Nose detection on 3D face images by depth-based template matching

Automatic detection of nose regions on 3D face images is highly important for 3D face registration and recognition. It can also be used in facial landmark detection which is important in facial feature segmentation, facial shape analysis, face synthesis and facial mesh reconstruction. In this paper, we propose a nose detection approach based on template matching of depth images. We have constructed three nose templates which correspond to the whole nose, the left and right half of nose, respectively. By doing this, we can detect nose regions even in situations such as occlusions and missing of facial part in which the symmetrical property of face is destroyed. By using Normalized Cross Correlation (NCC) method, the template matching procedure is time-efficient which allows for fast searching of frontal directions and obtains robustness against expression and pose variations. The experimental results indicate that the proposed method outperforms other two curvature-based algorithms in detection accuracy and obtain an overall 98.75% successful detection on the Bosphorus database.