Image restoration using a new regularized particle filter

In this paper, a new regularized particle filter is proposed and applied in mixed noisy image restoration. The general particle filter sample from discrete approximation distribution to cause inaccurate sample for not considering measurement information. In order to reducing the sample error, the regularized continuous distribution sample which is achieved by kernel density approximation function for posterior distribution is proposed when resampling. Meanwhile combing cumulative distribution function (CDF) which can be realized easily and minimize the variance in this new regularized resampling step, thus the degradation problem can be alleviated well. The experiments show the effectiveness of the algorithm, and demonstrated the superiority when comparing with wavelet threshold shrink methods and sample importance resampling (SIR) particle filter method.