基于自回归正则化和稀疏表示的图像超分辨率重建

Super-resolution Reconstruction Based on Sparse Representation and Auto-regressive Regularization

针对图像超分辨率重建中稀疏系数解不精确和重建图像质量不理想的问题,本文提出了一种空间自回归正则化的超分重构算法.该重构算法联合稀疏K-SVD方法训练一个具有相同稀疏系数解的相对应于高低分辨率图像块的字典对,在此基础上利用自然图像本身具有的局部自回归性先验知识来对图像进行处理,经过模型的训练和选择引入正则化项,实现图像的局部约束,从而完整构造了基于正则化的稀疏编码目标函数,为了进一步实现图像去模糊得到清晰图像,利用退化模型实现全局约束.实验结果表明:与Bicubic、NE和SCR等算法相比,本文算法在主观视觉效果和客观评价指标方面都有了一定地提升.

In order to solve the problem that the sparse representation coefficient is not accurate and the reconstruction effect is not ideal in the super-resolution reconstruction,a super-resolution reconstruction algorithm based on sparse representation and autoregressive regularization is proposed.Firstly,this paper used the K-SVD method to train the dictionary that suitable for high and low resolution image,so that the high and low resolution image patches have the same sparse representation coefficients under the dictionary.Secondly,the local autoregressive prior knowledge of the natural image is used to process the image,the regularization term is introduced through the training and selection of the model to realize the local constraint of the image,so that the regularization-based sparse coding the objective function.Finally,the clear and high resolution images are obtained by the whole constrained optimization.Experimental results validate that compared with the methods of Bicubic,NE and SCR,the proposed approach achieves better super-resolution reconstruction effects in both the subjective visual effect and the objective evaluation index by using the adaptive regularization term to further constraint the local structure.