Is Dense Sampling Good for
Lossy Data Compression?
EECS Professor and Associate Chair
Motivated by the scenario of a sensor network taking and encoding many neighboring samples from a correlated random field, e.g., a temperature field, this talk will characterize the efficiency of several lossy data compression methods operating on dense samples. The principal question is the following. As samples become denser, can the increasing correlation among samples be sufficiently exploited to mitigate the increasing number of samples? The answer ranges from no to yes, depending on whether or not the encoder uses scalar or vector quantization, is distributed or centralized, and uses a transform or not.
Thursday, November 6, 2008