基于抗混叠轮廓波变换系数分布模型的去噪算法研究

针对轮廓波变换存在频谱混叠致使其难以获得理想的去噪效果这一问题,本文提出一种基于抗混叠轮廓波变换系数分类的混合模型图像降噪算法.该算法通过计算变换系数的尺度间相关性,将系数分为重要系数和非重要系数两类,并对二者分别采用广义非高斯二元变量分布与零均值高斯分布建模,在Bayes框架下对原始图像进行估计.实验研究结果表明,以Barbara图像为例,当噪声方差σ=30时,本文算法不仅峰值信噪比(PSNR)超过Contourlet-HMT模型去噪2.72 dB,且主观视觉效果上亦均优后者,同时还具有较高的计算效率.

Novel denoising algorithm based on coefficient distribution model of non-aliasing contourlet transform

Frequency aliasing of contourlet transform poses difficulties for image denoising. To deal with this problem, a hybrid model based on the characteristics of non-aliasing contourlet transform (NACT) coefficients is proposed for image denoising in this paper. Coefficients of NACT are classified into two categories: important and non-important, in terms of their interscale correlations. Generalized non-Gaussian bivariant distribution and zero-mean local Gaussian distribution are used to model the important coefficients and non-important coefficients respectively, which are then incorporated into Bayesian framework for denoising. Experimental results show that, for Barbara image, the proposed algorithm is superior to contourlet denoising method based on hidden Markov tree model in terms of PSNR (2.72 dB with σ=30) and visual quality. In addition, much higher computational efficiency is also achieved.