@inproceedings{agr-ras-che-06-whrange,
    author = {Amit Agrawal and Ramesh Raskar and Rama Chellappa},
    title = {What is the Range of Surface Reconstructions from a gradient field },
    booktitle = {Proc. 9th European Conference on Computer Vision (ECCV)},
    location = {Graz, Austria},
    month = may,
    year = 2006,
    series = {Lecture Notes in Computer Science},
    volume = {3951},
    doi = {10.1007/11744023},
    pages = {578--591},
    publisher = {Springer},
    comment = {Describes the generic Poisson-based integration with weights, and several special cases, some old and some new. Claims that Frankot-Chellappa and shapelets are special cases too (I disagree).  The truly Poisson methods are unit-weight Poisson, $\alpha$-surface, M-estimators, Regularization (which seems mathematically incorrect), and Diffusion.  In M-estimators and Regularization, the goal is to minimize a non-quadratic functions; uses iterative weight adjustment to reduce the nonquadratic problem to a quadratic one.  The Diffusion method uses non-isotropic Laplacian and divergent estimators.}
}