Infrared image denoising using mixed statistical model in nonsubsampled contourlet domain

Aiming for the denoising problem for infrared image, a novel algorithm is presented based on mixed statistical model in nonsubsampled contourlet domain. The noise coefficients which affect infrared image quality are generally considered to obey Gaussian distribution in nonsubsampled contourlet transform domain. At the same time, the original signal coefficients have the features of sharper peak at the base point and long tail. Therefore, the proposed method describes the noise´s coefficients with Gaussian distribution and depicts the original signal´s with generalized Laplacian distribution. On this basis, by use of the Maximum a Posteriori estimation theory under Bayesian framework, a new estimation formula for the original image´s coefficients is deduced. In the end, the inverse transform for nonsubsampled contourlet transform is done and the denoising result image is obtained. The experimental results show that the method given by this paper, can suppress the Gaussian noise effectively which is produced during infrared imaging´s procedure and has the value of Peak Signal Noise Ratio higher than some standard algorithms. Otherwise, the proposed algorithm can keep most of the image´s detail information and the denoising image has better visual effect.