Feature extraction and localisation on 2.5D face range images

In this paper, we propose a method for semi- and fully automatic landmarking detection on raw face data using feature extraction and feature localisation methods. This approach is essential in any face application such face registration, analysis and face recognition methods. This approach involves locating distinct face features, such as the corners of the eyes, the tip of the nose, the chin and etc., without human manual landmarking intervention. Automatic landmarking has a number of advantages over manual landmarking. The process of manual landmarking is time consuming, error prone and limited in accuracy. We will present the accuracy of the landmark detection based on the threshold values and the interactive tool that was also developed to give a better visualisation of the landmarking process. The threshold values are analysed and generalised based on the best detected and extracted important keypoints or/and regions of facial features. We employed the proposed approach on 2.5D range face images. The results of the automatic detection and localisation based on the extracted facial features and candidate landmarks will be shown in this paper. © 2014, UK Simulation Society. All rights reserved.