多变换域马尔可夫树图像去噪算法

Algorithm of multi-transform domain HMT image denosing

针对图像去噪时单一变换方法的不足,提出了一种新的基于Contourlet变换和小波变换的多变换分级图像降噪算法。根据Wavelet变换和Contourlet变换系数对图像中不同频带信号的稀疏表示特点,利用隐马尔可夫树(HMT)模型可以描述相邻尺度变换域系数的互相关性。首先使用小波域HMT方法进行第一级降噪,然后将其作为先验估计,利用Contourlet变换进行迭代阈值降噪。通过与几种传统的小波域HMT和Contourlet域HMT去噪算法相比,本算法改善了去噪图像的可视性并使PSNR值有所提高。

To overcome the drawback of the preference of one kind of transform used in image denosing a proposed, new multi-level denoising approach based on Contourlet transform and Wavelet transform. Analysed the statistical information of image sparse representation of the wavelet domain coefficients and the Contourlet domain coefficients. Captured the character of more cross-correlations of transform domain coefficients between two neighboring scales in the HMTmode. The first image denosing was used by wavelet domian HMT method. Applied the threshold iterative denosing based on Contourlet transform in the next one by using the first result for prior estimates value. The experimental results indicate that it is better than several traditional wavelet HMT denosing and Contourlet HMT denosing algorithms in smoothing noise and preserving texture and details, and improving the PSNR.