全变差与曲波联合稀疏表示模型与原对偶算法

Total Variation-Curvelet Joint Sparse Representation Model and Primal-Dual Algorithm

全变差模型因能有效捕捉图像与视频中的细节信息而被广泛应用于机器视觉中,曲波变换具有较强捕捉二维信号中线状跳变信息的能力.文中结合全变差模型和曲波变换的优点,提出一类能更好地捕捉二维信号特征的联合稀疏表示模型,并用原对偶算法求解该模型,即原对偶全变差曲波算法.实验结果表明,用文中模型及求解算法处理后的图像,其客观质量及主观视觉效果均优于现有算法.文中算法也可用于解决图像去模糊、超分辨率等其它具有挑战性的图像处理问题.

Total variation model is widely used in machine vision due to its strong ability of capturing the details of the images and the videos. Curvelet transform can capture the edges and curved lines of the 2D signals easily. Combining both advantages, a class of joint sparse representation model is proposed, i.e. total variation and curvelet （TVC）. This model can represent the characteristics of the 2D signals more effectively. Primal-dual （PD） scheme is used to solve the model, which is called PDTVC algorithm. Experimental results show that PDTVC outperforms the existing algorithms in both subjective visual effect and objective image qualities. PDTVC can be applied to various challenging image processing tasks as well, such as deblurring and super resolution.