Multi-pose facial features localization based on skin color models

Abstract

Localization of multi-pose human facial features is one of the key issues in face recognition technology. The paper presents an algorithm to achieve the accurate positioning of facial features and the corresponding practical steps to realize it. Firstly, we establish the chrominance component mapping function of the feature points based on the chrominance information of high Cb and low Cr values around the key facial feature points. Secondly, we present a mapping function of the brightness components of feature points based on algorithms of the expansion and corrosion of the grayscale morphology. Thirdly, we obtain the map of facial feature points by combining the mapping of chrominance components and brightness components and then attain the binary image from the above map. Finally, we identify the area with feature points by employing four-connected detecting method and then eliminate the areas without feature points according to the proposed rules. The results from the experiments on the face database of ColorFeret and Cas-Peal clearly demonstrate the proposed algorithm is a useful and effective one to achieve accurate positioning of multi-pose human facial features.