明度画像を用いたエッジを含む三次元物体形状推定アルゴリズム

概要

論文の詳細を見る
A conventional method to reconstruct a 3-D shape involves the design of an objective functional H (X) over surface norm distribution X, which is a weighted average of the prior constraints (regularization terms) and posterior constraints (data terms) ; the reconstruction is then the surface norm distribution X, which minimizen H. In the Ikeuchi-Horn algorithm with a quadratic-type regularization term, a prominent weakness is the difficulty in recovering discontinuities in surface norms at edges. In order to overcome this drawback, we therefore examine an algorithm with concave-type regularization term for reconstructing a 3-D shape with edges, since it is suggested by Geman and Reynolds in their study on image reconstruction that such a regularization term is effective in the recovery of discontinuities. We theoretically consider the convergence of the algorithm and a condition for coordinate-wise minimization of H, and derive an upper bound of the relative weight λ of the data term to the regularization term, as a function of other model parameters, such as noise variance and partial derivatives of the reflectance map.
一般社団法人日本機械学会の論文
1995-04-25