Time frequency analysis of two dimentional signals For denoising using probability graphical models

Abstract

Denoising of real-world signals that are corrupted by Gaussian noise is a long established problem in statistical signal processing. For real-time signals, the techniques like denoising and detection needs the models to be non-Gaussian in nature. The existing models that use the technique of time-frequency analysis typically model the coefficients on the basis of their independency or their modeling to jointly Gaussian. Time-frequency analysis provides a sparse and approximately de-correlated representation of signals. Probabilistic Graphical Models designed along with time-frequency analysis of wavelet coefficients provide powerful models that allow achieving of compression of signals. These models in the time-frequency domain accurately model the behavior of signals regarding statistics at various scales. Expectation Maximization algorithms are developed that are used in the probabilistic graphical models to achieve the required de-noising of two dimensional signals. Signal processing techniques in time-frequency domain have a broad range of applications such as signal estimation, signal prediction and detection, classification and synthesis of signals.