Abstract:
Wavelet based statistical image denoising is vital preprocessing technique in real world imaging. The existing techniques are based on time-frequency domain where the wavelet coefficients need to be independent or jointly Gaussian. In denoising arena there is a need to exploit the temporal dependencies of wavelet coefficients with non-Gaussian nature. Here we present a denoising strategy based on Hidden Markov Model (HMM) based on Multiresolution Analysis in the framework of Expectation-Maximization algorithm. Proposed algorithm applies denoising technique independently on each frame of the video. It models Non-Gaussian statistics of each wavelet coefficient and captures the statistical dependencies between coefficients. Denoised frames are restored inversely by processing the wavelet coefficients. Significant results are visualized through objective as well as subjective analysis.
