Research On Nose Tip Detection Based On Near-Infrared Images

The detection and location of facial organs are of great significance in the human face pose estimation, face recognition, facial expression recognition, face tracking and other aspects. Nose tip, which is most difficult affected by changes in expression, is one of the facial organs, and thus plays an important role in the face pose estimation. However, since the texture features near the nose tip are not clear, the nose tip detection becomes more difficult than the other organs of the face.For nose tip detection when the nostrils are visible, researchers have proposed some effective methods. But when the nostrils are not visible, there appears to be no effective method to detect the nose tip. In this paper, by studying and observing a large number of face images which are taken in different poses, we conclude that when the nostrils are not visible, nose texture can be divided into two broad categories:(1)Obviously the nose lower contour, which corresponds to the front face((left and right rotation is less than 15 °)when the neck bows; (2) the clear bridge of nose, which corresponds to the side (left and right rotation is greater than 15°) when the neck bows. For Class 1, we propose a nose tip detection algorithm based on the nose lower contours. Firstly, we extract the nose lower contour roughly, and then use the least square method to fit the ideal nose lower contour, finally we make use of the proposed extreme point of the fitting curve to calculate the position of the nose tip. For Class 2, we propose a nose tip detection algorithm based on the nose bridge of the nose. Firstly, we extract the candidate bridge of nose, and then use Hough transform method to detect the true bridge of the nose, finally make use of the 3D face model to determine the nose tip.In order to prove the effectiveness of the new algorithm mentioned in this paper, we design two experiments on near-infrared image database in which the eye distance is 125 pixels in the normalized images. On 260 near-infrared face images, we use the nose tip detection algorithm based on the nose lower contour to detect it, and the nose tip detection rate, which the error of nose tip detection is less than 5 pixel, was 94.7%. On 547 near-infrared face images, we use the nose tip detection algorithm based on the bridge of nose to detect it, the nose tip detection rate, which the error of nose tip detection is less than 5 pixels, is 87.4%.