Bosphorus Database. 3D Face Database · Hand Database · 3D Face Database · 3D/2D Database of FACS annotated facial expressions, of head poses and of. The Bosphorus Database is a database of 3D faces which includes a rich set of IEEE CVPR’10 Workshop on Human Communicative Behavior Analysis, San. Bosphorus Database for 3D Face Analysis Arman Savran1, Neşe Alyüz2, Hamdi Dibeklioğlu2, Oya Çeliktutan1, Berk Gökberk3, Bülent Sankur1, Lale Akarun2 1.

Author: Dubar Morisar
Country: Barbados
Language: English (Spanish)
Genre: Automotive
Published (Last): 27 September 2005
Pages: 446
PDF File Size: 17.26 Mb
ePub File Size: 13.71 Mb
ISBN: 156-9-47447-310-5
Downloads: 31690
Price: Free* [*Free Regsitration Required]
Uploader: Kajilmaran

This is my PhD webpage. I submitted my PhD thesis on automatic 3D face landmarking in December I passed my viva in March Nick Pears and Prof. My work was to evaluate new techniques for automatic face landmarking and face recognition. I mainly focus on machine-learning-based 3D-shape-analysis techniques for facial landmarking. Landmarking using only 3D data is a difficult problem, especially in real-world non-cooperative cases where facial features has to be detected despite the presence of occlusions, pose variations, and spurious bosphoruz in the captured 3D scans.

This page provides information about this 3 years research project with links to publications and downloadable source code.

For information beyond the scope of this page, please contact me directly by email. The first expected outcome for analyais research is a face recognition technique more robust to face bospuorus than the current state-of-the-art. The applications for that kind of research are mainly Security ex: Automatic landmark detection techniques can also help in all domains where face labelling is needed on big database, from computer vision to psychology.

Automatic landmark detection can be used in face recognition for two purposes. It can help to find correspondences between two faces before matching and it can help to extract discriminative information about the face being treated.

The information can be featural and be supported by the neighbourhood of each landmark e. For the landmark detection, I combine sets of simple fields, for example several types of curvature and volumetric information as well as crestline and isolines on the surface to detect points.

For both landmark labelling and face matching, we construct hypergraphs upon the detected landmarks and match them on models using hypergraph matching techniques. The hypergraph structure in our code allows us to store and match relational information of any degree. In the case of complete hypergraph all-to-all connectivity we usually don’t go above degree 3 for speed concerns. Our keypoint detection system evaluate for every vertex a score of being a point of interest by using a pre-computed statistical dictionary of local shapes.

Clement Creusot, PhD

The correlation between the input vertices and the learnt features are computed using a large number of local shape descriptors mainly based of discrete-differential-geometric properties of local area patch around the vertex. The number and nature of the local descriptors, as well as the size of the neighborhoods on which they are computed and the way they are combined can be optimized using basic matching learning techniques such as LDA linear discriminant analysis or Adaboost adaptative boosting.

Here are examples of landmarking results obtained in October on the two bbosphorus using our keypoint-detection system coupled with a RANSAC geometric-registration technique. All the scripts and applications provided here have only been tested on Linux Ubuntu and Linux Mint. All the programs provided here are under GPL v3except if specified otherwise in the headers. Where is the interesting stuff?


In order to comply with the university regulation, I can not provide the sources databaes program that might be sensitive in term of intellectual property. However if you have a request about a specific program I have developed feature detector, hypergraph matcher, Code extracted from our framework that compute maps over a surface mesh for a set of local-shape descriptors.

The output mesh is made of triangles defined counterclockwise when seen from the camera position. Produce a 2D depth map of at most x by y pixels. The pixel bospnorus is forced to square.

For each pixel i,j the z i,j coordinate is interpolated on the triangle intersected by the vector passing by x i,j ,y i,j and colinear to axis z.

If several triangle intersect this line, the maximal value of z is kept.

Bosphorus Database

Try to create a minimal number of vertex from the triangles except if option -f is given. Generate only points and triangle faces. The last column is a 4×4 matrix format: The transformation analtsis computed using ICP on a cropped part of the face. All first faces of each individual have been registered to wnalysis first face of the first individual of the database d Then, all following faces per individual have been registered on the first one.

