中文摘要：

针对目前基于稀疏表示的图像盲复原算法计算量大且细节恢复能力有限等问题，提出一种新的图像盲复原方法．首先针对现有稀疏表示模型中重叠分块计算复杂度高的问题，提出一种多模式非重叠分块策略，在每种模式下独立求解复原图像，然后对各模式下复原图像求平均以消除“伪像”；另外，用12ll范数作为稀疏性度量，将图像梯度稀疏先验融入基于稀疏表示的图像盲复原模型．最后，本文提出了联合新型分块字典稀疏表示和图像梯度稀疏先验的盲复原模型，采取迭代方法交替估计模糊核和待复原图像．实验结果表明，该方法在主观和客观评价下均取得较好的复原结果，并显著降低算法整体复杂度．

英文摘要：

In the light of the heavy computational loads and the limited ability in detail preserving of sparse representation-based blind image deblurring algorithms,a new method based on patch-dictionary approach and gradient prior was proposed in this paper.First,in order to increase the computational efficiency,a multimode non-overlapping strategy,in which the image recovery was independently carried out in each mode,was presented,and then the results of the subproblems were averaged to eliminate artificial effects;In addition,using the 1 2l l norm as the sparsity measure,we incorporated the gradient prior into the sparse deblurring model.Finally,we designed a novel combined new patch-dictionary sparse representation and gradient prior image deblurring model in which the deblur kernel updating and the deblurred image estimating were performed in turn via the iteration.Experimental results show that the approach proposed achieves better results in both subjective and objective evaluation criteria and significantly reduces the overall complexity of the algorithm.