中文摘要：

本文提出了一种整数递推GCV（Integer Recurrent Generalized Cross Validation, IR-GCV）算法，能高效获取去噪阈值。在对遥感图像做小波变换后，统计子带中小波系数的幅值分布，进行同值小波系数合并运算，然后通过整数优化递推加速GCV计算过程，减少相邻阈值下GCV函数的冗余计算。在对多幅遥感图像仿真实验中，当噪声标准差为10至30时，IR-GCV算法耗时仅为GCV算法的2%至0.5%，且保证了去噪效果一致，能有效提高图像PSNR0.66db至6.03db。此外，GCV算法耗时随噪声增大及遥感图像尺寸增长而迅速升高，IR-GCV算法耗时则相对平稳。

英文摘要：

A computationally efficient IR-GCV （integer recurrent generalized cross validation） algorithm is proposed. Firstly the integer grey-scale pixels are transformed into integer wavelet coefficient. Then its distribution is calculated, followed by the incorporative numeration of coefficient of the same value. Furthermore the integer recurrent procedure based on integer wavelet coefficient is applied, which take full advantage of the relevance between GCV functions under adjacent thresholds. The comparison among GCV and IR-GCV is carried out on multiple remote sensing images. When the noise standard deviation varies from 10 to 30, the time complexity of IR-GCV is only 2 percent to 0.5 percent of that of GCV, while the de-noising results still remain unchanged. The PSNR of de-noising image is 0.66db to 6.03db higher than that of noisy image. Besides, the time complexity of GCV increases rapidly as the noise standard deviation and scale of remote sensing image increase, while the time complexity of IR-GCV is relatively more stable.