群稀疏残差约束的引导字典学习算法及其单幅图像去雨

Guided Dictionary Learning Algorithm with Group Sparse Residual Constraints for Single Image Deraining

为了更有效地进行单幅图像去雨,提出一种群稀疏残差约束的引导字典学习算法.该算法特点在于利用混合高斯模型从自然图像中学习外部字典,面向有雨图像构建了基于外部字典引导的内部字典学习模型,并将内部字典的低秩性增加到字典学习目标函数中.该模型可以有效地利用自然图像与有雨图像先验知识之间的互补性,有助于同时恢复潜在稀疏的与稠密的图像细节.其次,基于图像的非局部自相似准则,利用群结构稀疏表示确保了相似图像块的编码系数尽可能接近,并对该模型引入残差约束,可有效地提高学习字典的重构能力与泛化能力.实验结果表明,在合成图像与真实图像上与其他算法相比,使用所提算法去雨后的图像具有更丰富的细节信息,图像更清晰,大大改善了整体视觉效果.

In this paper,guided dictionary learning algorithm with group sparse residual constraints is proposed for single image deraining efficiently.The key of this algorithm is to learn the external dictionary from natural images using Gaussian mixture model,and then we exploit the learned external dictionary to guide internal dictionary learning.Meanwhile,internal dictionary with low-rank constraint is incorporated into the objective function of dictionary learning.The proposed algorithm can effectively utilize the complementarity of prior knowledge between natural images and rainy image,which helps to recover more latent sparse and dense details.Furthermore,based on the criterion of image nonlocal self-similarity,the group structure sparse representation is introduced to ensure that similar image patches have the similar coding coefficients.Additionally residual constraint is incorporated into the proposed algorithm,which can effectively improve the reconstruction and generalization ability of learned dictionary.Compared with other algorithms in the synthetic image and the real image,the experimental demonstrate that the reconstructed image with the proposed algorithm has better high-quality and more detailed information,and visual effect can be significantly improved compared with the state-of-the-art other algorithms.