Nose Tip Detection and Face Localization from Face Range Image Based on Multi-angle Energy

In this paper, we propose a novel method to detect nose tip and localize face from face range image. The nose tip detection procedure of the method is based on the idea of Multi-angle Energy (ME) and works in scale-space. The face localization procedure of the method is based on the position of the nose tip and a modified version of Multi-angle Energy. The scale-space is established by robust smoothing the input face range image. In the nose tip detection procedure, for each scale of the scale-space, we compute the Multi-angle Energy for each point of the face range image. For the points whose values of ME are not equal to zero, hierarchical clustering method is used to cluster them into several clusters. In the obtained first h largest clusters, we can find a nose tip candidate by using a cascading scheme. For all scales of the scale-space, we get a series of nose tip candidates. We apply hierarchical clustering again for them. Nose tip can be found in the largest cluster. In the face localization procedure, we present a modified version of ME. With the modified ME, we use a similar cascading scheme to detect one endocanthion for the input face range image. Based on the distance between nose tip and endocanthion, face localization is achieved by using a sphere which is centered on the nose tip to crop the face region. We evaluate our method on two well-known 3D face databases, namely FRGC v2.0 and BOSPHORUS, and compare our method with other state-of-the-art methods. The experimental results show that the nose tip detection rates of our method are higher than those of the state-of-the-art methods. The face localization results are fine and can adapt to the face scale variance.