【摘要】 本文针对小波域隐Markov树模型(hidden Markov tree model,HMT)的块效应问题,分析了块效应的产生机理,给出了以图像去噪为基础的块效应评价准则,并提出小波域隐Markov交叠树模型(hidden Markov overlappingtree model,HMOLT)和基于该模型的图像去噪算法。该模型通过对每个节点的数据扩展,使每个节点包括相邻的3个(1维)或9个(2维)小波系数,实现同一尺度相邻节点数据的交叠,有效地减轻HMT因树状结构而产生的块效应。实验表明,本文给出的模型和去噪算法,无论是在均方误差(MSE)、块效应指标,还是在主观视觉方面,都优于HMT和基于HMT的去噪算法。更多还原

【Abstract】 For solving the artifact problem of wavelet-domain hidden Markov tree mode (HMT), the mechanism of artifacts is analyzed and an artifact rule for evaluating the artifacts is proposed. Then, a new hidden Markov overlap- ping tree mode (HMOLT) and a denoising algorithm based on HMOLT are developed. By extending the data dimen- sion in each node from 1 to 3 (1D) or 9 (2D), the wavelet coefficients in adjacent nodes at the same scale are overlap- ping, and so the artifacts are reduced effectively. Simulations show that the new model and the denoising algorithm perform better than HMT and HMT based denoising algorithm not only in mean square error (MSE) and the artifact rule, but also in human vision.更多还原