hello-luck / 3d-img-face-analysis

Code to automatically analyse 3D point clouds to identify characteristic points.

Code used in my BSc in Engineering dissertation:

T. Kusmierczyk: Point descriptors in morphological analysis of three-dimensional face images. Warsaw University of Technology 2010.

---

Functionality

The system is prepared to automatically analyse 3D (face) scans (x,y,z point clouds) to identify characteristic points. Our approach relies on normal vectors estimation and local surface descriptors.

The system has the following functionality:

• preprocessing, filtering and smoothing
• normal vectors calculation
• point descriptors calculation
• subregions (e.g., face and nose ) localization
• and more...

Dissertation abstract

The dissertation presents a system prepared to automatically analyse three-dimensional face scans. Its design and implementation process is described. The system allows for face and nose localization in source cloud of points. Its functionality also includes identification of characteristic points on nose.

In theoretical introduction a review of known methods of free-form objects analysis is presented. An approach based on point descriptors is explained. Basing on its computational and algorithmic features the choice of this technique is clarified. Selection of sources is motivated. Measurement data characteristics are presented.

The practical part describes the design and implementation of proposed system. In the first phase source data is filtered and preprocessed. Algorithms used for normal vectors estimation and descriptors calculations are explained. Innovative procedure of hierarchical region localisation is presented. The process of face, nose and characteristic points identification, based on this procedure is revealed. Each element of the system was tested. Results are discussed and interpreted.