Robust nose tip detection for face range images based on local features in scale-space

Being the most distinct feature point in 3D facial landmarks, nose tip plays a significant role in 3D facial studies such as face detection, face recognition, facial features extraction, face alignment, etc. Successful detection of nose tip can facilitate many tasks of 3D facial studies. In this paper, we propose a novel method to detect nose tip robustly. The method is robust to noise, needs not training, can handle large rotations and occlusions. To reduce computational cost, we first remove small isolated regions from the input range image, then establish scale-space by robust smoothing the preprocessed range image. In each scale of the scale-space, the Multi-angle Energy (ME) of each point is computed and sorted in descending order. Then the first m points in the descending order list are obtained and hierarchical clustering method is used to cluster these points. In the first h largest clusters, we can find one point with the largest ME. For all scales of the scale-space, we get a series of such points which are treated as nose tip candidates. For these candidates, we apply hierarchical clustering again. In the obtained largest cluster, we compute the mean value of ME. The ME of nose tip will be closest to the mean value. We evaluate our method in two well-known 3D face databases, namely FRGC v2.0 and BOSPHORUS. The experimental results verify the robustness of our method with a high nose tip detection rate.