Vinay K. Ingle

Backward and forward linear prediction applied to ultraspectral image processing Effects on rate-distortion

Atmospheric infrared sounder images are ultraspectral data cubes that comprise over two thousand spectral bands accounting for well over 25 megapixels of information. In this paper, we focus on the analysis of backward and forward linear prediction (LP) applied in the context of ultraspectral image compression. We start by introducing a detailed analysis of the differences and similarities between them and proceed to present a mathematical model that integrates not only error signal but also LP coefficient encoding. In addition, to overcome some of the limitations of backward LP, we present a hybrid LP scheme where both, backward and forward LP, are put into consideration by dynamically interleaving them in order to minimize the mean square error of the error signal. The model is further extended to compare all three techniques, and both experimental and theoretical samples are contrasted to verify that hybrid LP provides most efficient compression method.

Generalized Linear Models for count time series

In this paper we discuss a class of models for time series of low count data based on the Generalized Linear Model (GLM) approach. Unlike the traditional Auto-Regressive Moving-Average (ARMA) models for continuous Gaussian data, these models capture both the temporal correlation structure and the discrete marginal distribution of count data. We focus on the properties, parameter estimation, and model adequacy aspects for count time series with Poisson or Negative Binomial conditional distributions. The properties and performance of these models are illustrated with synthetic and real data.