Image Restoration Using Gaussian Particle Filters

Sequential Monte Carlo method has received intense attention among the literature due to its promising applicability to non-linear and non-Gaussian problems. However, while adopting the standard particle filtering method to the area of image restoration, two main drawbacks are found. Firstly, the computational complexity, which mainly comes from a procedure called resample (in a serial implementation), of particle filters would render it too resource-requiring for image restoration. Secondly, the sample impoverishment introduced by resample can affect the filter's performance. In this paper, we adopt a new type of particle filters which do not require resample to the area of image restoration-the Gaussian Particle Filters (GPF). Simulation results are presented to show the GPF's better performances over conventional particle filters. Keywords: particle filter, Gaussian particle filter, resample, sample impoverishment