Virtual multi-pose faces produced method with single sample on homography transformation
Zhang Yan, Zhu Hong, Mao Wen, Jiao Li · 2016
In recent decades, face recognition has been a hot topic in image processing, pattern recognition and other areas. To solve the issue of the lack of feature information brought by single training face sample, this paper proposed a method on homography transformation, which is about producing virtual multi-pose face by a front view image of face. First, assume that the single sample face is positive to camera, then let the plane of the face which the camera optical axis is perpendicular to, respectively, relative to the camera rotates a certain angle in up, down, left and right direction. There is a homography transformation relationship between the rotated three-dimensional(3D) pose projection images and the front view of face image. The relationship is related to camera view angle and the angle of rotation, finally, the paper proposed a model which can calculate the homography matrix to produce virtual multipose faces images. The experimental results demonstrate that the method in this paper achieves good recognition effect.