|
3D Reconstruction of a Morphable Face from Finite 2D Profiles |
|
The research theme is to render a simulation of faces with several profiles and the profiles could be in different pose and/or expression. Therefore, in this research, a subject’s faces are going to be tracked, modeled and rendered. A simulation of facial gestures transformation from the first profile and the final profile would be the final result of this research. |
|
3D Face Database |
|
Figure 1.0 |
|
Figure 2.0 |
|
3D Face Alignment |
|
For the first few months, we have been working with FRAV3D database. FRAV3D, developed by Universidad Rey Juan Carlos, Spain, is a database with 106 subjects. However, the database does not contain data with occlusion problem; no facial landmarks information as well as the point-wise correspondences between models is not provided.
The drawbacks of FRAV3D database are the reason we have decided to use Bosphorus Database in our research project. There are 105 subjects with 53 different face scans per subject in Bosphorus Database . The database provides a rich set of expressions, systematic variation of poses and different types of realistic occlusions. Each scan is intended to cover one pose and/or one expression type. Furthermore, Bosphorus database is complete with 24 facial landmark points; provided that they are visible in the given scan (i.e., the right and left ear lobe can not be seen from frontal pose). The landmark points are manually labelled by the developer and each segmented 3D face approximately consists of 35, 000 points.
|
|
3D face alignment is needed to ensure that the shapes of the data are as close to each other as possible while keeping that particular shapes are unchanged. For the scanned 3D faces with high dimensional dense points, the data is too large to do a global transformation for all points. Thus, the alternative solutions to use sub-sampling sparse point sets and in this project, only three feature points are used. Figure 1.0 shows the faces without alignment process and figure 2.0 shows the three facial features that are used in 3D face alignment process. |
|
Figure 3.0 |
|
Figure 4.0 |
|
Figure 3.0 shows the unaligned and aligned for the three landmarks on 14 faces of neutral faces.. Figure 4.0 shows the unaligned and aligned for the all landmarks on 14 faces of neutral faces. The red dots indicate the three significant landmarks that are first used to find the transformation set. Figure 4.0(c) is the reference for aligned landmarks, 8 and 9 is the inner eye corner and 14 is the nose tip. As we can see in figure 4.0(b) the red dots are separated into three parts which correspond to three landmarks used. |
|
Future Plan |
|
Our future work is focused on the following tasks and the time frame in describe using the Gantt chart.
1. Test the developed 3D coarse alignment with all neutral profiles available as well as profiles of different poses and/or expressions. 2. Set up the GNU Scientific Library (GSL), a numerical library to solve the least square problems face in PCA and 3D fine alignment. 3. Equalize the number of vertices. 4. Develop a 3D fine alignment (i.e. ICP), followed by series of experiments with different type of data. 5. Develop a 3D face model based on 3DMM. 6. Develop a 3D face rendering system. |
|
Biography |
|
Hamimah obtained his first degree, BSc.(majoring in Computer Graphics) from Universiti Teknologi Malaysia( UTM) in 2003. She worked as a research assistant in UTM from April 2003 until October 2003. In November 2003 she had been granted a full scholarship (National Science Fellowship fund) from Malaysian Ministry of Science, Technology and Innovation for her MSc. With the fund, she further her master degree (by research) in UTM. Her research during her master study was about A Heuristic Coordinate Filtering Scheme to Aid Online Indexing of Spatio-Temporal Trajectories. While on her report-writing phase, she has been offered a tutor post in Universiti Malaysia Sarawak (UNIMAS). On 19 December 2005, she started to serve UNIMAS and has been upgraded to a lecturer post on 2 November 2006. For her career in UNIMAS (courses taught, research projects and et cetera), please refer to Hamimah@UNIMAS. In November 2008 , she has been awarded a full scholarship from Malaysian Ministry of Higher Education for her PhD study in University of Birmingham., United Kingdom which commenced on 1st March 2009.
|
|
Research Interests |
|
3D face rendering, 3D facial recognition, 3D shape recovery, 3D face reconstruction |