基于非局部自相似性的遥感图像稀疏去噪算法

基于非局部自相似性的遥感图像稀疏去噪方法研究,在为后续的图像分析、识别以及较高层次的处理提供保证方面具有重要意义。针对遥感图像中存在非局部自相似性和稀疏性,在分析传统稀疏去噪模型的基础上,将具有相似结构的非局部块构建成组,用组作为稀疏表示单元,利用基于组正则化稀疏模型进行图像去噪。此外,针对采用整幅图像进行字典学习具有高计算复杂度,分析组特点,为每个组自适应学习一个字典。最后,为获得有效的去噪结果,利用迭代收缩阈值算法解决L0最小化问题。以"资源三号"遥感图像为数据进行实验,结果表明,该算法能较好地去除遥感图像的噪声,提高图像的峰值信噪比,保持图像结构信息。基于非局部自相似性的遥感图像稀疏去噪算法能够充分利用图像块信息有效的去除图像中的噪声,提高图像质量。

Remote Sensing Image De-noising via Sparse Representation based on Non-local Self-similarity

Through the Image De-noising method study of remote sensing via sparse representation based on non-local self-similarity,the following image analysis,recognition and higher levels of processing can be provided assurance,which is important.Due to non-local self-similarity and sparsity of remote sensing images,on the basis of traditional sparse representation de-noising model,the group is composed of non-local patches with similar structures,which is exploited as the unit of sparse representation,and group-based sparse representation is used for image de-noising.In addition,because learning a dictionary of the whole image has high computational complexity,the characteristics of group is analyzed,and self-adaptive dictionary of each group is learned.Finally,in order to obtain an effective de\|noising result,the iterative shrinkage thresholding algorithm is developed to solve the L0 minimization problem.The results of the "Resource III" remote sensing images showed that the algorithm can better remove noise of remote sensing images,improve the peak signal to noise ratio and keep the structural information.Based on non\|local self-similarity,the information of patches can be fully used for the image de-noising,so this method can improve the image quality.