为了充分利用广义极小化残量方法在处理大规模线性问题时的优势，将其同正则化技术相结合应用于图像恢复领域，提出了一种新的图像恢复方法。该方法基于Arnoldi过程，用一系列规模远小于原不适定问题的最小二乘问题来逼近原问题，并应用截断奇异值分解正则化技术保证稳定求解这些最小二乘问题。其中，根据图像恢复问题的具体特点，在确定截断奇异值分解的截断次数时，对传统的L 曲线准则进行了少许修改。数值试验结果表明，试验数据与肉眼观察恢复图像的清晰程度相吻合，说明新方法是有效的。

Based on Generalized Minimal RESidual (GMRES) method and regularization technique, a new image restoration method is presented. The method is based on the Arnoldi process, which yields a sequence of smaller least squares problems to approximate the original ill\|posed problem. Truncated Singular Value Decomposition (TSVD) regularization is used to solve these least squares problems stably. According the characters of image restoration, Lcurve criterion is modified while selecting iteration in TSVD method. Numerical results illustrate good restoration effect.