3D Structure Estimation Using Evolutionary Algorithms Based on Similarity Transform

In this paper, an efficient 3D reconstruction algorithm is proposed based on multiview 2D images of human face by means of similarity transform. The pose and depth estimation from 2D feature points of the respective face images is considered as an optimization problem and solved using soft computing techniques. Four real parameter optimization techniques (Differential Evolution, Genetic algorithm, Particle Swarm Optimization and Simulated Annealing) are used to estimate the pose and 3D structure. To evaluate the accuracy of the pose estimation to corroborate depth estimation, Head Pose Database is used. The experimental results show that the Differential Evolution significantly outperforms the other optimization techniques in estimating the pose and depths of important features, preserving facial symmetry.