基于双非凸约束的遥感图像高密度条带去除算法

High density stripe removal algorithm based on double non-convex constraints for remote sensing images

针对去除高密度遥感图像条带时残留较多且易产生块效应的问题,通过分析条带和干净图像的特征,提出一种新的条带分离方法。条带的稀疏性既存在于在空间域又存在于梯度域,因此,对条带在空间域和梯度域进行非凸L0范数约束,同时采用L1范数约束干净图像的空间域连续性特征。该非凸模型可以通过均衡约束数学规划(MPEC)将其转化为凸优化问题并进行快速求解。实验结果表明:所提算法可以去除高浓度条带噪声的同时,较好地保留原图像的结构和细节信息,而且在峰值信噪比(PSNR)和结构相似性指数(SSIM)指标上显著优于当前最优秀的算法。

Aiming at the problem that there are still much residues and block effect when removing high-density remote sensing image stripes, a new stripe separation method is proposed.The sparsity of stripe exists both in the spatial domain and gradient domain.Therefore, non-convex L0 norm constraint is applied to stripe sparsity in spatial domain and gradient domain, and L1 norm is adopted for constraining spatial continuity of clean images.The proposednon-convex model can be transformed into a convex optimization problem by mathematical programming of equilibrium constraint(MPEC) and solved quickly.The experimental results show that the proposed algorithm can remove high-density stripe noise, while retaining the structure and detail information of the original image, and is significantly prior to the state-of-art algorithms in terms of peak noise ratio(PSNR) and structural similarity index(SSIM).