The Muliti-gesture Face Image Build Method On Stereo Vision

In order to overcome the problem of sensitivity to pose which was common in conventional 2D face recognition algorithms, a method for reconstructing face images of specific poses by the use of 3D face model was proposed. This method could improve the performance of 2D face recognition algorithms. The main problem in reconstructing 3D face model was to overcome the defects of impracticability, high computation costs and sensitivity to noise. An method for 3-D face model reconstruction using three range images was presented. The range images were obtained by the means of stereo vision. This method made use of manually chosen and computer calibrated feature points in the face. Then it applied the improved ICP algorithm to get the 3D face model. After the 3D face model was obtained, rotation transform in Euclidean space and the projection matrix were used to produce multi-pose face images. Experimental results showed that the algorithm can produce realistic face images. The average error of integration is about 1.32mm.