基于马尔科夫随机场学习模型的图像模糊核估计

发布时间：2018-05-17 17:50

本文选题：图像恢复 + 模糊核　；参考：《东南大学学报(自然科学版)》2016年06期

【摘要】：为在图像模糊核估计中充分利用图像的区域特征和结构信息作为先验知识,提出一种基于马尔科夫随机场学习模型的模糊核估计方法.首先,由滑动的子窗口构成马尔科夫随机场的节点集,以每个子窗口的曲率方向能量滤波器的响应和边缘分布组成的特征向量作为模型的输入;然后,利用对数伪似然优化算法估计模型参数,在模型训练阶段,采用交叉熵相似性度量模糊核的相似性以标记训练样本;最后,利用置信度传播算法推测最优图像子块.运用所提方法对仿真和实际模糊图像进行实验,结果表明,该学习模型可以精确地估计模糊核,在主观视觉对比和客观评价方面均具有较好的效果,同时也具有较好的自适应性.与其他3种方法相比,模糊核相似度分别提高了1.55%,5.64%和7.02%.
[Abstract]:A fuzzy kernel estimation method based on Markov random field learning model is proposed in order to make full use of the region feature and structure information of image as prior knowledge in image fuzzy kernel estimation. Firstly, the Markov random field node set is composed of sliding subwindows, and the eigenvector composed of the response and edge distribution of the energy filter in the curvature direction of each sub-window is taken as the input of the model. The logarithmic pseudo-likelihood optimization algorithm is used to estimate the model parameters. In the model training stage, cross-entropy similarity is used to measure the similarity of the fuzzy kernel to mark the training samples. Finally, the confidence propagation algorithm is used to estimate the optimal image subblocks. Experiments on simulation and actual blurred images are carried out by using the proposed method. The results show that the proposed learning model can accurately estimate the fuzzy kernel, and has good results in subjective visual contrast and objective evaluation. At the same time, it also has good adaptability. Compared with the other three methods, the similarity of fuzzy kernel is increased by 1.555.64% and 7.2% respectively.
【作者单位】：东南大学机械工程学院;
【基金】：国家自然科学基金资助项目(51275090);国家自然科学基金科学仪器基础研究专款资助项目(21327007) 中央高校基本科研业务费专项资金资助项目江苏省普通高校研究生科研创新计划资助项目(KYLX15_0208)
【分类号】：TP391.41

【相似文献】