Spherical-Based Iterative Closest Point With Nose Tip Localization For Unconstraint Viewpoint 3D Face Registration
3D face registration remains as a critical area of research as it can be applied for image fusion, face recognition, and motion analysis to name a few. It is a challenging task to register a viewpoint face image due to facial occlusion. Generally, the occluded face image due to viewpoint is enhanced...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/55400/1/Spherical-Based%20Iterative%20Closest%20Point%20With%20Nose%20Tip%20Localization%20For%20Unconstraint%20Viewpoint%203D%20Face%20Registration.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | 3D face registration remains as a critical area of research as it can be applied for image fusion, face recognition, and motion analysis to name a few. It is a challenging task to register a viewpoint face image due to facial occlusion. Generally, the occluded face image due to viewpoint is enhanced and restored using generic image processing techniques prior to face registration. On the other hand, such an approach is approximate. The first contribution of this study is the proposed spherical-based iterative closest point (SICP) for unconstraint viewpoint 3D face registration prior to any kind of image enhancement or restoration. In addition, SICP
face registration method able to assist nose tip localization using the genetic algorithm (GA). The proposed SICP+GA method was evaluated on publicly available 3D face database that
provides various face poses (including viewpoint up to approximately ±90° yaw angle) and was compared against coherent point drift (CPD) and iterative closest point (ICP) based
methods. The experimental finding suggests that the proposed SICP+GA is superior compared to coherent point drift (CPD) and iterative closest point (ICP) based methods, yielding an
average mean registration error of 1.0758 mm compared to 1.2465 and 1.3540 mm of later two methods, respectively. More importantly, SICP+GA successfully registered all the frontal
and non-frontal images while CPD and ICP failed to register profile posed when the viewpoint yaw angle was about ±90°. In addition, the proposed coarse-to-fine nose tip localization with the assistance of SICP face registration method using GA has improvised the nose tiplocalization distance with an average mean of 1.6767 mm. Also, the experimental finding
shows that the proposed technique comes handy when the coarse nose tip annotation is approximate. Contrarily, the face images can be rather noisy and contain outliers which may skew the registration performance. In the effort of investigating the robustness of the proposed SICP face registration method against noise and outliers, post- and pre-filtering methods were performed on SICP face registration method using existing 2.5 |
|---|