invariant 3D face recognition with a Morphable Model
Face recognition based on fitting a 3D morphable model
IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI)
Discriminative 3D morphable model fitting.
Three-dimensional face recognition in the presence of facial expressions: an annotated deformable model approach
IEEE Trans. Pattern Anal. Mach. Intell. (T-PAMI)
Reanimating faces in images and video
Comput. Graphics Forum
Static and dynamic 3D facial expression recognition: a comprehensive survey
Image Vis. Comput. (IVC)
Recognition of 3D facial expression dynamics
Image Vis. Comput. (IVC)
Robust Learning from Normals for 3D Face Recognition
Multimodal face recognition: combination of geometry with physiological information
dI4D - 4D Capture Systems
Robust Discriminative Response Map Fitting with Constrained Local Models
Global Correspondence Optimization for Non-rigid Registration of Depth Scans
Multilinear Wavelets: A Statistical Shape Space for Human Faces
3D Face Template Registration Using Normal Maps
Cited by (56)
Gaussian process model based restoration of damaged Buddha statue head
2024, Journal of Cultural HeritageIntelligent paving and compaction technologies for asphalt pavement
2023, Automation in ConstructionGabor Log-Euclidean Gaussian and its fusion with deep network based on self-attention for face recognition
2022, Applied Soft ComputingCitation Excerpt :The image obtained with the RGB-D camera can be converted into a point cloud image, which is the mainstream data used for 3D face recognition. Early methods did not consider facial features such as texture in 3D space, but directly used methods such as ICP to match point cloud images [13]. Like 2D face recognition, 3D face recognition is currently fully transformed into a deep learning model.
Linearly augmented real-time 4D expressional face capture
2021, Information SciencesCitation Excerpt :Then the template was iteratively deformed according to this target face. Cheng et al. [16] presented a non-rigid ICP method for 3D face alignment. Trimech et al. [17] presented a 3D facial expression recognition method.
Robot self-calibration using actuated 3D sensors
2024, Journal of Field Robotics
- *
This paper has been recommended for acceptance by Sinisa Todorovic.