Cutset Image Sampling for Compression, Reconstruction and Source Localization

Motivated by Markov random field models, cutset image sampling is a new approach in which a two-dimensional field, e.g. an image, is sampled densely on a grid of lines, e.g. a Manhattan grid, as an alternative to conventional uniform lattice sampling, e.g. rectangular or hexagonal. It is useful, for example, when sampling is spatially restricted, such as when sampling from a moving vehicle, when sampling with a wireless sensor network, in which case it reduces the energy needed for intersensor communication, or as a first step in image compression, in which case the close sample spacing increases intersample correlation and compressibility. This talk introduces cutset sampling, as well as methods for reconstructing nonbandlimited images from cutset samples, a sampling theorem for cross-bandlimited images, lossless and lossy image compression methods based on cutset sampling, and low energy methods for localizing a radiating source from signal strength measurements in a wireless sensor network.