Uniformly sampling the space of exposure times until every pixel is correctly recorded at least once (i.e. it is not always
clipped) can result in an unnecessarily large image stack with a sub-optimal Signal-to-Noise ratio. For the scene shown in the
tonemapped image on the left, this results in a 5-image stack. Our method determines that for this scene only 3 images suffice to
capture the whole range while sampling the important intensity levels better. The insets for the 5-image stack (top row) and for
our method (bottom row) are not tonemapped; rather they are linearly mapped to fit in 8 bits so as to preserve the noise characteristics.
Notice the huge improvement in terms of noise with the smaller set of images selected by our method.
bibtex:
@article{GalloEG12,
    title = {Metering for Exposure Stacks},
    author = {Gallo, O., and Tico, M., and Manduchi, R., and Gelfand, N., and Pulli, K.},
    journal = {Computer Graphics Forum (Proceedings of Eurographics)},
    year = {2012},
    volume = {31},
    pages = {479-488}
}