小波域HMT模型参数的快速估计及其在图像降噪中的应用

小波域隐马尔可夫树(Hidden Markov Tree，HMT)模型可以很好地刻画尺度内与尺度间小波系数的相关性，但模型参数的训练过程复杂，计算量大。针对这个缺点，提出了一种不经训练的HMT模型参数快速估计方法。该算法首先用一种自适应阈值将每个子带小波系数分成不同的类，然后分别对每类进行统计，这种统计是局部的，因而有很好的局部自适应性，最后模型参数可以利用这些局部的统计特性来描述。将估计出的参数模型运用到图像降噪中，实验结果表明这种快速估计的HMT参数模型不仅可以大大提高计算速度，降低计算复杂度，而且从峰值信噪比和主观视觉效果上都不逊于传统的经过迭代训练的HMT模型降噪算法。

Fast estimation of parameter in wavelet-domain HMT model and its application in image denoising

Although HMT model captures intrascale and interscale dependencies of wavelet coefficients, model parameter training is complex and computationally expensive. To solve this problem, a fast parameter estimation algorithm in which training is not needed was proposed. Firstly, each subband coefficients were classified by spatially adaptive threshold. Secondly, the local statistical features of different classes were computed respectively, and HMT model parameters can be estimated by computing local statistical features. Finally, a non-training HMT was applied to image denoising. Experimental results show that this fast parameter estimation algorithm can not only reduce computing expense and accelerate computation, but also provide an improved denoising performance of PSNR and human vision beyond other methods.