基于二元广义正态分布模型的双树复数小波图像去噪

Image denoising based on bivariate generalized normal model of DTCWT

双树复数小波变换具有平移不变性和多方向选择性,适用于图像去噪.对小波系数统计分布进行建模,提出了一种二元广义正态分布的概率模型.在此先验分布的基础上,通过运用最大后验概率估计方法,从含噪系数中去除高斯噪声.实验表明,该方法不仅在直观视觉上去噪效果明显,在信噪比方面也要优于Bayes-Shrink、W iener2、SureShrink等方法.

Complex wavelet transform is suitable for image denoising clue to its characteristics of shift invariance and multi -directional selectivity. A model based on statistical distribution for complex wavelet coefficients is proposed which using the bivariate generalized normal probability density func- tion. Under such prior distribution, MAP （Maximum a Posteriori ） estimator is used to restore the wavelet coefficients from the noisy observations. Experiments show that the proposed method is better than recently published methods, such as BayesShrink, Wiener2 and SureShrink.