Deconvolution of Poisson Images in Bayesian Domain Using Fuzzy Median Filter

The inverse problem associated with reconstruction of Poisson blurred images arising from many fields ranging from engineering to medicine has attracted much attention in the recent years. Digital images captured at medium to low light levels are often corrupted by sensor noise which is dominated by Poisson statistics. The zoom optics, sensor PSF and atmospheric turbulence cause blurring in acquired images. In this paper, we propose to address the problem of reconstruction of Poisson blurred images captured by surveillance imaging / digital photography in Bayesian domain using fuzzy median filter as a Gibbs prior. The addition of prior information makes the problem well-posed and aids to distinguish between an edge and noisy pixels thereby effects noise filtering. The performance of the proposed algorithm at several intensity levels have been studied qualitatively and quantitatively using mean absolute error (MAE) and Universal Quality Index (UQI) as metrics. The results are compared with other state-of-the-art approaches.