Stereoscopic face reconstruction from a single 2D face image using orthogonality of normal surface and Y-Ratio

Abstract

This research proposes a 3D face reconstruction method with single image and condition of Lambertian model. This method can be divided into 4 parts: surface albedo estimation, normal surface calculation, actual height calculation and accuracy evaluation. The first part which is surface albedo estimation started with initial normal surface from Average face data and morphed it to match an input image by my pattern morphing method. Then the estimated normal surface, Intensity map and Light source vector of input image are used to estimate the albedo of each point on input image. The second part, normal surface calculation, started with finding the relation equation of Normal vector in X and Y axis and calculate the normal surface. The third part is pixels’ height calculation. And the last part is a accuracy evaluation of the method and surface display. The experiment was done with 114 face’s images from 2 face’s database compared with the minimization approach and the result shown that my approach gave the better accuracy and consumed the less process time.