A robust subspace algorithm for projective reconstruction from multiple images

All the points are visible in all views and mismatched datum (outliers) are not presented in measure matrix are necessary conditions of the existing subspace method. To eliminate the harmful effect of outliers and missing datum to factorization method, a robust subspace projective reconstruction method is proposed in this paper. Augmented Lagrange multipliers (ALM) imposed rank constraints can be used for solving a convex optimization problem. By minimizing a combination of nuclear norm and L1-norm, low-dimension subspaces of measure matrix can be obtained in this convex optimization. Projective shape and projective depths are alternatively estimated in the subspace projective reconstruction method. The two sub-problems are formulated in a subspace framework and same objective function is iteratively minimized on two independent variable sets. The above improvements can effectively ensure the convergence of the iterative process. Comparing with Tang's subspace method, experimental results are provided to illustrate the validity and reliability of the proposed algorithm in this paper.