It should be good enough as initial transformation but may require refinement for boshporus applications. 3dd address the problem of automatically detecting a sparse set of 3D mesh vertices, likely to be good candidates for determining correspondences, even on soft organic objects. We focus on 3D face scans, on which single local shape descriptor responses are known to be weak, sparse or noisy.

Our machine-learning approach consists of computing feature vectors containing D different local surface descriptors. These vectors are normalized with respect to the learned distribution of those descriptors for some given target shape landmark of interest.

Then, an optimal function of this vector is extracted that best separates this particular target shape from its surrounding region within the set of training data. We investigate two alternatives for this optimal function: Our system achieves state-of-the-art performance while being highly generic. Bibtex File [bib] Plain text. We present a machine learning framework that automatically generates a model set of landmarks for some class of registered 3D objects: The aim is to replace heuristically-designed landmark models by something that is learned from training data.

The value of this automatically generated model is an expected improvement in robustness and precision of learning-based 3D landmarking systems. Simultaneously, our framework outputs optimal detectors, derived from a prescribed pool of surface descriptors, for each landmark in the model. The model and detectors can then be used as key components of a landmark-localization system for the set of meshes belonging to that object class.

Automatic models have some intrinsic advantages; for example, the fact that repetitive shapes are automatically detected and that local surface shapes are ordered by their degree of saliency in a quantitative way. We compare our automatically generated face landmark model with a manually designed model, employed in existing literature.


Keypoints on 3D surfaces are points that can be extracted repeatably over a wide range of 3D imaging conditions. They are used in many znalysis shape processing applications; analysi example, to establish a set of initial correspondences across a pair of surfaces to datavase matched. Typically, keypoints are extracted using extremal values of a function over the 3D surface, such as the descriptor map for Gaussian curvature.

That approach works well for salient points, such as the nose-tip, but can not be used with other less pronounced local shapes. In this paper, we present an automatic method to detect keypoints on 3D faces, where these keypoints are locally similar to a set of previously learnt shapes, constituting a ‘local shape dictionary’.

The local shapes are learnt at a set of 14 manually-placed landmark positions on xnalysis human face. Local shapes are characterised by a set of 10 shape descriptors computed over a range of scales.

For each landmark, the proportion of face meshes that have an associated keypoint detection is used as a performance indicator. Repeatability of the extracted keypoints is measured across the FRGC v2 database.

Most 3D face processing systems require feature detection and localisation, for example to crop, register, analyse or recognise faces. The three features often used in the literature are the tip of the nose, and the two inner corner of the eyes. Failure to localise these landmarks can cause the system to fail and they become very difficult to detect under large pose variation or when occlusion is present.

Bosphorus 3D Face Database > Publications

analgsis In this paper, we present a proof-of-concept for a face labelling system, capable of overcoming this problem, as a larger number of landmarks are employed.

A set of points containing hand-placed landmarks is used as input data.

The aim here is to retrieve the landmark’s labels when some part of the face is missing. By using graph matching techniques to reduce the number of candidates, and translation and unit-quaternion clustering to determine a final correspondence, we evaluate the accuracy at which landmarks can be retrieved under changes in expression, orientation and in the presence of occlusions. Local-descriptor map computation 1. LIC and generic solve. Idem to localSum, but return mean values volume: Compute coarce tetrahedron volume in a given neighborhood willmore: Compute a willmore energy dataase field from the discrete mesh not tested distToLocalPlane: Compute the DLP using the given normals and neighborhood byProduct: Compute a field as a function of different other fields mean: Local symetry descriptor LSD read: A progress bar is generated on stderr showing percentage, remaining time and current file name.

More convenient than a bash equivalent: Output mesh after obj2abs. Mean faces – Using mean depth map of registered model sets low resolution 5. Capture of the mesh. If you improve this ground truth, please publish the changes you have made. Download Registration Video [avi, 8 MB] – faces in 2: In the right picture, The green points represent the ground truth, the blue points our datahase.