Mixed Noise Removal Using TV Inpainting with Adaptive Parameter Adjustment

Abstract

We propose a nonlinear filter that is effective in removing both Gaussian and impulse noise. The proposed filter achieves an excellent performance in the restoration of images degraded by mixed noise, combining smoothing for Gaussian noise removal and inpainting for impulse noise removal on the basis of total variation minimization principle. We also propose how to determine smoothing parameter λ that is important for the quality of restored images, according to a surface model of λ with respect to σ_n and σ_s, standard deviations of Gaussian noise and original image respectively. In applying the proposed method to images degraded by various mixed noise, we verified that both PSNR and subjective image quality are significantly improved, compared with the conventional filters.