Automatic landmarking on 2.5D face range images

This project presents a novel approach for automatic landmarking process on face range images. The approach consists of feature extraction and feature localisation methods. In recent years, methods to improve existing face processing applications have increased rapidly. This includes automatic landmarking method which could be an important intermediary step for many face processing applications, such as for face recognition, face analysis and etc. The approach aims to locate facial feature points automatically, such as the nose tip, the mouth corners, chin, etc., without the intervention of human. Automatic landmarking holds a number of added advantages over manual landmarking especially if dataset is large, the landmark selection would be less time consuming. Identifying features on a face automatically may be a challenging process for computing. Our human vision system can perceive salient feature easily without any difficulties. For instance, a human is able to detect the eyes, the nose tip and/or the mouth of a person at a first glance. However, a computer system is unable to do such task easily and effortlessly. Therefore, a method to automatically detect and label landmarks on the features of the face is developed. Firstly, features or primitive surface types are extracted from range images. The primitive surfaces are derived using Mean (H) and Gaussian (K) curvatures from a down-sampled by Gaussian Pyramid approach. Otsu’s algorithm is used to place landmark on the extracted facial features and/or regions. In summary, we have successfully implemented an automatic landmarking method and an interactive tool has been developed to ease the visualisation of the overall processes