Mixed Guassian and uniform impulse noise analysis using robust estimation for digital images

Abstract

Previous work on mixed Gaussian and impulse noise (MGIN) reduction has impressive quantitative results. However, the estimation of the statistical properties of the MGIN model that varies within a wide range has not been fully investigated. In this paper, statistical properties of the MGIN model are analyzed in detail with a robust estimation. The paper also proposes a two-stage impulse-then-Gaussian filter for MGIN suppression. which makes use of the estimated statistical properties of MGIN. The proposed filtering scheme applies a impulse proportion adaptive median filter (IPAMF) to impulse noise suppression, and a state-of-the-art discrete cosine transform (DCT) domain filter to Gaussian noise reduction. Numerical results, in terms of the peak signal-to-noise ratio (PSNR), and visual samples demonstrate that the proposed filtering scheme achieves better performance of noise reduction than two existing MGIN filtering schemes.