2.5D SIFT Descriptor for Facial Feature Extraction

This paper presents an application of SIFT (Scale Invariant Feature Transform) in 2.5D facial feature extraction. As range images have more rich in geometric features than that in 2D Images, we intend to improve the SIFT algorithm to extract facial features in 2.5D images. According to face topology and differential geometric properties of surfaces, the extracted key points from 2.5D range images are divided into 9 different surface types for further match. The significance of work presented here is that 2.5D SIFT algorithm has more rotation invariance and robust in facial feature extraction.