一种改进的基于软判决自回归模型图像内插算法

提出一种改进的基于软判决自回归模型图像内插算法,分为正则化权值训练和图像内插两部分.在训练部分,根据局部窗口特征和其对应的最优正则化权值建立字典.在图像内插部分,利用加权最小二乘法估计自回归模型参数,并通过查找字典中距离当前局部窗口最近的样本获得最优正则化权值,然后使用软判决估计重建图像内插值.同时,使用EM算法修正估计出的模型参数和图像内插值.实验结果表明,与其它算法相比,所提的内插算法具有更好的主观和客观效果.

Improved Image Interpolation Algorithm Based on Auto-Regressive Model and Soft-Decision Estimation

An improved image interpolation algorithm is proposed denpending on the auto-regressive （AR） model and the soft-decision estimation,including regularization weight training phase and image interpolation phase.During the regularization weigh training phase,a dictionary is built according to the local patch feature and the optimal regularization weight.During the image interpolation phase,AR model parameters are predicted by the weighted least-squares estimation,and the optimal regularization weight is obtained by searching the nearest sample in the dictionary.Then,the missing pixels are reconstructed by the soft-decision estimation.Furthermore,model parameters and interpolated pixels are both corrected by an expectationmaximization （EM） algorithm.Experimental results show that the proposed algorithm can yield better performance than other leading interpolation algorithms in terms of subjective and objective effects.